Method of The Management of Garbage Coll-201-300

Utilization of Redundant Communication Network 181
Throughput
connected to two or more independent buses realizing quite the same
functionality. The MuNetIN node is equipped with three networks, but the
management of those connections is different than it would be in a
redundant system. The communication process on redundant system is
usually managed independently and the set of communication tasks is the
same on every redundant bus. Mean- while, in the presented solution there
is one manager common for all buses and the communication process on
every bus may vary.
4.1 General Idea

Three MuNetIN buses have the same physical layer with Master-Slave
protocol. The MuNetIN node is the Master node. The Slave subscribers may
be connected to one, two or all three buses. The slaves addressing is unique
for every slave and common on every bus. For example, when slave S1 is
connected to buses A and B with an addr (the same for both buses) then
the address addr is also considered occupied on the third C bus (Fig. 1).
Fig. 1. Communication buses in multiple network interface node MuNetIN
In systems with redundancy, communication with every Slave would be

executed to use every available network connection. For the above
example, communication with slave S1 would be done on both A and B
buses consuming twice the throughput necessary to send all user data. The
MuNetIN on the other hand, would realize the data exchange only on all
buses the given subscriber is connected to, but only in parts so the
payload is distributed. That should increase the efficiency of the
communication system.
Greater communication efficiency allows implementation of additional
features to the system, such as additional data exchanges that are not
critical to the operation of the whole industrial computer system, but useful
from the point of view of maintenance or supervising activities. That
requires to divide of the network data exchanges into two groups – one with
exchanges critical to the system operation and the other with non-critical
exchanges. The exchanges from the latter group would be executed only in
fault-free circumstances, taking into consideration the current temporal
parameters of the communication system, so that the real-time constraints
are satisfied. The membership in these groups is based on data exchanges
priorities defined by the system developer.
182 A. Kwiecien´ et al.
4.2 Triple Network Interface Communication System

In Master-Slave networks, the communication is based on request-response
queries. The Master station, according to previously prepared list of queries,
sends requests to the Slaves. The queries may be of several types with
general functionality of two kinds – read or write queries.
In typical applications, all queries are exchanged on one communication
bus and the Master station only determines which query should be send at
a given time. The MuNetIN node has triple communication interfaces
TNICS, so it also has to decide which interface the query should be sent
through. Therefore, it requires one common data exchange scheduler which
would control the communication process. In the MuNetIN node, a single
Master station is implemented for all buses with the controls algorithms
extended for the usage of multiple network connections. Moreover, the
MuNetIN node has to include precise information about the availability of
every Slave subscriber on every network.
Having information about Slaves availability, it is possible to transmit
the queries on multiple communication interfaces. To realize the master
functionality, a program module was implemented to service to multiple
interfaces – Queries Distributor QD (see Fig. 2). The QD module checks the
busy status of the network interfaces. When an interface is not occupied,
then the QD asks the Queries Scheduler QS for a query to be realized on
that interface. The QS may choose either periodic or aperiodic query to be
realized. It may be either critical or not critical (optional), too. Then the
distributor passes over the received query definition to a Query Handler QH.
The QD at this point sets the bus status to busy and the QH sends it
through the network. The busy status is cleared by the QH after realization
of the query.
4.3 Query Scheduler

The queries in Master-Slave communication networks have two transmission
rules – periodic and aperiodic. For every periodical query it is defined how
often
Fig. 2. Query distribution

Throughput
it should be realized (e.g. every 100 ms). According to that parameter, the
master creates a schedule of the queries transmission. Realization of all the
queries from the schedule (some of the queries possibly more than once) is
called the network communication cycle NC.
In Master-Slave networks, the aperiodic queries are usually sent immedi-
ately (out of the order of the scheduled periodical queries) or after
realization of all periodic queries in the network cycle. When there is a
necessity of sending more than one aperiodic query, the queries are
sequenced in FIFO order. The maximum number of aperiodic queries in one
NC is limited hence the real-time constraints defined for the communication
network are satisfied.
The Master station that operates on one communication interface, may
realize the queries based on a static schedule created once on the system
startup. Having three independent interfaces it is no longer possible. The
scheduling have to be performed dynamically taking into consideration the
current status of all interfaces.
In the MuNetIN node, the query scheduler QS, on the Queries
Distributor QD requests, selects from the list an appropriate query in two
steps. Firstly, it excludes from the list all the queries that cannot be realized
on the bus given by the QD. Secondly, it chooses from the result list the
query with the shortest deadline which is calculated basing on the defined
periodicity and the subtraction of the current time and the timestamp of the
last successful realization of the given exchange. The aperiodic queries are
passed over to the QD for realization when it would not jeopardize the
periodic exchanges but no later than after realization of all queries from the
schedule.
5 Formal Description of Distribution Algorithm

5.1 Periodic Tasks
Let BD (bus/device) denote a matrix which defines connections between
buses and devices. Assuming there is n devices and m buses we are
considering the matrix
⎡ . ⎥⎤
BD = bd.1,1 · ·. ·. bd. 1,n
⎢
⎣ ⎦
where bdm,1 · · · bdm,n ,
1 if device j is connected to bus i,
bdi,j =
0 otherwise.
Let matrix DT (device/task) similarly define devices served in particular
communication tasks. Assuming there is n devices and r tasks, the matrix
DT is as follows:
⎡ .. ⎥⎤
DT = dt1,1
. · · · . dt1,r
⎢
⎣ . ⎦
,
dtn,1 · · · dtn,r
Throughput
where
1 if device i is involved in task j,
dti,j =0 otherwise.
We are considering communication tasks related to exactly one device,
thus every column of matrix DT contains exactly one 1, i.e.:
∀j=1...r ∃i dti,j = 1
∀j=1...r dtx,j = dty,j = 1 ⇒ x = y
Matrix BT (bus/task) is a matrix product BT = BD · DT defining tasks
which may be served on particular buses.
⎡ . ⎥⎤
bt.1,1 · ·. ·. bt1,r
BD · DT = BT = . ⎦
where ,
btm,1 · · · btm,r
i,j ⎢ 1 if task j may be served on bus i,
bt =⎣
0 otherwise.
Additionally, each from r tasks has preset periodicity, i.e. maximal time
tasks. P = p1 · · · pr , where: pi denotes periodicity of task i.
between serving successive task. Let P be a vector defining periodicity of
the
Communication tasks are served by system described in Sect. 4.3, where
Queries Distributor chooses successive task to serve. Distributor is called
peri- odically, for every bus, as soon as the bus becomes idle. Selection of
the next task is based on periodicity in order to choose the task which may
be handled on a given bus, and whose time period left to obey the declared
periodicity of that task is minimal. Therefore, additional information about
time left to deadlines is necessary. Let T be a vector defining time left
for serving each task.
T = t1 · · · tr , where ti denotes maximal time, in which task i may be
served to obey periodicity defined in P .
Matrices BD, DT, BT, P are constant, T is variable changed by
Distributor. Initially T = P . According to passing time T decreases.
Choosing task j to serve involves changing tj to initial value tj = pj,
defined by periodicity of task j.
Naturally, considering the system with defined time for serving single
communication task tsingle, identical for each task, it is possible to simplify
the model to handle discrete time, where time flow is modeled by decreasing
each element of T by tsingle, after sending tasks in each bus.
Choosing the task to serve on bus i, Distributor is looking for a
minimum value in these elements of T , where corresponding elements in
vector BTi,1..r are 1, i.e. among the tasks which may be served on a given
bus. Additionally, tasks, which are just being served on another bus, should
be omitted. In order to decide if the query related to the task i has been just
sent or not, such information will
be stored in vector Q = q1 · · · qr , where:

1 if task i is just served,
qi =
0 otherwise.
Hence, Distributor for given i should choose such j, that:
tj = min {tj : bti,j = 1 ∧ qj = 0}.

j∈{1..r}
5.2 Aperiodic Transactions

Aperiodic query may be sent if such transaction will not jeopardize timely
realization of periodic tasks. The main problem is to define such condition.
Let pMAX = max pi be maximal periodicity defined in P . We are considering
a sys-
P
tem with the following constraints:
– n devices, m buses, and r periodic tasks,
– a pool of aperiodic queries, without time constraints,
– constant time for serving single communication task tsingle, identical for
each query.
In such conditions communication will be performed in time slots, defined
tsingleby
. Let S define number of available slots during pMAX time, for all buses:
pMAX ∗ m
S= .
tsingle
Each task i during pMAX time is handled approximately pMAX pi times. Thus,
to handle allr periodic tasks in given periodicity constraints, during pMAX time
Σ pMAX
on average
pi slots are needed.
i=1
Let cf be a coefficient, calculated in every cycle, similarly as above, but
basing on ti instead of pi. Such coefficient might be understood as a number
of slots needed for all periodic tasks, if periodicity of each task was equal to
its actual time to deadline. According to passing time, such coefficient for
each task increases until the task is served, then it returns to its initial
value.
Σ
cf = r pMAX
ti if ti is not zero,
i=1 pMAX otherwise.
In the cycles where cf < S − m, i.e. dedicating m slots for aperiodic
queries (each query being served in one slot) does not expose deadlines for
periodic
transactions to risk, successive aperiodic queries from the pool will be
handled on each bus.
6 Algorithm Realization Example

Let us assume that connections between devices and buses are defined in the
following matrix:
⎡ ⎤
110 1
BD = ⎣ 0 1 1 1⎦
1110
According to Sect. 5, each column of the matrix defines connection of a
particular device to buses. Slave 1 is connected to buses A and C, slave 2
to all three buses, slave 3 to B and C, whereas slave 4 is connected to buses
A and B.
DT matrix defining devices involved in particular tasks, is as follows:
1100000 0
⎡ ⎤
DT = 0⎣0 0 1 0 1 1 0
⎢
⎢0 0 1 0 0 0 0
⎥ 00001001
0 ⎦
Each column defines the device serving given task. The first task is
served by slave 1, the second also by slave 1, the third by slave 3, the
fourth by slave 2, etc.
BT matrix is calculated as matrix product of BD and DT .
⎡ ⎤
1101111 1
BT ⎣0 0 1 1 1 1 1 1⎦
= 11110110
Each column defines buses, where given communication task may be
served. The first and second tasks may be served on buses A or C, the third
on B or C, the fourth on all buses, etc.
Periodicity vector for the tasks is as follows:
P = 50 50 50 100 100 150 150 300

Tasks 1–3 have to be served at least every 50 time units, 4 and 5 at least
every 100 time units, tasks 6 and 7 at least every 150 time units, and 8 at
least every 300 time units.
Time for serving a single communication task equals tsingle = 10. The
number of slots calculated according to Sect. 5.2 equals S = 90. Resulting
list of queries is shown in Table 1. For brevity, each aperiodic query has
been denoted as Q+, however such value should be understood as one of the
queries from the pool of aperiodic queries defined in Sect. 5.2, but not
necessarily the same query each time.
The initial value of time to deadline for each task equals its periodicity
(column 1 in Table 1). Coefficient cf calculated for these values equals
29.0. It is
The value of 27 is less than S − m = 87 (the cf = 87 is marked with dotted
shown in the cf row in Table 1 and in Fig. 3 in the first graph from the
bottom. line in the bottom graph), so in the first cycle aperiodic queries Q+
are handled
on each bus. The moments when queries Q+ may be sent are marked by
gray rectangles with dots in the graph. No periodic queries have been sent,
thus time to deadline for each task decreases by 10 (tsingle) each, as shown in
column 2 of Table 1 and in graphs Q1–Q8 depicting the change of deadlines
of each query in time.
Throughput 1
Table 1. List of queries
Time to 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
deadline
Q1 50 40 30 20 10 50 40 30 20 10 50 50 40 30 20
Q2 50 40 30 20 10 50 40 30 20 10 50 40 30 20 10
Q3 50 40 30 20 10 50 40 30 20 10 50 40 30 20 10
Q4 100 90 80 70 60 50 40 30 20 10 0 100 90 80 70
Q5 100 90 80 70 60 50 40 30 20 10 0 100 90 80 70
Q6 150 140 130 120 110 100 90 80 70 60 50 40 30 20 10
Q7 150 140 130 120 110 100 90 80 70 60 50 40 30 20 10
Q8 300 290 280 270 260 250 240 230 220 210 200 190 180 170 160
cf 29.0 34.5 43.2 59.7 106.6 37.2 45.4 58.8 84.9 161.4 631.5 43.6 55.8 79.3 145.4
Query Bus A Q+ Q+ Q+ Q+ Q1 Q+ Q+ Q+ Q+ Q1 Q4 Q+ Q+ Q+ Q2
Bus B Q+ Q+ Q+ Q+ Q3 Q+ Q+ Q+ Q+ Q3 Q5 Q+ Q+ Q+ Q3
Bus C Q+ Q+ Q+ Q+ Q2 Q+ Q+ Q+ Q+ Q2 Q1 Q+ Q+ Q+ Q6
Fig. 3. Deadlines and queries realization
In the second cycle cf equals 34.5 which is still less than S − m.

Therefore, 5th cycle (column 5) the value of the cf is greater than S − m
and aperiodic aperiodic exchanges are sent again. Similarly, in the cycles
3rd and 4th. In the exchanges with minimal time to deadline are chosen,
namely Q1, Q3, and Q2
respectively for buses A, B, and C. The fact of sending the queries on
buses is depicted in the bottom part of Fig. 3 (by table with grayed cells).
Time to deadline for these queries is reset to the initial value (50), time to
deadline for other queries Q4–Q8 decreases by 10 (column 6). In the 6th
cycle cf equals 37.2, so again aperiodic queries are handled, and time to
deadline for each task
decreases by 10. Periodic queries are sent again in the 10th and 11th cycle as the
cf is higher than S −m. After that again aperiodic queries are sent 3 more times.
For brevity, only 15 cycles have been shown, however numerous cycles and
different examples have been analyzed during the algorithm development.
7 Conclusions and Future Work

The main idea presented in the paper is the concept of an algorithm for
queries distribution to be implemented in multi network interface node
MuNetIN which is dedicated to network control systems NCS. The goal is
to take advantage of redundant communication interfaces with real-time
communication protocols. The algorithm should allow both, increased
reliability and greater communication network efficiency.
The performed simulation tests show that the algorithm is able to sched-
ule periodic exchanges so that they are realized with at least the periodicity
defined for each query. Moreover, it is possible to make room for realization
of aperiodic exchanges. Nevertheless, the presented solution needs further
development. At this point, the periodic exchanges and aperiodic
exchanges are always “grouped” and being realized in “bursts” (e.g. couple
of Q+ exchanges on all buses). Besides, in the presented example the
exchange Q1 was always realized on the bus A. It would be better to realize
it also on every other bus, to which the slave associated with that exchange
is connected to. That would allow some diagnostics of the state of the
network connections.
What is the strong point of the algorithm, the choice of queries for
realization in the proposed algorithm is dynamic. As a result, the Master
node is capable of adjusting to change network topology. Importantly, the
cause of the change could be planned (moving a system node from one
network to another) or unplanned (failures affecting the accessibility of the
system nodes). Any changes in the topology will be automatically serviced
by the query distributor so that the communication link is maintained using
the available resources in most efficient way.
References
1. Gaj, P., Malinowski, A., Sauter, T., Valenzano, A.: Guest editorial: distributed
data processing in industrial applications. IEEE Trans. Ind. Inform. 11(3), 737–
740 (2015)
2. Gaj, P., Jasperneite, J., Felser, M.: Computer communication within industrial
distributed environment - survey. IEEE Trans. Ind. Inform. 9(1), 182–189
(2013)
3. Birkholz, H., Sieverdingbeck, I.: Link-failure assessment in redundant ICS
networks supported by the interconnected-asset ontology. In: 2014 IEEE
International Work- shop Technical Committee on Communications Quality and
Reliability (CQR), pp. 1–6, May 2014
4. Kopetz, H.: Real-Time Systems: Design Principles for Distributed Embedded
Applications. Springer, Heidelberg (2011)
Throughput
5. Sauter, T.: The three generations of field-level networks-evolution and
compatibility issues. IEEE Trans. Ind. Electron. 57(11), 3585–3595 (2010)
6. Flak, J., Gaj, P., Tokarz, K., Wide-l, S., Ziebinski, A.: Remote monitoring of
geo- logical activity of inclined regions - the concept. In: Kwiecien´, A., Gaj,
P., Stera,
P. (eds.) Computer Networks. Communications in Computer and Information
Sci- ence, vol. 39, pp. 292–301. Springer, Heidelberg (2009)
7. Popovic, M., Mohiuddin, M., Tomozei, D.C., Boudec, J.Y.L.: iPRP: parallel
redundancy protocol for IP networks. In: 2015 IEEE World Conference on
Factory Com- munication Systems (WFCS), pp. 1–4, May 2015
8. Yu, X., Jiang, J.: Hybrid fault-tolerant flight control system design against
partial actuator failures. IEEE Trans. Control Syst. Technol. 20(4), 871–886
(2012)
9. Kwiecien´, A., Stój, J.: The cost of redundancy in distributed real-time systems
in steady state. In: Kwiecien´, A., Gaj, P., Stera, P. (eds.) Computer Networks.
Com- munications in Computer and Information Science, vol. 79, pp. 106–120.
Springer, Heidelberg (2010)
10. International Electrotechnical Commission: IEC 62439-2: Industrial communica-
tion networks: High availability automation networks - part 2: Media
redundancy protocol
11. Zuloaga, A., Astarloa, A., Jiménez, J., Lázaro, J., Araujo, J.A.: Cost-effective
redundancy for ethernet train communications using HSR. In: 2014 IEEE 23rd
International Symposium on Industrial Electronics (ISIE), pp. 1117–1122, June
2014
12. International Electrotechnical Commission: IEC 62439–3 (2012): “industrial
communication networks: High availability automation networks” - part 3:
Parallel redundancy protocol (PRP) and high availability seamless redundancy
(HSR)
13. Giorgetti, A., Cugini, F., Paolucci, F., Valcarenghi, L., Pistone, A., Castoldi, P.:
Performance analysis of media redundancy protocol (MRP). IEEE Trans. Ind.
Inform. 9(1), 218–227 (2013)
14. Kirrmann, H., Weber, K., Kleineberg, O., Weibel, H.: HSR: zero recovery time
and low-cost redundancy for industrial ethernet (high availability seamless
redundancy, IEC 62439–3). In: 2009 IEEE Conference on Emerging
Technologies Factory Automation, pp. 1–4, September 2009
15. Rentschler, M., Heine, H.: The parallel redundancy protocol for industrial IP
networks. In: 2013 IEEE International Conference on Industrial Technology
(ICIT),
pp. 1404–1409, February 2013
16. Hoga, C.: Seamless communication redundancy of IEC 62439. In: 2011 Interna-
tional Conference on Advanced Power System Automation and Protection, vol.
1,
pp. 489–494, October 2011
17. Nsaif, S.A., Rhee, J.M.: Improvement of high-availability seamless redundancy
(HSR) traffic performance. In: 2012 14th International Conference on Advanced
Communication Technology (ICACT), pp. 814–819, February 2012
18. Nsaif, S.A., Kim, S., Rhee, J.M.: Quick removing (QR) approach using cut-
through switching mode. In: 2016 18th International Conference on Advanced
Communi- cation Technology (ICACT), pp. 170–147, January 2016
19. Saadi, I.M.A.: Dynamic message transmission scheduling using can protocol. Int.
J. Sci. Technol. Res. 2(9), 158–162 (2013)
20. Wey, C.L., Hsu, C.H., Chang, K.C., Jui, P.C.: Enhancement of controller area
network (CAN) bus arbitration mechanism. In: 2013 International Conference
on Connected Vehicles and Expo (ICCVE), pp. 898–902, December 2013
21. Kwiecien´, A., Kwiecien´, B., Maćkowski, M.: Algorithms for transmission
failure detection in a communication system with two buses. In: Gaj, P.,
Kwiecien´, A., Stera, P. (eds.) Computer Networks. Communications in
Computer and Informa- tion Science, vol. 608. Springer International
Publishing, Cham (2016)
22. Kwiecien´, A., Kwiecien´, B., Maćkowski, M.: A failure influence on
parameters of real-time system with two buses. Computer Networks.
Communications in Com- puter and Information Science, vol. 608. Springer
International Publishing, Cham (2016)
Software Defined Home Network
for Distribution of the SVC Video
Based on the DASH Principles
Slawomir Przylucki1(✉), Artur Sierszen2, and Dariusz Czerwinski1

1
Lublin University of Technology, 38A Nadbystrzycka Street, 20-618 Lublin, Poland
[email protected], [email protected]
2
Lodz University of Technology, 116 Zeromskiego Street, 90-924 Lodz, Poland
[email protected]
http://www.pollub.pl
http://www.p.lodz.pl
Abstract. With the spread of Dynamic Adaptive Streaming over

HTTP (DASH) systems, appeared the problem related to the resources
management in end-users’ networks. The article proposes a new
solution for the distribution of Scalable Video Coding (SVC) video
streams in home networks. It relies on a Software Defined Network
(SDN) controller, which based on the parameters of transmitted SVC
video and the usage status of network resources allows for dynamic
traffic shaping. At the same time, it was assumed that the proposed
solution is compatible with Server and Network Assisted DASH
(SAND) standard and its implementation is possible without the use
of additional, dedicated hardware. The article presents the analysis of
influence of proposed SDN on the improvement of the SVC video
quality received by DASH clients in a typical home network. Test
results confirm the superiority of the proposed system over the
conventional solutions.
Keywords: Home network · Software Defined Network (SDN) · Scal-

able Video Coding (SVC) · Dynamic Adaptive Streaming over HTTP
(DASH) · Server and Network Assisted DASH (SAND) · Docker
containers
1 Introduction
Nowadays, the importance of multimedia transmissions for the entire
market of Internet services, is no longer subject to discussion. The undoubted
success of services based on the transmission of video is not only due to the
rapid development of network infrastructure, both within the Internet backbone
and access networks. An important role in their wide acceptance also played
the development of techniques for efficient video encoding, new methods for
video streaming and scalable content distribution. In the article, special
attention is paid to two techniques: HTTP Adap- tive Streaming (HAS) [1]
and Scalable Video Coding (SVC) [2].
Oc Springer International Publishing AG 2017
P. Gaj et al. (Eds.): CN 2017, CCIS 718, pp. 195–206, 2017.
DOI: 10.1007/978-3-319-59767-6 16
196 S. Przylucki et al.
Despite the obvious advantages of the HAS approach [3], there are still
issues that need attention in the implementation of video streaming
services. The most important challenges can certainly include issues related
to ensuring the fairness allocation of available bandwidth to individual video
streams [4].
Most of the current proposals to guarantee fair allocation of resources
were focused on systems implemented in distribution networks or networks
of local Internet providers [5]. Much less attention was paid to home
networks or networks in general, which are attached to the end-user of
DASH services. Such an approach does not seem to be justified in the
context of the rapid increase in the number of individual users of video
streaming services. Unfortunately, in home networks, dedicated hardware
and complex algorithms for monitoring and traffic shaping, could not be
easily applied [6]. A promising basis for such solutions is the idea of
Software Defined Network (SDN) [7]. So far, two basic restrictions
significantly affect the popularity of the solutions of this type. The first was
the limited availability of network devices supporting the protocol
OpenFlow. The second reason was related to the complex process of
installation and configuration of SDN controllers. Currently, the first cause
becoming less important due to the availability of efficient, home
networking equipment. In turn, the second limitation can be effectively
eliminated using the lightweight virtualization. In this context, particularly
promising seems to be the idea of microservices and especially the
containerization techniques based on Docker containers [8].
The remainder of the paper is organised as follows. Related works are
presented in Sect. 2. Section 3 contains the description of the proposed
SDHN solution for video stream distribution in a home network. The test
scenario and performance evaluation are presented in Sect. 4. Finally, a
brief conclusion is made in Sect. 5.
2 Background and Related Works

The fundamental principle of the HAS systems is based on the distribution
of an audio-video content that is stored on HTTP servers in the form of
multiple representations. These representations are associated with
different video encoding parameters and thus require a different amount of
available bandwidth along the transmission path. Depending on the
utilization of network resources between the server and client and also
considering the parameters of the receiv- ing devices, it becomes possible to
choose the best representation and thus to adapt to changing network
conditions. The effectiveness of these solutions has led standardization
organizations to try to develop a uniform standard for systems HAS. As a
result, in 2012, ISO/IEC published the standard: Dynamic Adap- tive
Streaming over HTTP (DASH) [1]. In the case of DASH, the progressive
download of a video representation was replaced by the so-called segment
approach. This solution is based on segmentation of each representation
into chunks containing the video fragments of the fixed length. A
description of segmented data structure, stored on the server DASH, is
contained in a dedicated Media Presentation Description files (MPD).
Software Defined Home Network for Distribution of the SVC 197
Video
DASH clients, based on the information
contained in these files, can switch between different representations

(chunks) during the video playback. The DASH adaptation algorithm,
described in the standard, unfortunately does not contain mechanisms
related to the reducing the impact of the mutual influence of the individual
video streams. Each DASH client realizes the process of adaptation in
isolation from the others, competing with the other receivers. In other
words, DASH client are trying, in very selfish way, to secure the best
quality of the received video. This leads to decreasing the values of Quality
of Experience (QoE). This degradation is observed at all receivers in a given
network segment [9]. Recently these issues were reflected in the new
standard developed by ISO/IEC, which was called Server and network
assisted DASH (SAND) [10].
The introduction of multi-layer coding simplifies the preparation, storage
and video distribution in DASH systems [11]. The idea of SVC is to use the
hierarchical structure of the video stream, which allows receivers to select
only the portion of the video data while maintaining accuracy and
continuity of the decoding process. This feature has been obtained by
defining within the structure of a video stream, the following types of
layers: the base layer and the enhanced layers [12]. The base layer is always
necessary for decoding of a video. It provides video playback with minimal
values of the parameters (resolution, image smoothness, image quality). The
enhanced layers allow for completion of data from the base layer and
thereby to obtain better video quality. An important advantage of the
SVC-DASH systems is that the video content is encoded only once. SVC-
DASH clients realize the process of adaptation by selection of the amount of
decoded layers from a single representation of video content. This process is
often called as switching the Operation Point (OP) [13, 14].
Recent years have brought increased interest in the possibility of using
the SDN principles to the process of adaptation in DASH systems [6, 15,
16]. In this context, the most promising areas of the research are the
methods for traffic shaping in general and ensuring a fair distribution of
SVC video streams as a special case [17, 18]. Simultaneously, thanks to
advances in the field of virtualization, it became possible to implement
these solutions on the equipment available for most users of home computer
networks [19].
3 Proposed Solution
The proposed solution is designed for typical home networks. Thus,
following assumptions have been made:
– Multiple DASH clients are connected to an external network (typically,

the ISP network) through the gateway. The link between the home
gateway and ISP is the bottleneck.
– The proposed SDHN should be compatible with the principles of SAND.
The SDN controller should work with any network devices supporting the
protocol OpenFlow 1.3 and above.
– The SDHN controller should be installed on any hardware platform,
supporting Docker environment, version 1.9 and above.
Video
From the perspective of the DASH clients, the SDHN is a kind of service
responsible for ensuring the best possible parameters of video stream. Thus,
the functionality of the proposed SDNH can be called as QoS as a Service
(QaaS). The structure of the SDNH controller consists of four functional
components:
– Network resources module (NRM).

– QoS controller (QoSC).
– Traffic shaping module (TSM).
First two modules are associated with a few fundamental functionalities.

Thus, their structure can be further divided into blocks and databases. The
NRM consists of two functional blocks, respectively: monitoring block and
policy block. The QoSC contains fairness distribution block (FDB) and
two databases, respectively: DASH clients database and video stream
database. The TSM is responsible for generating the final rules for each
identified video flow. The parameters required by these rules are provided
by QoSC. The TSM module is typical component of SDN controller and by
using the OpenFlow, it controls the tasks performed by switch (e.g.
OpenFlow vSwitch). The structure of SDHN controller and the internal
relationship between its components are presented in Fig. 1. The elements
which use (are compatible with) the SAND principles are indicated by black
backgrounds.
SDHN controller
SAND messages
DASH clients database
optional GUI/CLI
video streams database
Northbound (REST API)
Fairness distribution
block (FDB)
policy block
QoS controller (QoSC)
Network resources module (NRM)
Traffic shapping module (TSM)

monitoring block
Southbound (OpenFlow)
klient DASH
gateway/home network
Fig. 1. The architecture of the proposed SDHN solution.
3.1 Network Resources Module

The main task associated with this module is gathering information about
all the traffic within a home network. The source of data is gateway and
mutual communication is performed based on the OpenFlow protocol. The
monitoring block, which is a part of this module, is also responsible for
classification of video flow. This task included the identification of new flow
and termination of given video transmission. Considering that target area of
application is a home
0
network, the identification process base on a destination IP address. The
second block under this module is the policy block. In general case, it
includes any policy associated with DASH service. A common use of this
module is to implement the predefined Service Level Agreement (SLA),
resulting from a specific network operator tariffs. The software
implementation of the policy block allows to define restrictions on whole
DASH service or on the individual DASH clients.
3.2 QoS Controller

The QoS controller is the element that has benefited greatly from the SAND
architecture. Two of its elements directly use the Status and Metrics
messages as defined in SAND specification [10]. These messages are
exchanged between DASH clients and QoSC. In the case of proposed
solution, the representation of SAND DANEs are DASH clients database
and video streams database. The first database consists of parameters of all
registered clients (set of operating points such as desired bandwidth and
quality) but also their QoS metrics (average throughput, buffer level, initial
playout delay). The second database comprise of data originated from the
MDFs associated to requested video flow (stream structure, chunk
descriptions, OPs parameters). The third component of the QoSC, fairness
distribution block (FDB) uses these databases to calculate fair provisioning
of available bandwidth.
Fairness Distribution Block. The algorithm implemented in the FDB is

based on the utility function (UF) described in our earlier research [14].
Generally, this utility function is non-linear function which represents the
relationship between video stream bitrate and relative utilization (RU)
indicator. The RU indicator represents the ratio of the number of the OPs
that DASH client can (or wishes to) received to all available OPs for given
SVC stream. One of the most important advantages of the method of
creation of the UF is fact that it requires only the data contained in the
MDF file.
Next task of the QoSC is to define the set of stream bitrates that ensures
fairness distribution of available bandwidth across all registered DASH
clients. Other words, the QoSC should find the optimal value of Yoptimal
satisfying the Formula (1).
Yoptimal = max[min(UFi(xi))] Σ(xi) ≤ Blimit (1)

i
where UFi be a utility function, xi the bitrate for i-th video stream and Blimit
represents the total amount of available bandwidth for all streams.
To solve this optimization problem, the integer-programming algorithm
was used [20]. We discussed this algorithm in our earlier article [14]. The
results are the optimum bitrate values. However, these values do not
necessarily correspond to the OPs bitrates defined within SVC stream. For
this reason, when it is necessary, calculated values of bitrate must be
changed to the nearest, lower value of the bitrate associated with the
particular OP. Finally, the calculated
values are sent to the DASH clients as SAND PER Messages (message:
Share- dResourceAssignment) [10]. Summarizing, the QoSC block performs
sequentially the following three tasks:
– reads the data contained in DASH clients and video stream databases,
creates the utility function,
– finds solution to optimization problem (Formula (1)),
– downgrades the optimal bitrates to the bitrate (lower) of the closest OP.
4 Test Scenario, Evaluation and Results

The proposed SDHN solution has been tested against the advantages it
brings in comparison to the standard DASH solutions. Thus, two software
solutions were configured. The first one was consistent with DASH
principles. This means that DASH clients were responsible for the process of
dynamic adaptation according to DASH standard and no traffic shaping
algorithms were implemented inside the home network. This solution, later
in the article, is referred to as legacy DASH. The second one is the
implementation of the proposed SDHN and it introduced two additional
elements, the SDN controller (QoSC and TSM modules) and databases
(NRM module), respectively. Both new components were implemented as
Docker containers. This solution is referred as SDHN DASH.
Considering specific features of home networks in context of the video
streaming services, we decided to conduct the two, separate tests. The first
test (test A) assumed that a couple of DASH clients were switching on and
off (playback/pause) during the whole period of test. The second test (test
B) was a representation of continuous video playback of all clients, during
the entire test. These tests correspond to typical scenarios in the home
networks with multiple video receivers.
4.1 The Testbed Configuration

In order to assess the effectiveness of the SDHN solution as well as to
compare it with legacy DASH, we developed a evaluation testbed as shown
in Fig. 2.
All the network components were connected to the TP-LINK TL-
WR1043ND home switch. That switch served as a gateway. The software
platform for the operation of the switch was OpenWRT with OpenFlow 1.3
[21]. The bandwidth on the link between gateway and HTTP server which
store SVC videos, was limited to 18 Mb/s by means of TC (Linux traffic
control tool) rules. DASH Server was implemented based on DASH-JS
library [22]. The implementations of SDHN controller was divided into two
Docker containers. All of them were installed automatically, from local
repository configured on the x86 64-bit, 8 GB RAM PC. All necessary
resources (images, dockerfiles, docker-compose files) can be stored on any
local or remote repository (e.g. DockerHub) and installation on Docker-ready
systems requires no user intervention. The testbed includes a few DASH
client. The base for them was the scootplayer [23], the an experimental
MPEG-DASH request
Video 1
scootplayers scootplayers SDHN controller
client DASH client DASH

1080p 1080p databases
720p 720p NRM container

server DASH
DASH-JS 360p 360p
PC + LXD PC + LXD
ONOS
MDF files
QoSC & TSM container
video SVC
(chunks) limiter scootplayer remote control
Openflow + OpenWRT
12 Mb/s Docker local repository
IMQ + TC PC + Docker
PC TP-LINK switch
Fig. 2. The structure of the testbed.
engine with support for accurate logging. It was modified to support the
SAND messages listed in Sect. 3. The DASH clients were implemented as
separate, virtual instances of LXD (Linux Container Deamon) containers.
All tests (the clients activities) were orchestrated by remote control scripts.
The “Elephants Dreams” video sequence (1080 p, 24 f/s) was used for all
tests. The video was encoded using the reference software JSVM (Joint
Scalable Video Model) what resulted in two SVC video files with resolutions
1080 p and 720 p, respectively. In the encoding process the Profile High was
used. The parameters of the individual layers within the SVC stream were
as follows:
– three temporal layers (T0 - 6 f/s, T1 12 f/s and T2 24 f/s),

– four SNR(MGS) layers (initial QP: SNR0 - 42, SNR1 - 42, SNR2 - 32
and SNR3 - 32).
4.2 Evaluation Metrics

To evaluate SDHN solution, we choose three metrics. Their aim were to
reflect the feature of our video distribution system when multiple SVC-
DASH video flows share the same bottleneck link.
Fairness index. Publications of recent years bring many proposal of metrics

for fairness assessment [17, 24]. We decided to base on the wide-used Jains
Index [25] with modification which reflects the structure of SVC video
stream. The fairness index FairI according to Jains approach is expressed
by the Formula (2).
Σn 2
Σi=1 ai
FairIa = n
(2)
n · i=1 2
ai
Here, n is the number of competing users. The parameter a signifies the
fairness metrics which depends upon the application. In the case of SVC
streams, the metric can be associated with fraction of demand. The DASH
adaptation process causes the change of OP (change of demand). Also,
clients have unequal demands. Thus it is reasonable to measure fairness by
closeness of the allocations
to respective demands. As the result, we propose the following definition of the

fairness metric 3.
OPi if OPi < OPmax i
ai = OPmaxi (3)
1 otherwise
Here, OPmaxi is the number of available (accepted) OPs inside the SVC
stream. In other words, it is the maximum demand of ith DASH client. OPi
is the number of received OP by ith DASH client.
Fluctuation index. Based on our experience [14], the bitrate fluctuations

(switches between video representation) have great influence on the
perceived quality (QoE) of video on clients’ side. The system for
distribution of video in the DASH service should prevent or at least
minimize the occurrence of bitrate- switches. To asses this issue, the
Fluctuation Index FlucI has been used. Because the bitrate-switches (the
OPs switches) may occur only between chunks thus the index is defined by
the Formula 4.
n
n
FlucI
Σ Cn (4)
F
= i=1
Here, Fn is the number of switches for SVC video stream n and Cn is a total
number of received chunks.
Link Utilization Index. It is obvious that the limited capacity of outgoing

link should be utilized as good as possible. In order to asses this issue, we
used well-known idea of measure of utilization on a network layer. Thus, the
Link Utilization Index LUI is defined by the Formula 5.
Σn
Bn
i=1
LUI = Blimit (5)
Here, Blimit is the bandwidth of the bottleneck and Bn is the summary

bitrate of all SVC video streams.
4.3 Results of SDHN Evaluation

The evaluation carried out included five DASH clients. Two of them were
configured for the maximum 1080 p video resolution and others, for the
video of maximum 720 p resolution. Both tests (A and B) were repeated
fifty times. In the case of Test B (SDNH DASH), five schedules of clients’
activities were used (each repeated ten times). Every single test lasted 150 s
(the fifty chunks, each 5 s long). The example of such schedule and
corresponding bitrates of video streams are presented in Fig. 3.
The schedule on the Fig. 3 defines two strong congestions. First one
(between 30th and 90th second) grows slowly while the next one (between
100th and 135th second) is of the burst type. The behaviour of legacy
DASH and SDHN DASH solution is presented in Fig. 4.
Video
Fig. 3. The example of the schedule of the test B (left) and corresponding bitrates of
the SVC video streams received by five DASH clients (right).
Fig. 4. The utilized bandwidth, the total demand on bandwidth and the bottleneck
capacity for two video distribution systems: legacy DASH (upper) and proposed
SDHN DASH (lower).
In the most of tested cases, the legacy DASH was sensitive to

congestions. It is worth noting that the highest levels of overload (resulting
from exaggerated trials of adaptation) were seen when congestions had been
disappearing (between 85th and 100th second). As it was expected, the
SDHN DASH system was able to
calculate near-optimal bitrate values what in combination with playout

buffer, allowed for a significant reduction of fluctuations (number of
switching). An additional effect is much better link utilization. The values
of the Fluctuation Index and Link Utilization Index, for two compared
video distribution system and for all 50 test, are presented in the Table 1.
Table 1. The mean values of the FlucI and LUI indexes.
Scenario – system Indexes

FlucI LUI
Test A – legacy DASH 2.39 0.93
Test A – SDHN DASH 0.16 1.0
Test B – legacy DASH 2.88 0.88
Test B – SDHN DASH 0.89 0.94
In all the tests conducted, the proposed SDHN solution demonstrated

superiority over the standard DASH system. Particularly, in the case of
continuous playback, SDHN offers almost perfect solution. Of course, the
100% utilisation of the link capacity is not always possible due to the
parameters associated with chunks (parameters of the video encoding).
The fairness defined according to the formulas 2 and 3 proved to be
better in the case of proposed solution. The graph of the values of the FairI
index value during test B (the test according to schedule presented in Fig.
3) is shown in Fig. 5.
Fig. 5. The fairness Index for the test B.
For the standard DASH distribution system (legacy DASH), the

appearance of congestion results in significant fairness degradation. In case
of relatively longer congestions, the fluctuations of FaitI is also observed.
These phenomena are much weaker for proposed SDHN system.
5 Conclusion and Future Work

The combination of the SVC video coding and DASH principles is undoubt-
edly a step towards greater flexibility of video streaming services. However,
this
Video
approach requires the new adaptation methods which are able to provide
the high level of fairness during the process of video distribution. This issue
has been analysed from the providers perspective. The article presented the
different point of view. We proposed the video distribution system based
on DASH and employing SDN which is dedicated to the typical home
network. Through a combination of previously developed utility function for
SVC-DASH and SAND architecture, we elaborated the software solution
which can be implemented in the relatively simple SDN controller.
Therefore, it was possible to orchestrate and implement the proposed
solution by means of Docker containers. Home network users can
automatically install our SDHN solution if only have Openflow compliant
switch. No additional, specialized hardware is required beyond the typical
PC. In order to make the proposed solution even more practical, we are
working on a version optimized for single-chip computers and selected,
commercial gateways.
The tests carried out have shown the superiority of proposed SDHN
solution in terms of the guarantee fairness distribution of SVC video
streams and the use of available bandwidth on a link to an external
network. At the same time, the SDHN allowed for substantial reduction of
bitrate switching, which is important to maintain high, perceived quality of
received video.
References
1. ISO/IEC 23009-1:2014: Information technology - Dynamic adaptive streaming
over HTTP (DASH) - Part 1: Media presentation description and segment
formats (2014)
2. Reichel, J., Schwarz, H., Wien, M., Vieron, J.: Joint Scalable Video Model 9 of
ISO/IEC 14496-10: 2005/AMD3 Scalable Video Coding. Joint Video Team
(JVT), Docs. JVT-X202 (2007)
3. Stockhammer, T.: Dynamic adaptive streaming over HTTP standards and
design principles. In: Second Annual ACM Conference on Multimedia Systems,
San Jose,
pp. 133–144. ACM (2011)
4. Akhshabi, S., Anantakrishnan, L., Dovrolis, C., Begen, A.C.: What happens
when HTTP adaptive streaming players compete for bandwidth. In: Proceedings
of ACM Workshop on Network and Operating System Support for Digital Audio
and Video, Toronto, pp. 9–14 (2012)
5. Timmerer, C., Griwodz, C., Begen, A.C., Stockhammer, T., Girod, B.: Guest
editorial adaptive media streaming. IEEE J. Sel. Areas Commun. 32(4), 681–
683 (2014)
6. Abuteir, R.M., Fladenmuller, A., Fourmaux, O.: SDN based architecture to
improve video streaming in home networks. In: Proceedings of IEEE 30th
Interna- tional Conference on Advanced Information Networking and
Applications, Crans- Montana, pp. 220–226 (2016)
7. ONF: Open Networking Foundation Software-Defined Networking (SDN) Defini-
tion. https://www.opennetworking.org/sdn-resources/sdn-definition
8. Docker Inc. and Docker Project. https://www.docker.com/
9. Akhshabi, S., Begen, A., Dovrolis, C.: An experimental evaluation of rate-
adaptation algorithms in adaptive streaming over HTTP. In: Second Annual
ACM Conference on Multimedia Systems, San Jose, pp. 157–168. ACM (2011)
10. ISO/IEC FDIS 23009-5:2016: Information Technology - Dynamic adaptive

streaming over HTTP (DASH) - Part 5: Server and network assisted DASH
(SAND) (2016)
11. Kalva, H., Adzic, V., Furht, B.: Comparing MPEG AVC and SVC for adaptive
HTTP streaming. In: IEEE International Conference on Consumer Electronics,
Las Vegas, pp. 158–159 (2012)
12. Unanue, I., Urteaga, I., Husemann, R., Del Ser, J., Roesler, V., Rodriguez, A.,
Sanchez, P.: A tutorial on H264/SVC scalable video conding and its tradeoff
between quality, coding efficiency and performance. In: Lorente, J.D.S. (ed.)
Recent Advances on Video Coding. InTech, Rijeka (2011). Chap. 1
13. Schwarz, H., Marpe, D., Wiegand, T.: Overview of the scalable video coding
extension of the H.264/AVC standard. IEEE Trans. Circ. Syst. Video Technol.
17(9), 1103–1120 (2007)
14. Przylucki, S., Czerwinski, D., Sierszen, A.: QoE-oriented fairness control for
DASH systems based on the hierarchical structure of SVC streams. In: Gaj, P.,
Kwiecien´, A., Stera, P. (eds.) Computer Networks. Communications in
Computer and Infor- mation Science, vol. 608, pp. 180–191. Springer, Cham
(2016)
15. Mustafa, B., Nadeem, T.: Dynamic traffic shaping technique for HTTP adaptive
video streaming using software defined networks. In: Proceedings of 12th Annual
IEEE International Conference on Sensing, Communication, and Networking,
Seat- tle, pp. 178–180 (2015)
16. Cetinkaya, C., Ozveren, Y., Sayit, M.: An SDN-assisted system design for
improving performance of SVC-DASH. In: Federated Conference on Computer
Science and Information Systems, Lodz, pp. 819–826 (2015)
17. Mu, M., Broadbent, M., Farshad, A., Hart, N., Hutchison, D., Ni, Q., Race, N.:
A scalable user fairness model for adaptive video streaming over SDN-assisted
future networks. IEEE J. Sel. Areas Commun. 34(8), 2168–2184 (2016)
18. Tang, S., Hua, B., Liu, S.: A heuristic algorithm for optimal discrete bandwidth
allocation in SDN networks. In: Proceedings of IEEE Global Communications
Con- ference, San Diego, pp. 1–7 (2015)
19. Xingtao, L., Yantao, G., Wei, W., Sanyou, Z., Jiliang, L.: Network
virtualization by using software-defined networking controller based Docker. In:
Proceedings of IEEE Information Technology, Networking, Electronic and
Automation Control Conference, Chongqing, pp. 1112–1115 (2016)
20. Dakin, R.J.: A tree-search algorithm for mixed integer programming problems.
Comput. J. 8(3), 250–255 (1965)
21. GitHub Inc. https://github.com/CPqD/ofsoftswitch13/wiki
22. ITEC – Dynamic Adaptive Streaming over HTTP. http://www-itec.uni-klu.ac.at/
dash/
23. Scootplayer. http://scootplayer.readthedocs.io/en/latest
24. Liu, L., Zhou, C., Zhang, X., Guo, Z.: A fairness-aware smooth rate adaptation
approach for dynamic HTTP streaming. In: IEEE International Conference on
Image Processing (ICIP), Quebec, pp. 4501–4505 (2015)
25. Jain, R.K., Chiu, D.-M.W., Hawe, W.R.: A Quantitative Measure of Fairness
and Discrimination for Resource Allocation in Shared Computer System. In:
DEC Research Report TR-301 (1984)
Teleinformatics and
Telecommunications
Minimum Transmission Range Estimation
for Vehicular Ad Hoc Networks
in Signalised Arterials
Bartl-omiej Pl-aczek(✉) and Marcin Bernas
Institute of Computer Science, University of Silesia,

Be֒ dzin´ska 39, 41-200 Sosnowiec, Poland
Abstract. In this paper the connectivity is analysed of vehicular ad

hoc networks (VANETs) in arterial roads with signalised intersections.
A method is proposed to derive an estimate of minimum transmission
range (MTR), i.e., a minimum value of the vehicles transmission range
that allows all vehicles in a given road section to be connected via
multihop communication. The introduced method takes into account
the impact of traffic signals coordination on formation of vehicle
platoons. It is applicable for both the under-saturated and the over-
saturated traffic conditions. The main contribution in this paper is
formulation of a relationship, which allows the upper bound of MTR
to be estimated for VANETs in unidirectional and bidirectional
signalized arterial roads. The conducted experiments confirm that the
derived estimate fits well with results of simulation experiments.
Keywords: Vehicular ad hoc networks · Connectivity · Minimum

transmission range · Traffic signal control
1 Introduction
Vehicular ad hoc network (VANET) is considered as an important
component of the next generation automotive systems, which enables inter-
vehicle communication for a number of safety, traffic management, and
comfort-related applications [1–4]. Using the communication between
vehicles, VANET could provide a safer and a more convenient driving
experience. From the perspective of the VANET applications development,
network connectivity is very important since it might be difficult to provide
reliable services in the case of disconnections. Due to the dynamically
changing topology of VANET, the connectivity has direct influence on
channel contention and vehicle communications [5].
Transmission range is a critical parameter, which significantly affects
VANET connectivity [6]. On the one hand, the connectivity of VANET is
expected to be improved as the transmission range increases [7]. On the
other hand, a longer transmission range in a dense wireless network can lead
to a huge amount of interferences between neighbour nodes and a high
network overhead. Thus, the
DOI: 10.1007/978-3-319-59767-6 17
210 B. P-laczek and M.
Bernas
network will suffer from inefficiency due to packet collisions and losses [8].
To alleviate this issue, the transmission range in VANET should be locally
adjusted to the current traffic situation. So far, several transmission range
adjustment techniques for VANET have been investigated [8, 9]. Such
techniques require accurate estimation of a transmission range value, which
guarantees connectivity of all vehicles in a considered area.
In this paper, the minimum transmission range (MTR) is analysed for
VANET in arterial roads with signalised intersections. MTR corresponds to
the minimum common value of the vehicles transmission range, which
enables all vehicles to be connected via multihop communication in a single
network clus- ter. A method is proposed to derive an estimate of the upper
bound MTR. The analytical estimate of the upper bound MTR is verified in
a simulation environment for unidirectional and bidirectional signalized
arterial road.
The paper is organized as follows. A short review of the related literature
is provided in Sect. 2. The method of MTR estimation is introduced in Sect.
3. Simulation experiments and their results are described in Sect. 4. Finally,
conclusions are given in Sect. 5.
2 Related Works and Contribution

A number of efforts have been already accomplished to investigate the MTR
for VANET in highways. In [10] a cellular automata model of the
uninterrupted highway traffic was considered and an analytical lower-bound
for the MTR was derived. It was shown that a non-homogeneous
distribution of vehicles, due to traffic jams, results in the need for a higher
transmission range in comparison to that required for homogeneous vehicles
distribution. Based on the above- mentioned results, a traffic density
estimate was used in [11] to develop an algorithm, which adjusts the
transmission range of vehicles dynamically according to the local traffic
conditions on a highway.
The multihop connectivity of one-dimensional VANET, with given either
vehicle locations or traffic densities, was discussed in [12]. A model was
proposed, which can be used as a guideline for determining the required
transmission range, in order to achieve very high network connectivity.
However, that approach is limited to simple traffic scenarios without traffic
lights.
Inter-platoon connectivity for a bidirectional highway was analysed in
[7]. The authors have assumed that all vehicles drive in platoon-based
patterns, which facilitate better traffic performance as well as information
services. Impact was investigated of the transmission range on expected time
of safety message delivery among platoons. The simulation experiments have
shown that the transmission range is a critical parameter, which
significantly affects the inter-platoon communication in VANET.
The highway scenario was also considered in [9] to introduce a
transmission range control algorithm for safety applications. According to
that algorithm, the transmission range was adapted by taking into account
vehicle density, VANET load and packet generation rate of each vehicle.
Another scheme, which enables
Minimum Transmission Range Estimation 211
0
dynamic transmission range adjustment, based on both local vehicle density
and distance to destination, was presented in [8]. That study was devoted to
the broadcast-type communication between vehicles equipped with
directional antennas.
A connectivity robustness model was proposed in [13] to assess vehicle-
to- vehicle communication in connected vehicle environments. That model
accounts for the transmission range and multiple other factors affecting the
connectivity in VANETS. The results in [13] confirm that overall robustness
of connectivity increases for higher values of transmission range.
In [14] a cellular automata approach was applied to construct an urban
traffic mobility model. Based on the developed model, characteristics of
global traffic patterns in urban areas were analysed. It was shown that even
though the inter- vehicle spacing of both highway and urban traffic can be
approximated by the exponential distribution, the connectivity pattern of a
vehicle is very different in these two scenarios. The simulation results
presented in [14] indicate that the parameters of traffic control at signalised
intersections such as cycle duration, green split, and coordination of traffic
lights have a significant effect on inter- vehicle spacing distribution and the
required transmission range.
A simulation environment was also used in [6] to investigate the MTR
for a road network with signalised intersections. Based on the simulation
experiments, it was shown that the signalised intersections cause
significant fluctuations of inter-vehicle distance and thus the signal-
controlled traffic requires higher communication ranges to ensure
connectivity. The higher communication range values were needed to
connect the vehicles stopped at traffic lights with those moving freely.
According to the authors’ knowledge, the problem of analytical MTR
determination for arterial streets with signalised intersections has not been
studied so far in the VANET-related literature. The main contribution in
this paper is a relationship formulation, which allows the upper bound of
MTR to be estimated for VANETs in signalised arterial roads.
3 A Method for Minimum Transmission Range

Estimation
In arterial streets with traffic signals, vehicles are grouped into platoons at
exits of the signalised intersections. The vehicle platoons move from one
signalized intersection to the next. A well-coordinated traffic signal system
moves platoons of vehicles so that they arrive during the green phase of the
downstream intersection. In order to evaluate the minimum transmission
range, which allows maintaining connectivity of all vehicles in a signalised
arterial road, it is necessary to consider the distances between the vehicle
platoons.
In Fig. 1 the progression of platoons along a signalised artery is
illustrated by means of time-space diagrams. Vehicle trajectories are plotted
on the time-space diagrams as the sloped lines. Stopped vehicles are shown
as horizontal lines. The horizontal bars show indications of the traffic signals
at intersections. An example
Bernas
of the time-space diagram for unidirectional artery is presented in Fig. 1(a).
In this example the vehicles move northbound (from bottom to top of the
diagram). A time-space diagram of bidirectional artery is presented in Fig.
1(b), where the dashed lines correspond to the vehicles that moves
southbound (top to bottom). It should be noted that the vehicle trajectories
in time-space diagrams (Fig. 1)
are idealized for presentation purposes. The slope of the trajectories in these
diagrams corresponds to space-mean speed of the vehicles. Vertical distance
between the trajectories represents average space headway between vehicles.
Moreover, for sake of clarity, the examples in Fig. 1 do not include vehicles
travelling outside platoons. The simplified presentation of time-space
diagram, which is based on the average traffic stream parameters, does not
involve loss of generality. The simulation model, which was used for
evaluation of the proposed method (Sect. 4), involves random variations of
vehicle speed, presence of vehicles outside platoons as well as platoon
dispersion.
For the unidirectional artery (Fig. 1(a)), the longest transmission range
is necessary to connect the vehicle platoons, when one platoon approaches
the end of the considered road section (signal 3) and the second platoon
enters this road section (approaches signal 1) Thus, the MTR can be
evaluated according to the following formula:
r = d − w, (1)
where d denotes length of the considered road segment and w is length of
the vehicle platoon.
In case of the bidirectional arterial road (Fig. 1(b)), the longest
communication range is necessary to connect the two vehicle platoons that
move in opposite directions and approach ends of the analysed road section.
It means that lengths of the two platoons (w1 and w2) have to be taken into
account when evaluating the MTR. Thus, Eq. 1 is used for this case with w
= w 1 + w 2.
Fig. 1. Minimum transmission range in signalized arterial: (a) unidirectional traffic,

(b) bidirectional traffic.
Length of a vehicle platoon (w) can be evaluated as follows:

w = (n − 1) · h, (2)
where n is number of vehicles in the platoon and h is average space headway

between vehicles. The average space headway is equal to the inverse of
traffic density (h = 1/k). When taking into account the fundamental
equation of traffic flow [15], the average space headway can be expressed as:
h = 1/k = v/s, (3)
where k is traffic density, v is space-mean speed of the vehicles, and s is

saturation flow rate, i.e. the maximum number of vehicles that can pass
through the intersection during a time unit, when the traffic signal is green.
The saturation flow rate is used in (3) because the platoon is created if a
group of vehicles leave an intersection when the traffic signal turns green. It
should be also noted that in Eq. (1) the length of a single vehicle is ignored
(w = 0 if n = 1) for sake of simplicity. Substituting (3) into (2) yields,
w = max(0, n − 1) · v/s (4)
The platoon is a group of vehicles leaving an intersection when red light

ends. Therefore, the number of vehicles in the platoon (n) can be estimated
as a number of vehicles that have been stopped at an intersection due to the
red signal. Assuming that R is duration of the effective red signal and q is
intensity of incoming traffic, the number of vehicles stopped during time
period can be calculated by employing an analytical approach with fluid
approximation, as depicted in Fig. 2.
Fig. 2. Number of vehicles stopped at an intersection due to red signal (n). (Color
figure online)
Based on the arrival curve and the departure curve presented in Fig. 2,
it can be found that the total number of stopped vehicles, which
corresponds to the number of vehicles in the platoon, can be estimated as
follows:
R·q
n= (5)
1 − q/s
Bernas
The above considerations are related to the MTR estimation for under-
saturated traffic conditions, whereby vehicle queues are only created during
the red signal and are dissolved during green signal. In oversaturated traffic
conditions, when the intensity of traffic flow (q) exceeds the intersection
capacity (c), vehicle queues fill approaches of the intersection and spill over
from one intersection to another. In such case, the queues persist and
cannot be cleared totally during green signal.
According to the proposed approach, the MTR for oversaturated traffic
is estimated based on the length of the widest gap between vehicles, which
can be created when vehicles leave queue during green signal. The expected
number of vehicles that pass an intersection during effective green time (G)
is G · s. Aver- age road length occupied by one vehicle in the queue equals
1/kjam, where kjam denotes maximum vehicle density in traffic jams. Thus,
the MTR for oversaturated traffic conditions is estimated as G · s/kjam.
Finally, based on the above considerations, the upper bound of MTR for
VANET in a signalised arterial road of length d is given by
⎧ R·q
r(q) = d−
⎨ 1− (6)
q/s — 1 · v/s, q ≤
c,
⎩
G · s/kjam, q > c,
where c denotes capacity of the signalized intersection and the remaining

sym- bols were defined earlier in this section. The capacity is calculated by
using the following formula:
G
c= s·
R+ . (7)
G
4 Evaluation in Simulation Environment

The analytical relationships that were derived in the previous section are
verified through simulations of vehicle traffic for both unidirectional and
bidirectional arterial road.
4.1 Simulation Setup

In this study the stochastic cellular automata (CA) model of urban traffic
was used for the traffic simulation [16]. Each traffic lane is represented by a
one- dimensional cellular automaton. The traffic lanes are divided into cells
of equal lengths (7.5 m). An occupied cell in the cellular automaton
corresponds to single vehicle. At each discrete time step (1 s) the state of
CA is updated according to four steps (acceleration, deceleration, velocity
randomization, and movement). These steps are necessary to reproduce the
basic features of real urban traffic. Step 1 represents driver tendency to
drive as fast as possible; step 2 involves reaction to other vehicles and to
traffic lights; step 3 introduces non-deterministic acceleration due to random
external factors and the overreaction of drivers while
slowing down. Finally, in step 4 the vehicles are moved according to the
new velocity calculated in steps 1–3. Steps 1–4 are applied for all vehicles.
According to the definition of step 3, the speed of all vehicles that have a
velocity of at least 1, is reduced by one unit with a probability of p.
Parameter p is referred to as the deceleration probability. It should be
noted that without step 3, the model is deterministic [17]. Detailed
definitions of steps 1–4 can be found in [14].
A schema of the simulated arterial road, with eight signalized
intersections, is shown in Fig. 3. Total length of the arterial road is 7.5 km
(1000 cells). The above-mentioned CA model of the arterial road was
implemented in Matlab. The simulations were performed for two scenarios.
Unidirectional traffic was considered in Scenario 1 and bidirectional traffic
was analyzed in Scenario 2. Simulation data were collected in a section
between 4th and 6th intersection (the shaded area in Fig. 3).
Fig. 3. Simulated arterial road.
Intensity of the traffic flow was determined for the network model by
parameters qA and qB (in vehicle per second). At each time step vehicles
were randomly generated with a probability equal to the intensity qA or qB
in the entrances of the road model, as shown in Fig. 3. As a result,
binomially distributed traffic flows were obtained that approximate a
Poisson distribution. The simulation period was 10000 s for each considered
combination of the traffic intensities. The traffic signal control was
simulated assuming the green wave coordination strategy. Detailed
parameters of the simulation model are presented in Table 1.
4.2 Experimental Results

Simulation experiments were performed to determine the MTR needed to
connect all vehicles in the analysed road section. In general, the MTR is
calculated
Bernas
Table 1. Parameters of simulation model.
Parameter Value
Total road length 1000 cells
Length of analyzed road section (d) 222 cells
Distance between intersections 111 cells
Effective green time (G) 38 s
Effective red time (R) 38 s
Mean free-flow speed (v) 2.8 cells per second
Maximum speed 3 cells per second
Deceleration probability 0.2
Saturation flow rate (s) 0.54 vehicles per second per lane
Capacity of intersection (c) 0.27 vehicles per second per lane
Maximum vehicle density (kjam) 1 vehicle per cell
by constructing the minimum spanning tree in a graph K(N, E), which rep-
resents the VANET in the considered road section [10]. The vertices N in K
correspond to vehicles and edges E represent communication links. An
edge e = (ni, nj) exists in K if the distance between vehicles i and j is
shorter than or equal to min(ri, rj), where ri, rj denote transmission ranges
of vehicles i and j, respectively. The MTR is the longest edge in the
minimum spanning tree. In a single-lane VANET, the MTR is the widest
gap between any two consecutive vehicles. The MTR was computed at each
time-step of the simulation and stored along with the intensity of traffic
flow.
Results of both the simulation experiments and the analytical MTR
evaluation for unidirectional traffic (Scenario 1) are compared in Figs. 4, 5
and 6. The
Fig. 4. MTR in unidirectional signalized artery (Scenario 1, qB = 0 veh./s)

scatter plots present the MTR values that were determined for various
traffic intensities during simulation. The curves in the plots correspond to
the upper bound of MRT, which was estimated by using Eq. (6). These
plots show clearly that the MTR values are concentrated below the derived
upper bound for all analysed intensities of traffic flow.
Fig. 5. MTR in unidirectional signalized artery (Scenario 1, qB = 0.5qA).
Fig. 6. MTR in unidirectional signalized artery (Scenario 1, qA = 0.1 veh./s).
The results in Fig. 4 were obtained assuming that there are no vehicles
turn- ing at intersections from the side roads into the arterial road (qB = 0
veh./s). In this case, without the traffic inflows from side roads, only free-
flow traffic conditions are observed. Subsequently, the intensity of the traffic
entering arterial from side roads was set to half of the traffic intensity on
the arterial road,
Bernas
i.e., qB = 0.5 · qA (Fig. 5). For qA = 1.8 veh./s such setting led to
oversaturated traffic conditions since the total traffic intensity (qA + qB =
1.5 · qA) exceeds the capacity of intersections (c = 0.27 veh./s). This causes
a significant decrease of the MTR, as shown in Fig. 5.
The performed MTR analysis has also taken into account different ratios
of the side inflow intensity and the arterial traffic intensity (qB/qA). Figure 6
shows the MTR values that were obtained for qA = 0.1 veh./s and qB
between 0 and
0.1 veh./s. It can be observed in these results that the MTR decreases when
the side inflow intensity increases. The shorter MTR for higher qB
intensities are caused by the vehicles from side roads that join the platoons
on the arterial road and extend their lengths.
Fig. 7. MTR in bidirectional signalized artery (Scenario 2, qA = qB).
The MTR values determined for the bidirectional traffic (Scenario 2) are
shown in Figs. 7 and 8. It should be noted that in this scenario qA and qB
denote intensities of the traffic in arterial road for both directions. The
traffic inflows from side roads are not present. The results in Fig. 7 were
obtained for equal intensities of the traffic in both directions (qA = qB). The
MTR values for different qA and qB intensities are presented in Fig. 8. In
general, the results confirm that the proposed method enables correct upper
bound evaluation for MTR also in case of the bidirectional arterial road.
When comparing the results obtained for Scenario 2 with those of
Scenario 1 it can be observed that the MTR values decreases faster with the
intensity for the bidirectional traffic. The reason is that the number of
vehicles in the analysed section of the bidirectional road is higher than in
the unidirectional road and the distances between platoons are shorter, as
shown in Fig. 1b.
Fig. 8. MTR in bidirectional signalized artery (Scenario 2, qB = 0.1 veh./s).
5 Conclusion
Simulation results show that the proposed method allows the upper-bound
MTR to be accurately estimated for VANETs in unidirectional and
bidirectional signalized arterials. The derived MTR estimate fits well with
the results of simulation experiments for a wide range of traffic intensities.
In case of under-saturated traffic, the MTR decreases when the traffic flow
intensity increases. In comparison to the under-saturated traffic, a
significantly lower MTR value is obtained for the over-saturated traffic
conditions. For both the under-saturated and the over-saturated traffic
conditions, the traffic signals timing has a strong impact on the estimated
MTR. When the MTR is considered for under-saturated traffic, the distance
between intersections is an additional important factor, which has to be
taken into account. It was also observed that in case of bidirectional traffic
the VANET connectivity is improved by lowering the MTR requirements.
In practice this method may be used by control nodes at intersections
that are equipped with wireless communication and collect data reported by
vehicles via VANET in order to optimize the traffic signals. Such control
nodes could evaluate the MTR and broadcast this information to the
vehicles. The proposed approach can be combined with advanced data
collection strategies [18]. Further research will be necessary to integrate the
method into a dynamic range control protocol for VANETs. Another
interesting topic for the future research is to use the stochastic geometry
model [19] for the MTR analysis.
References
1. Al-Sultan, S., Al-Doori, M.M., Al-Bayatti, A.H., Zedan, H.: A comprehensive
survey on vehicular ad hoc network. J. Netw. Comput. Appl. 37, 380–392
(2014)
2. P-laczek, B.: Selective data collection in vehicular networks for traffic control
applications. Transp. Res. Part C: Emerg. Technol. 23, 14–28 (2012)
Bernas
3. Berna´s, M.: VANETs as a part of weather warning systems. In: Kwiecien´, A.,
Gaj, P., Stera, P. (eds.) CN 2012. CCIS, vol. 291, pp. 459–466. Springer,
Heidelberg (2012). doi:10.1007/978-3-642-31217-5 48
4. Berna´s, M.: WSN power conservation using mobile sink for road traffic
monitoring. In: Kwiecien´, A., Gaj, P., Stera, P. (eds.) CN 2013. CCIS, vol.
370, pp. 476–484. Springer, Heidelberg (2013). doi:10.1007/978-3-642-38865-1 48
5. Shao, C., Leng, S., Zhang, Y., Vinel, A., Jonsson, M.: Performance analysis of
connectivity probability and connectivity-aware MAC protocol design for
platoon- based VANETs. IEEE Trans. Veh. Technol. 64(12), 5596–5609 (2015)
6. Kafsi, M., Papadimitratos, P., Dousse, O., Alpcan, T., Hubaux, J.P.: VANET
connectivity analysis. In: IEEE Workshop on Automotive Networking and
Applica- tions (Autonet), no. LCA-CONF-2009-001 (2008)
7. Jia, D., Lu, K., Wang, J.: On the network connectivity of platoon-based
vehicular cyber-physical systems. Transp. Res. Part C: Emerg. Technol. 40,
215–230 (2014)
8. Soua, A., Ben-Ameur, W., Afifi, H.: Beamforming-based broadcast scheme for
multihop wireless networks with transmission range adjustment. In: 10th Annual
Conference on Wireless On-Demand Network Systems and Services (WONS),
pp. 107–109 (2013)
9. Javed, M.A., Khan, J.Y.: Performance analysis of an adaptive rate-range control
algorithm for VANET safety applications. In: 2014 International Conference on
Computing, Networking and Communications (ICNC), pp. 418–423 (2014)
10. Artimy, M.M., Robertson, W., Phillips, W.J.: Minimum transmission range in
vehicular ad hoc networks over uninterrupted highways. In: 2006 IEEE
Intelligent Transportation Systems Conference, pp. 1400–1405 (2006)
11. Artimy, M.: Local density estimation and dynamic transmission-range
assignment in vehicular ad hoc networks. IEEE Trans. Intell. Transp. Syst. 8(3),
400–412 (2007)
12. Jin, W.L., Recker, W.W., Wang, X.B.: Instantaneous multihop connectivity of
one- dimensional vehicular ad hoc networks with general distributions of
communication nodes. Transp. Res. Part B: Methodol. 91, 159–177 (2016)
13. Osman, O.A., Ishak, S.: A network level connectivity robustness measure for
connected vehicle environments. Transp. Res. Part C: Emerg. Technol. 53, 48–
58 (2015)
14. Tonguz, O.K., Viriyasitavat, W., Bai, F.: Modeling urban traffic: a cellular
automata approach. IEEE Commun. Mag. 47(5), 142–150 (2009)
15. Dhingra, S.L., Gull, I.: Traffic flow theory historical research perspectives.
Trans- portation Research E-Circular, E-C149 (2011)
16. Maerivoet, S., De Moor, B.: Cellular automata models of road traffic. Phys. Rep.
419(1), 1–64 (2005)
17. P-laczek, B.: A traffic model based on fuzzy cellular automata. J. Cell. Automata
8(3–4), 261–282 (2013)
18. P-laczek, B., Bernas, M.: Uncertainty-based information extraction in wireless
sen- sor networks for control applications. Ad Hoc Netw. 14, 106–117 (2014)
19. Steinmetz, E., Wildemeersch, M., Quek, T.Q., Wymeersch, H.: A stochastic
geometry model for vehicular communication near intersections. In: Globecom
Workshops (GC Wkshps), pp. 1–6. IEEE (2015)
The Possibilities and Limitations
of the Application of the Convolution
Algorithm for Modeling Network Systems
Adam Kaliszan(✉) and Maciej Stasiak
Chair of Communication and Computer Networks,

Faculty of Electronics and Telecommunications,
Poznan University of Technology, Poznan´, Poland
{adam.kaliszan,maciej.stasiak}@put.poznan.pl
http://nss.et.put.poznan.pl/
Abstract. This article verifies the applicability of the convolution algo-

rithm for modelling multi-service network systems with different call
streams, including non-Poissonian streams. The study includes inves-
tigations of traffic streams that are state-dependent and independent.
The service time is determined by the following distributions: Pareto,
uniform, gamma, normal and exponential. The results of the
simulation experiments are compared with the results of modelling by
convolution algorithm. The results obtained in this study allow limits
for the applicability of the convolution algorithm to be determined
and identified.
1 Introduction
The works [1, 2] propose a convolution algorithm to model full-availability
resources that are systems with state-independent call admission process. In
[3– 7], a modification to convolution algorithms for modelling a number of
selected systems with state-dependent call admission process, such as: the
full-availability group with reservation [2], limited-availability group [4],
threshold systems [4] and overflow systems [6] is presented. In all the above
mentioned works the application of the convolution algorithm is limited to
model systems that are offered mutually independent classes of traffic
generated by a finite (Engset traffic, Pas- cal traffic) or infinite (Erlang
traffic) number of traffic sources. A review of the literature also provides a
number of methods in which methods other than convolution algorithms
are used to model systems with finite and infinite number traffic sources and
different new call admission mechanisms [8]. However, the issue of a possible
application of convolution algorithms to model call streams other than
Poisson call streams has not been addressed as yet.
Hanczewski et al. [7] proposes the model of a queueing system with
cFIFO service discipline and considers an application of a convolution
algorithm to model systems which are offered traffic classes with other than
exponential distributions of time between subsequent events. The model
for cFIFO queueing systems is an approximated model and hence on the
basis of the results presented in [7] it is not possible to unequivocally
evaluate the influence of non-Poissonian call streams on the accuracy of the
convolution algorithm.
Bernas
DOI: 10.1007/978-3-319-59767-6 18
222 A. Kaliszan and M. Stasiak
Until now, no analysis of a possibility to model full-availability

resources, the so-called full-availability group), with non-Poissonian classes
of offered call streams has been carried out. This fact has induced the
authors to start a further investigation into available possibilities to apply
a convolution algorithm to analyse full-availability systems in which call
streams are determined on the basis of one of the following distributions:
uniform, normal, gamma, Pareto or exponential. The objective of the study
was then to determine limitations in the application of these distributions in
modelling telecommunications systems by convolution methods.
This article is structured as follows. Section 2 presents the traffic
parameters that describe the full-availability group, while Sect. 3 provides a
description of a call class with non-Poissonian streams. Then, Sect. 4
discusses the convolution algorithm for modelling a full-availability group
with any call classes. Section 5 presents a simulation method for a
determination of occupancy distributions in single-service systems with non-
Poissonian traffic call streams. Section 6 compares the results of analytical
calculations with the results of simulation experiments for different
(diverse) distributions that describe call streams. Finally, Sect. 7 is a brief
summary chapter that sums up the article.
2 Full-Availability Resources (The Full-Availability

Group)
The basic model of resources with multi-service traffic is the so-called full-
availability group. This group is offered m call classes from the set M that
correspond to the services offered by a telecommunications system under
investigation. In a full-availability group, a call of a given class is always
admitted for service if only the group has unoccupied resources required for
this call to be serviced. Each call class is characterised by a different
number of resources di that are needed for a single call to be serviced, which
is expressed in AU (Allocation Units). While constructing multi-service
models for telecommunications systems it is assumed that the AU is the
greatest common divisor (or its multiplicity) of the bit rates of all calls
offered to the system [7]. After a determination of a given value that
corresponds to the AU it is also possible to express the capacity of a full-
availability group in AUs.
A full-availability group for a large number of Erlang traffic classes of
Erlang type can be described by the model [9, 10]. In the case of many
other traffic classes (Engset, Erlang and Pascal), the so-called BPP traffic
(after the names of call streams in Engset, Erlang and Pascal traffic, i.e.
Bernoulli, Poisson and Pascal traffic streams, respectively), the model [11,
12] or the model [13] can be used. In addition, the convolution algorithm [1,
2] can be used to model a multi- service full-availability group with BPP
traffic. The results of studies presented in [9–12, 14–21] are based on an
analysis of multi-dimensional Markov processes that occur in the full-
availability system and therefore cannot be used to model non-Poissonian
(i.e., those that are different from BPP) call streams. A possibility to
apply convolution algorithms for modelling multi-service full-availability
groups with non-Poissonian call streams have not been analysed as yet.
The Possibilities and Limitations of the Convolution 223
Algorithm
A full-availability group for one traffic class is described by the Erlang
model (for Erlang traffic: Poisson stream, exponential service time) [22],
Engset (for Engset traffic: Bernoulli call stream, exponential service time)
[23] or Pascal (model) (for Pascal: Pascal call stream, exponential service
time) [24].
Fig. 1. A diagram of the full-availability group with multi-service traffic
A diagram of the full-availability group is shown in Fig. 1. In line

with the adopted notation, fi(t) is the function of time distribution between
subsequent calls of class i. The parameter xi in function fi(xi, t) indicates the
dependence of a distribution on the number xi of serviced calls of class i. For
these distributions that are dependent on the number of serviced calls, the
number of traffic sources Si is a significant parameter. The parameter µi
defines the average service intensity for calls of class i. In traffic
engineering issues, particular traffic classes are characterised by the
following parameters: the average traffic intensity Ai and the average
number of AUs demanded by a single call of a given class, denoted in the
article by the symbol di. A Poisson call stream of class i is characterised by
the average call intensity λi, whereas Bernoulli and Pascal streams by the
call intensity γi from one unoccupied source (sources in active state, i.e.
in a state in which a subsequent call can be generated). The dependence
between the call intensity λi and the average value of the intensity of
offered Erlang traffic Ai can be expressed by the following formula:
λi
A = . (1)
i
µi
In a Bernoulli stream, call arrival intensity decreases along with the number
of serviced calls. The dependence between the number of traffic sources Si,
intensity γi on a source in active state and the average value of the
intensity of offered traffic Ai can be written as follows [25]:
A = (S
— A ) · γi . (2)
i i i
µi
In a Pascal stream, call arrival intensity increases along with the
number of serviced calls. The dependence between the parameters Si, γi
and the average value of offered traffic Ai can be written with the
following Formula [25]:
A = (S + A ) · γi . (3)
i i i
µi
3 Non-Poissonian Call Streams

Consider now those call classes that are derivatives of Poisson classes in
terms of the independence of call intensity on the occupancy state in the
system. In these classes, a call stream can have any distribution that defines
time tint between subsequent calls. The time tint distribution can be described
on the basis of two parameters: the expected value Ei(tint) and variance
Di(tint) of the time between subsequent calls. The expected value Ei(tint) is
the inverse of the average intensity λi, which makes it possible to write Eq.
(1) in the following form:
1
Ei(tint) = . (4)
Aiµi
Let us define now the call classes that are derivatives of the Bernoulli
class in terms of the dependence of the call stream on the occupancy state
of the group. In these streams, along with the increase in the number of
serviced calls, time tint between consecutive calls also increases. Further on
in the article, these classes will be called call classes with the state-
dependent arrival process (with descending traffic pattern). Let tact
determines the time between a termination of call service and the arrival of
a new call for a single traffic source, i.e. the time when the source is active.
The expected value Ei(tact) determines thus the average value of active time
of the source. The expected time Ei(tact) is the inverse of the average
intensity of the source in active state γi. The time tact can be described by
any (randomly selected) distribution whose parameters depend on the
expected value Ei(tact). As a result, the dependence between the expected
value Ei(tact) and the average traffic intensity of offered traffic Ai can be
expressed, on the basis of (2), as follows [25]:
Ei(t Si − A i
act
)= . (5)
Aiµi
It is also possible to define in a similar way classes that are state-
dependent with ascending traffic pattern that are derivative classes from
Pascal call classes. The dependence between the average traffic intensity and
the expected value Ei(tact) can be written in this particular case, on the basis
of (3), in the following way:
Si + A i
Ei(tact ) = . (6)
Aiµi
Algorithm
Let us consider now the way in which appropriate parameters for
selected distributions used in the investigation into the convolution
algorithm are determined. These distributions make a determination of the
time between calls (state-independent call classes) or the time between a
termination of a service and a subsequent call coming from the same source
(state-dependent call classes) possible. The distribution parameters can be
determined on the basis of the expected value Ei(t∗) and variance Di(t∗),
where t∗ = tint for state-independent streams and t∗ = tact for state-
dependent streams.
In the uniform distribution the probability density function is written
with the following formula:
fi(t) = 1
t i,max — ti,min , (7)
where ti,min denotes the minimum value after which a call of class i arrives,
whereas ti,max is the maximum value. The parameters ti,min and ti,max can be
determined on the basis of the expected value and variance:
√
ti,min = Ei(t∗) − 3 · Di(t∗), (8)
√
ti,max = Ei(t∗) + 3 · Di(t∗). (9)
In the case of the normal distribution, the probability density function is
determined directly on the basis of the expected value and variance for time
t∗:
1
2
(t−Ei(t∗))
−
fi(t) = √ e 2Di(t∗)
(10)
Di(t∗) 2π
In the gamma distribution the density function f (x) is typically written
in the following way:
tαi−1 e−t/βi , (11)
fi(t) =
βiαi Γ (αi)
where αi shape parameter (tail index), while βi is the scale parameter.
Those parameters (αi and βi) can be determined on the basis of the
expected value and variance:
2
αi = [Ei(t∗)] , (12)
D (t )
i ∗
βi = Di(t∗)/Ei(t∗). (13)
In the Pareto distribution the density function can be written as follows:
αi · βiαi
∀ fi(t)
t> βi , (14)
= tαi+1
where αi shape parameter (tail index), while βi is the scale parameter. Due
to a particular specificity of the distribution, the parameter αi should be
greater than 2. This value of the parameter αi limits variance to finite value.
The coefficients αi and xi can be determined on the basis of the following
formulas:
Di(t∗)
α = 1 + 1+ , (15)
i
[Ei(t∗)]2
i ∗
[ED(t(t )]2 )
βi =
E (t ) 1 + i i ∗∗ . (16)
1+ 1 + [EDi(ti(t∗)]2
∗)
4 Convolution Algorithm
In convolution algorithms the input data that describe the offered call class
i include: the number of demanded resources di and occupancy distribution V
[P ]{i}. The single element
V [P (n)]{i} of the distribution determines the
occupancy prob-
ability (of) n AUs by calls of class i in a system with the capacity V AUs
that services calls of class i only. Figure 2 shows a diagram of a full-
availability group for the convolution algorithm.
Fig. 2. A diagram of a full-availability group used for convolution modelling
In the next step of the convolution algorithm the distributions of single

classes [P ]{i} are aggregated to determine the occupancy [P ]M for a multi-
service
V
system V
that services the set M of call classes. The distribution [PV]M for the system to
which all call classes are offered makes a determination of the QoS
parameters, such as the blocking probability or loss probability, possible.
The convolution algorithm can be described by the three following steps:
1. Determination of the single occupancy distributions [PV]{i} for each offered

traffic class i,
2. Determination of the aggregated, normalised occupancy distribution [P V ] ,
M
3. Determination of the blocking probability Bi.
In the first step, for each class i, the distribution [P

V ]
{i}
of the occupancy
probability for a given number of AUs by calls of class i in a system with
the capacity
V AUs is determined. The distribution [PV]{i} can be determined on the basis
Algorithm
of the probabilities pi(x) that determine the number of calls x serviced in
the single-service system:
{i} pi(x) for x = n/di ∈

[P (n)]V = 0 (17)
in all other
N cases.
The probabilities pi(x) can be determined on the basis of

– single-service models (for BPP traffic),
– simulation experiment,
– measurements in a real system.
In the second step, with an application of the convolution operation, the non-
normalised occupancy distribution of the system [P ]M is determined.
[P ]M = [P ]{V1} ∗ [P ]V{2} ∗ . . . ∗ [PV ]{m}. (18)
The convolution operation [P ]A ∗ [P ]B can be written in the following way:

n
Σ
[P (n)] A∪B
= [P (l)]A · [P (n − l)]B, (19)
l=0
where A and B are the sets of single or previously aggregated call classes.
Then, the distribution [P ]M is normalised to the distribution
V
[P ]M .
[P (n)]VM = G · [P (n)]M , (20)
where G is the normalisation constant equal to:

V
Σ
G = 1/ [P (n)]M . (21)
n=0
In the third step, appropriate system characteristics, e.g. the blocking

probability Bi, can be determined:
Bi = Σ
V
[P (n)]VM . (22)
n=V −di+1
5 Occupancy Distributions for Non-Poissonian Call

Streams
For the purpose of this article, the occupancy distributions pi(x) for single,
non-Poissonian call streams have been determined in a simulation [26].
Appro- priate time distributions between events used in the simulators used
in the study were as follows: uniform distribution (7), normal distribution
(10), gamma distribution (11) and Pareto (distribution) (14). The
assumption was that in all
considered systems the service time had exponential character, while the
service intensity µi for each class i (i ∈ M ) was equal to 1 [(time unit)−1].
∀ µi = 1. (23)
i∈ M
Another assumption in the investigations discussed in this article was that

for each considered class∈i (i M ) with non-Poisson distribution the ratio of
square of expected value [Ei(t∗)]2 to variance Di(t∗) was equal to 3.
[Ei(t∗)]2
∀ = 3. (24)
i∈ M D i(t ∗ )
On the basis of (4)–(6) for required value of traffic intensity of a given class,
the expected value Ei(t∗) can be determined. Then, taking into
consideration the conditions (23) and (24) it is possible to determine all
necessary parameters for new call arrival time distributions. Figure 3 shows
the necessary parameters for the simulation of state-independent traffic
streams in relation to the average intensity of traffic offered to the system.
The graphs do not include the value tmin for the uniform distribution that,
consistent with the adopted assumptions, is always equal to 0. An
exemplary occupancy distribution in a system with the capacity 10 AUs and
average intensity of offered traffic equal to 7 Erl. is shown in Fig. 4. Each of
the curves presented in the diagram defines appropriate call streams in
which the time between calls is described by the uniform, normal, gamma
and Pareto distributions.
Fig. 3. The dependence of the para- Fig. 4. Occupancy distribution in a

meters of call stream distributions single-service system with the capac-
under study on the average intensity ity 10 AUs and average offered traffic
of offered traffic intensity 7 Erl.
Algorithm
Fig. 5. The dependence of the para- Fig. 6. The occupancy distribution in a

meters of the studied call stream dis- single-service system with the capacity
tributions on the average intensity of 10 AUs and average intensity of offered
offered traffic Ai traffic 7 Erl., and the number of
sources 20
Figures 5 and 7 show in turn the necessary parameters for a simulation

of state-dependent streams. The assumption is that the number of traffic
sources Si is equal to 20.
Fig. 7. The dependence of the para- Fig. 8. The occupancy distribution in a

meters of the studied call stream dis- single-service system with the capacity
tributions on the average intensity of 10 AUs and average intensity of offered
offered traffic Ai traffic 7 Erl., and the number of
sources 20
The exemplary occupancy distributions for the systems with the capacity
10 AUs to which state-dependent streams generated by 20 sources are
offered are shown in Figs. 6 and 8. The presented curves show the
occupancy distribution for a system which is offered traffic with the average
intensity of 7 Erl. Each of the curves presents appropriate call streams in
which the activity time of a source is described by a uniform, normal,
gamma and Pareto distributions.
6 A Comparison of the Results of the Simulation

with the Convolution Algorithm
In order to determine the boundary possibilities of a possible application of
convolution algorithms a comparison of the results obtained on the basis of
analytical algorithms with the results of corresponding simulation
experiments was made. The simulator applied for the purpose is described
in [27]. For all
Fig. 9. Full-availability group V = 20, d1 = 1, d2 = 2, d3 = 3, A1d1 : A2dd : A3dd = 1

: 1 : 1, state-independent call streams, exponential service time µ1 = µ2 = µ3 = 1
Algorithm 1
systems presented in this chapter the group capacity is equal to 20 AUs.
The group was offered 3 call classes that demanded d1 = 1, d2 = 2, d3 = 3
AUs, respectively. The assumption was that the proportions in the intensity
of offered traffic A1d1 : A2dd : A3dd were equal to 1 : 1 : 1. This means
that calls of all classes occupy on average the same number of resources in
the system with infinite capacity. The total traffic offered to the system was
equal to aV :
m
Σ
Aidi = aV, (25)
i=1
where a is the average traffic offered per one AU.

Figure 9 for a selected number of time distributions tint between
subsequent calls shows the dependence between the blocking probability Bi
for all call classes and offered traffic a that falls on a single AU. On the basis
of the presented results
Fig. 10. Full-availability group V = 20, d1 = 1, d2 = 2, d3 = 3, A1d1 : A2dd : A3dd = 1

: 1 : 1 descending state-dependent call streams, exponential service time µ1 = 1, µ2
= 1, µ3 = 1
it can be claimed that the convolution algorithm can be successfully applied

to models to which non-Poissonian call streams are offered. Slight variances
in the results obtained in the performed simulation experiment and the
results of analytical calculations follow from the fact that the distributions
under investigation have no sequence property. This means that the
probability of a new call arrival in such streams depends on the moment of
arrival of the preceding call and, in consequence, the convolution algorithm
proves to be an accurate method only for traffic classes with Poisson call
streams that have no sequence property.
Figure 10 shows the dependence between the blocking probability Bi and
offered traffic a per a single AU for streams with state-dependent call
arrival process (with descending traffic pattern). A comparison of Figs. 9 and
10 confirms that for call streams with state-dependent call arrival process in
which the call arrival intensity decreases with the number of serviced calls,
the accuracy of
Fig. 11. Full-availability group V = 20, d1 = 1, d2 = 2, d3 = 3, A1d1 : A2dd : A3dd =

1 : 1 : 1 ascending state-dependent call streams, exponential service time µ1 = 1,
µ2 = 1, µ3 = 1
Algorithm
the convolution algorithm increases. This results from the fact that the
sources of calls are mutually independent, whereas the influence of sequence
property appears “within the source” only. The total stream offered to the
system is generated by a large number of call sources. Each source has its
own memory and a rejection of a call from one source has no influence on
the call generation process in the remaining sources.
Similar conclusions can be drawn for the streams with state-dependent
call admission process in which the intensity of call inflow increases along
with the number of serviced calls. Figure 11 shows the dependence between
the blocking probability Bi and offered traffic a per a single AU.
The presented results confirm high accuracy of the convolution method
for systems with state-dependant call streams.
7 Conclusions
The results of the study presented in the article show that the convolution
algorithm can be successfully used to model systems with non-Poissonian
call streams, including state-dependent and state-independent call streams.
The approximation error is just slight and in extreme instances does not
exceed 20%. This article presents the results for groups with low capacities
(V = 20 AUs). The investigations carried out by the authors show that the
approximation error decreases along with an increase in the capacity of the
system. In their future studies the authors intend to determine the accuracy
of the convolution algorithm for a case where the group is offered a
mixture of call streams that differ in their time distributions between calls.
The authors also intend to study the accuracy of the convolution algorithm
for service times that are not exponential.
References
1. Iversen, V.: The exact evaluation of multi-service loss system with access
control. In: Seventh Nordic Teletraffic Seminar, Lund, pp. 56–61, August 1987
2. Ross, K.: Multiservice Loss Models for Broadband Telecommunication Network.
Springer, London (1995)
3. G-labowski,
֒ M., Kaliszan, A., Stasiak, M.: Asymmetric convolution algorithm
for blocking probability calculation in full-availability group with bandwidth
reservation. IET Circ. Devices Syst. 2(1), 87 (2008)
4. G-labowski,
֒ M., Kaliszan, A., Stasiak, M.: Modeling product-form state-dependent
systems with BPP traffic. Perform. Eval. 67, 174–197 (2010)
5. G-la֒bowski, M., Kaliszan, A., Stasiak, M.: Two-dimensional convolution
algorithm for modelling multiservice networks with overflow traffic. Math. Prob.
Eng. Article ID 852082, 18 (2013)
6. G-la֒bowski, M., Kaliszan, A., Stasiak, M.: Generalized convolution algorithm for
modelling state-dependent systems. IET Circ. Devices Syst. 8(5), 378–386 (2014)
7. Hanczewski, S., Kaliszan, A., Stasiak, M.: Convolution model of a queueing
system with the cFIFO service discipline. Mob. Inf. Syst. 2016, 1–15 (2016)
8. Moscholios, I., Logothetis, M., Vardakas, J., Boucouvalas, A.: Performance met-
rics of a multirate resource sharing teletraffic model with finite sources under the
threshold and bandwidth reservation policies. IET Netw. 4(3), 195–208 (2015)
9. Kaufman, J.: Blocking in a shared resource environment. IEEE Trans. Commun.
29(10), 1474–1481 (1981)
10. Roberts, J.: A service system with heterogeneous user requirements –
application to multi-service telecommunications systems. In: Pujolle, G. (ed.)
Proceedings of Performance of Data Communications Systems and their
Applications, pp. 423–
431. North Holland, Amsterdam (1981)
11. G-la֒bowski, M.: Recurrent method for blocking probability calculation in multi-
service switching networks with BPP traffic. In: Thomas, N., Juiz, C. (eds.)
EPEW 2008. LNCS, vol. 5261, pp. 152–167. Springer, Heidelberg (2008).
doi:10.1007/
978-3-540-87412-6 12
12. G-labowski,
֒ M., Stasiak, M., Weissenberg, J.: Properties of recurrent equations
for the full-availability group with BPP traffic. Math. Prob. Eng. (2012)
13. Delbrouck, L.: On the steady-state distribution in a service facility carrying mix-
tures of traffic with different peakedness factors and capacity requirements.
IEEE Trans. Commun. 31(11), 1209–1211 (1983)
14. Rácz, S., Gerö, B.P., Fodor, G.: Flow level performance analysis of a multi-
service system supporting elastic and adaptive services. Perform. Eval. 49(1–4),
451–469 (2002)
15. Vassilakis, V.G., Moscholios, I.D., Logothetis, M.D.: Call-level performance
modelling of elastic and adaptive service-classes with finite population. IEICE
Trans. Commun. E91.B(1), 151–163 (2008)
16. Moscholios, I., Vardakas, J., Logothetis, M., Boucouvalas, A.: Congestion proba-
bilities in a batched poisson multirate loss model supporting elastic and adaptive
traffic. Ann. Telecommun. - annales des télécommunications 68(5), 327–344(2013)
17. Moscholios, I.D., Logothetis, M.D., Vardakas, J.S., Boucouvalas, A.C.: Congestion
probabilities of elastic and adaptive calls in Elang-Engset multirate loss models
under the threshold and bandwidth reservation policies. Comput. Netw. 92, Part
1, 1–23 (2015)
18. Iversen, V.B.: Reversible Fair Scheduling: The Teletraffic Theory Revisited.
Springer, Berlin (2007)
19. Sobieraj, M., Stasiak, M., Weissenberg, J., Zwierzykowski, P.: Analytical model
of the single threshold mechanism with hysteresis for multi-service networks.
IEICE Trans. Commun. E95–B(1), 120–132 (2012)
20. G-labowski,
֒ M., Hanczewski, S., Stasiak, M., Weissenberg, J.: Modeling Erlang’s
ideal grading with multi-rate BPP traffic. Math. Prob. Eng. Article ID 456910,
35 (2012)
21. G-labowski,
֒ M., Kaliszan, A., Stasiak, M.: Modelling overflow systems with
distributed secondary resources. Comput. Netw. 108, 171–183 (2016)
22. Erlang, A.: Solution of some problems in the theory of probabilities of
significance in automatic telephone exchanges. Elektrotechnikeren (1917)
23. Engset, T.: On the calculation of switches in an automatic telephone system.
Telek- tronikk 99–142 (1998)
24. Wallstrom, B.: A distribution model for telefone traffic with varying call
intensity, including overflow traffic. Ericsson Technics (2) 183–202 (1964)
25. Iversen, V.: Teletraffic Engineering Handbook. ITU-D SG 2/16 and ITC Draft
(2001)
Algorithm
26. G-la֒bowski, M., Kaliszan, A.: Simulator of full-availability group with
bandwidth reservation and multi-rate bernoulli-poisson-pascal traffic streams. In:
EUROCON 2007 - The International Conference on “Computer as a Tool”, pp.
2271–2277. IEEE (2007)
27. Hanczewski, S., Kaliszan, A.: Simulation studies of queueing systems. In: 2016
10th International Symposium on Communication Systems, Networks and
Digital Signal Processing (CSNDSP), pp. 1–6, IEEE, July 2016
An Efficient Method for Calculation
of the Radiation from Copper Installations
with Wideband Transmission Systems
✉
Piotr Zawadzki( )
Institute of Electronics, Silesian University of

Technology, Akademicka 16, 44-100 Gliwice, Poland
[email protected]
Abstract. The recent innovations in connecting hardware and smart

encoding schemes make it possible the old copper installations are
capable to support T1/E1 bit rates. It is obvious that old networks
destined for POTS, will intensively dissipate energy transmitted by
new wideband services and they should be regarded as a potential
source of electromagnetic hazards. This note reports results of the
study towards developing a computer-aided method for numerical
modeling of an electromagnetic environment in a close vicinity of
cables carrying signals of wideband transmission systems. The
proposed method is formulated in the frequency domain and uses
multiconductor transmission line theory for determination of field
sources. Subsequently, fullwave approach is applied to find near
electromagnetic field in the vicinity of the cable.
Keywords: Wideband transmission system · Electromagnetic hazards
1 Introduction
Bandwidth “hungry” computing applications, the increasing need for
information sharing and the explosive growth in higher data transmission
rates demand the greater speed and larger bandwidth from LANs (Local
Area Networks) or local premises telephone installations. One method of
improving the information infrastructure is to install an all-fibre network.
However, the all-fibre premises distribution network has its price:
implementing it requires the replacement of copper cables with fibre and the
installation of more expensive active components, such as fibre optic
interface devices. With recent innovations in the design of connecting
hardware and smart encoding schemes UTP (Unshielded Twisted Pair) or
FTP (Foiled Twisted Pair) cables are now capable of supporting high bit
rates.
DSL (Digital Subscriber Line) refers to several types of advanced
modems that enable fast access at speeds 300 times faster than most analog
modems. Since DSL works on regular telephone lines they are considered a
key means of opening the bottleneck in the “last mile” of the existing
telephone infrastructure, as telephone companies seek cost-effective ways
of providing much higher
Algorithm
DOI: 10.1007/978-3-319-59767-6 19
An Efficient Method for Calculation of the Radiation 237
speed to their customers. It is obvious that old premise networks, destined

for the POTS, will intensively dissipate energy for the new modulation
schemes and they might become a source of not negligible electromagnetic
radiation. The above observation leads to the conclusion that new services
installed on existing networks should be treated not only as victims of the
external noise, but also as potential sources of electromagnetic emissions.
This, in turn, calls for the elaboration of the fast, reliable, computer aided
method for prognosis of electromagnetic hazards arising in this new
environment.
The proposed method is formulated in the frequency domain and uses
multiconductor transmission line theory for determination of field sources.
Subse- quently, fullwave approach is applied to find near electromagnetic
field in the vicinity of the cable.
2 Numerical Model
Numerical models for the radiation of the cable with the given parameters
can be derived on a basis of computational intensive methods based on a
direct Maxwell equations discretization. However everyday EMC engineering
practice requires maybe less accurate but more efficient methods.
We use multiconductor transmission line theory to determine current
distribution along the cable [1, 2] and full-wave approach [3] to determine
field of a specified current distribution.
2.1 Calculation of the Current Distribution

For a system of N conductors (plus reference conductor) telegraphers
equation can be written in the matrix form
dV(s)
+ Z′I(s) = 0 (1a)
ds
dI(s)
+ Y′V(s) = 0 (1b)
ds
where V(s) and I(s) are the voltage and current vectors and Z′ and Y′
are the square N × N matrices describing per unit length line parameters.
These two first order differential equations can be transformed into two
second order
equation for the current distribution
2
d I(s)
− R′I(s) = 0
(2)
ds2
where R′ = Y′Z′. The solution of this equation has the
form
I(s) = Zc−1S−1 E+(s)a − E−(s)b (3)

238 P. Zawadzki
where 1
Zc = S−1 SZ′ (4)
γi
⎡ ⎤
γ1 0 . . . 0
[γi] = 0 γ2 . . . 0
⎢. . . . . . . . . . (5)
.. ⎣
⎡ ⎥ 0 0 . . . γn ⎤
⎦
e ∓γ 1 s 0 ... 0
E±(s) = ⎢ 0 e ∓γ 2 s . . . 0 ⎥ (6)
. .⎣. . . . . . . . . . . . . . . . . . . . . .⎦
0 0 . . . e∓γns
and matrix S is a diagonalisation matrix of R′, γi is a propagation constant
of the i-th mode and vectors a and b have to be calculated from the
boundary conditions.
2.2 Decomposition on Elementary Dipoles

Due to linearity of Maxwell equations linear extended source (i.e. cable) can
be decomposed into collection of short sections and each section can be
treated as elementary dipole. If the cable section length l holds the
condition [3] l < λ/10 (λ is the wavelength) then one can neglect the spatial
dependency of the current on a dipole. We have in this case the staircase
approximation of the current distribution along the cable, and in the net
effect the entire structure is replaced by a set of elementary dipoles.
The resulting field in point P can be determined as a sum of respectively
retarded contributions. The rigorous approach to field calculation in the
presence of a lossy ground requires computation of the so called Sommerfeld
integrals [4]. The cost of computation of those integrals is prohibitively large
and unacceptable in most practical implementations. Fortunately, the finite
ground conductivity has in many cases only a small impact on obtained
results and it is acceptable to use approximated method based on perfect
soil assumption and method of electrical images.
2.3 Field of an Elementary Dipole

The elementary dipole is characterized by its directional vector 1l (see Fig.
1), length l that is small compared to all distances of interest, and current
I0. The solution of Maxwell equations in free space for this source
configuration has the form [5, 6]
ηI0l
E = −j (kR)2 − (1 + jkR) 1
4πk l
, e−jkR
— (kR) − 3 (1 + jkR) (1 ·1 ) (7a)
2 l R R
R3
1
I 0l
H = (1
×1 ) (1 + jkR) e−jkR
l R . (7b)
4π R2
z P
l
rp
Fig. 1. Elementary dipole in a free space
3 Method Validation
Method validation can be performed by comparing its results with
predictions of other methods that implement commonly accepted and
verified models or by the comparison with independently conducted
computations based on the same mathematical model of the reality. In the
following sections we present method validation as a two stage process:
1. comparison with independent multiconductor transmission line analysis

code developed by Excem [8]
2. comparison with MOMIC code implementing rigorous full-wave approach
to scattering and radiation problems [7]
The mentioned above programs were used for verification of current and
field distributions for some canonical examples. All field calculations are
performed with perfect ground assumption. Keys for presented plots have
the following meaning: line code denotes results obtained with the validated
program, Excem code – results from [8], MOM code – results obtained by a
rigorous approach to electromagnetic waves radiation based on solution of
an integral equation for current distribution with method of moments [5, 7].
3.1 Comparision with Excem Code

The distribution of sources uniquely determines the electromagnetic field.
The effective, nonbalanced current flowing along the multiconductor cable is
the field
240 P. Zawadzki
source in our case. The report [8] was chosen as a source of reliable data to
validate current distribution along the cable. We considered four pair UTP
cable with each pair loaded with R0 = 120 Ω and one pair forced with
voltage V0 = 2 V. The cable has length L = 10 m and was positioned on
height h = 10 cm. The current distribution along cable for two frequencies
is presented in Fig. 2.
0.0003
60
Excem code, f=2.818kHz
0.00025
|I(f=2.818kHz)| [µA]
line code, f=2.818kHz 50
|I(f=2.818MHz)| [µA]
Excem code, f=2.818MHz
0.0002 line code, f=2.818MHz 40
0.00015
30
0.0001
20
5e-05
10
0 0
0 2 4 6 8 10
y [m]
Fig. 2. Effective current distribution along multiconductor line.
3.2 Comparision with Method of Moments

Lets consider transmission line composed by a horizontal conductor over
perfectly conducting ground (see Fig. 3) and grounded at both ends by
loading impedances. This structure (together with its electrical image) was
analysed by a rigorous approach and with proposed method. The observed
quantities were: current distribution along the line for two fixed frequencies
and electric field produced at point P (L, L/2, 2h).
V
h
ZL1 ZL2
L
y
Fig. 3. Transmission line geometry, L = 2 m, h = 1 cm, a = 1 mm, ZL1 = ZL2 = 50 Ω.
It is clearly seen from Fig. 4 that antenna mode gives no contribution to

the current on the line, and in consequence, radiated fields predicted by the
transmission line theory and rigorous approach are the same (see Fig. 5).
1
0.01
MOM code, f=20 MHz
line code, f=20 MHz
MOM code, f=200 MHz
0.008 line code, f=200 MHz
0.006
|I/V0| [Ω ]
-1
0.004
0.002
0
0 0.5 1 1.5 2
y [m]
Fig. 4. Current distribution along horizontal conductor

90
MOM code
85 line code
80
75
70
|E| [dBµ V/m]
65
60
55
50
45
40
35
100000 1e+06 1e+07 1e+08
f [Hz]
Fig. 5. Field at point P (L, L/2, 2h).
4 Parametric Study
A motivation in developing a model is to have the capability to perform
parametric studies, in which independent model parameters can be varied
in order to gain a global understanding of the phenomenon. We used
presented method to investigate the dependence of the field in the vicinity
of the multiconductor transmission line on the following parameters:
– cable type,
– geometrical configuration of the line,
– line forcing.
4.1 Calculation Setup

The considered line is composed from sections of different length of UTP
four pair cables S278 and S298. Each pair is terminated with the resistance
120 Ω. Two cases of line forcing were considered: first pair forced by a
sinusoidal voltage of amplitude 2 V, or all pairs forced with the voltage
same as previously. This forcing conditions are identified on graphs by keys
V1 and V4, respectively. Geometrical configurations further referenced as I-
like and U-like are presented in Fig. 6. Configurations are parametrized by
the cable section length L and the height h above ground, which is assumed
to be ideal. Observation point was positioned at point P(1 m, L/2, 1 m).
The calculations were performed for two heights and three lengths (Table
1).
242 P. Zawadzki
z
h
L
y
a) “I-like” x
b) “U-like”
Fig. 6. Analysed geometries
Table 1. Parameters of calculations
Cable Height Geometry Length Forcing

S278 h = 15 cm I-like L = 5 m V1
S298 h = 1 m U-like L = 10 m V4
L = 15 m
4.2 Discussion
The strength of the field at the resonant frequency strongly depends on
cable type but resonant frequency itself does not (see Fig. 7).
130
E/S278/h=1m/I/L=10m/V1
120 E/S298/h=1m/I/L=10m/V1
110
|E| [dBµ V/m]
100
90
80
70
60
50
40
100000 1e+06 1e+07 1e+08
f [Hz]
Fig. 7. Amplitudes at resonant frequencies.
The resonant frequencies are determined by geometrical configuration

(see Fig. 8) of the system, and it is not possible to design modulation
schemes to program them out i.e. make PSD function small at resonant
frequencies. However, it is still possible to design new modulation schemes to
avoid resonant frequencies of potential victims, for instance radio amateur
receivers.
130
E/S298/h=1m/I/L=10m/V4
120 E/S298/h=1m/U/L=10m/V1
110
|E| [dBµ V/m]
100
90
80
70
60
50
40
100000 1e+06 1e+07 1e+08
f [Hz]
Fig. 8. Dependency of resonant frequencies on geometrical configuration.
Moreover, field amplitude at resonant frequency do not depend on

geometrical cable configuration. The results for two geometries with the
same total length Ltot = 15 m are presented in Fig. 9. This behavior can be
explained by a special choice of the observation point. It is positioned close
to the cable, and far ends of the cable do not give significant contributions
to the field because those contributions vanish as 1/R.
130
E/S298/h=1m/I/L=15m/V4
120 E/S298/h=1m/U/L=5m/V4
|E| [dBµ V/m]
110
100
90
80
70
100000 1e+06 1e+07 1e+08
f [Hz]
Fig. 9. Field dependency on geometrical configuration.
5 Conclusion
The proposed method is general and flexible. Virtually, it is capable to
analyse any service supported over any cable, unless reliable cable model and
PSD (Power Spectral Density) function of the transmitted signal are known.
The method is also fast as the computation time grows with square of the
number of pairs in a cable and linearly with the cable length. Moreover,
method is able to analyse not only straight section of the cable, but also
handles networks with cable bends.
244 P. Zawadzki
Acknowledgements. Author acknowledges support by the Ministry of Science and

Higher Education funding for statutory activities and Rector of Silesian University
of Technology grant number 02/030/RGJ17/0025 in the area of research and
development.
References
1. Paul, C.R.: Analysis of Multiconductor Transmission Lines. Wiley, New York (1994)
2. Tesche, F.M., Ianoz, M.V., Karlsson, T.: EMC Analysis Methods and Computa-
tional Models. Wiley, New York (1997)
3. Stutzman, W.L., Thiele, G.A.: Antenna Theory and Design. Wiley, New York (1981)
4. Somerfeld, A.: Uber die ausbreitung der wellen in der drahtlosen telegraphie.
Annalen der Physics 28, 665 (1909)
5. Zawadzki, P.: Elaboration of software for prediction of signals induced by a
lightning discharge on telecommunication cables placed inside buildings.
Technical report NT/LAB/QFE/143, CNET, Lannion, December 1996
6. Zawadzki, P.: Evaluation of the near field radiated by cables used for wideband
transmission systems. Technical report NT/CNET/6338, CNET, Lannion,
Septem- ber 1999
7. Harrington, R.F.: Field Computation by Moment Methods. MacMillan, New
York (1968)
8. Comaparaison de l’emission des cablages de telecommunication. Excem doc. no.
98042102A
A Videoconferencing System Based on WebRTC
Technology
Robert Bestak(✉) and Jiri Hlavacek
Czech Technical University in Prague, Technicka 2, Prague 6, Czech Republic

Abstract. Last years, videoconferencing systems are rapidly evolving

and they are becoming more and more popular as a real time commu-
nication tool among users. A technology such as VoIP can be used in
conjunction with desktop videoconferencing systems to enable low-cost
face-to-face business meetings without a necessity to travel, especially
for internationally oriented companies that have offices all around the
word. In this paper, we focus on and discuss a videoconferencing
platform based on the WebRTC technology. We analyze an impact on
the multiplexing server’s CPU load and memory requirements for
different number of communicating users while taken into account
different HW/SW configurations of end-point devices.
Keywords: Real time communication · Videoconferencing systems ·

WebRTC · Measurement
1 Introduction
Historically, a real time communication, has only been available to large
companies who could afforded to pay expensive licensing fees, and/or to
buy specific proprietary depend HW. Since that, videoconferencing systems
evolved to a low cost standard-based technology that is available to the
general public. Besides the audio and video communications, current
videoconferencing systems also support variety of other services, including
instant messaging, documents sharing, screen sharing, video recording, etc.
Technological developments in the 2010 s have further extended the
capabilities of videoconferencing systems for use not only for in the static
environment but in the mobile environment as well.
A videoconferencing technology typically employs both audio and video
chan- nels jointly interconnecting two (point-to-point) or several users
(multi-point) located at different sites. The multi-point communication is
possible via a unit called Multipoint Control Unit (MCU) that represents a
bridge interconnecting the communicating users (similarly to the audio
conference call). Either all participating users can call the MCU, or the
MCU can call the involved users itself. A MCU can be characterized
according to the number of simultaneous supporting calls, its ability to
provide transposing of data rates and protocols, or the way it is
implemented (stand-alone HW devices, vs. embedded unit in a dedicated
videoconferencing system). Certain videoconferencing platforms employ
246 P. Zawadzki
DOI: 10.1007/978-3-319-59767-6 20
246 R. Bestak and J. Hlavacek
a standards-based H.323 technique known as “decentralized multi-point”

where each involved end-point device in a multi-point communication
exchanges video and audio directly with other end-point devices without
using a MCU.
Videoconferencing systems can be operated in two modes: (a) Voice-
Activated Switch (VAS), or (b) Continuous Presence. In the VAS mode, the
MCU switches end-point devices in such way that it can be seen by the
other end-point devices by the levels of one’s voice. For an example, if there
are four users in a video-conference, the only one that will be displayed is
the currently talking user. In the Continuous Presence mode, all users are
displayed at the same time; the user streams are assembled together by
MCU into a single video stream that is transmitted to all users.
Though, the videoconferencing systems gain increasing popularity, there
are still several issues the systems have to face, among them: complexity
(not all users are technically skill, users prefer a simple interface and usage),
interoper- ability (not all systems can readily be interconnected - there are
different standards, features and qualities), or bandwidth (it is not always
possible to have a high quality connection to support a good quality
conferencing communication).
In general, the videoconferencing systems can be classified into the
following three categories:
– Dedicated systems: these systems have all required components packed

into a single platform (console), which contains all necessary interfaces,
codecs, and control unit.
– Desktop systems: they are designed as add-ons (HW boards or SW
codecs) to common PCs and laptops, and they transform them into
videoconferencing platforms. A range of different cameras and
microphones, which contains the necessary codecs and interfaces, can be
used.
– WebRTC platforms: they represent videoconferencing solutions that are
available through standard web browsers. Typically, a browser uses the
local camera and/or microphone to establish the connection.
Current approaches supporting real time communications are based

either on utilizing a separate application or on employing a specific plug-in
(e.g., Flash plug-in). By using a separate application means leaving the web
browser and launching a dedicated application, i.e. the browser content and
the real-time content are independent. The plug-in solution provides a
tighter integration between the real-time and browser contents. However,
plug-ins are proprietary solutions and they do not work in all environments.
Contrary, the WebRTC platforms include all necessary audio and video
components to support the real-time communications directly via web
browsers.
The rest of paper is organized as follows. The next section briefly
describes basic features of the WebRTC technology, including related
works. A comparison of selected real-time communication platforms with
WebRTC is provided in Sect. 3. Impact on the multiplexing server’s CPU
load and memory requirements for different number of communicating end-
point devices are analyzed in Sect. 4. Finally, Sect. 5 concludes the paper.
A Videoconferencing System Based on WebRTC 247
Technology
2 Web Real-Time Communication
2.1 Basic Features
Web Real-Time Communication, or WebRTC, is a specification plug-in free,
real time communication via a web browser. It provides communication
protocols and defines application programming interfaces allowing two or
more users to mutually communicate in real-time. The WebRTC technology
is being standardized by World Wide Web Consortium (W3C) organization at
the application level and by Internet Engineering Task Force (IETF)
organization at the protocol level.
WebRTC is composed of three main elements:
– Web browser: To send and receive audio and video streams directly from a
web browser, the web browser has to be enhanced with capabilities for
controlling the local audio and video elements at the device at which the
web browser is running (e.g., a PC or smartphone).
– Web application: A user is typically asked to download a Java script
from a web server. This script runs locally at the user’s device and
interacts with the web server.
– Web server: The server provides the Java scripts for users and executes
the necessary application logic.
Initially, the WebRTC technology was designed to provide a real time
communication between two users. Later on, it was enhanced with the
multi-connection support. There are 2 approaches how to support the multi-
connections: (a) centralized mode, and (b) fully meshed mode. In the first
case, a user agent is in charge of dialog and mixing of media streams. The
assumption here is that only the host mixes streams and sends them to each
end-point device; the host could be a dedicated server or a standalone device
with sufficient computational power. To join the videoconference call, a new
end-point device has to notify the host. In the centralized mode, the
communication fully depends on the host; if a host leaves the conference, all
connections are disconnected. In the second mode, fully meshed model, end-
point devices are interconnected to each other and any end- point device
can invite and leave the call without effecting the other end-points. The
WebRTC technology can be nowadays integrated within a SIP-based
system and thus offering additionally functions [1, 2]. A videoconferencing
architecture can include a unit which allows legacy SIP user agents and SIP
WebRTC clients to interoperate. A real-time chatting tool based on HTML5
and WebRTC can also be proposed; it includes a basic information
management module and a communication module allowing text
communication by means of a dedicated server and voice/video sessions
through a point-to-point connection between web
browsers.
2.2 Related Works

The WebRTC technology has received a lot of attention by researchers
couple of last years.
In [3], a study of WebSockets and WebRTC usage from the point of

mobile device power consumption in LTE networks is provided. Based on
realized power consumption analysis, a couple of recommendations for
standardization process are provided by authors.
Performance of WebRTC’s adaptive video streaming capabilities in an
IEEE 802.16e network environment is evaluated in [4].
The applicability of HTML5 and WebRTC technology for a P2P video
streaming is investigated in [5]. The work describes a Bit Torrent P2P video
streaming solution which is used to identify the performance bottlenecks in
WebRTC-based P2P video streaming implementations.
In [6], the WebRTC-based communication performance is evaluated for
case of star and full-mesh network topologies, with focus on the performance
of the congestion control mechanism used with WebRTC.
A feasibility of live video streaming protocols via the WebRTC
technology is investigated in [7]. The experiments illustrate a possibility to
implement a pull-based P2P streaming protocol with WebRTC, at least, for
small-scale P2P networks.
A performance of WebRTC on mobile devices, while taking into account
different type of mobile devices, wireless network connectivity and web
browser configurations, is evaluated in [8].
In [9], authors present a benchmarking tool, call WebRTCBench, which
mea- sures WebRTC peer connection establishment and communication
performance. Authors discuss performance of the WebRTC technology across
a range of implementations and devices.
WebRTC communications over the LTE mobile network simulated by
NS-3 tool is investigated in [10]. Several multimedia WebRTC streams
between the two end-point devices are analyzed and empirical CDFs of
typical performance figures including throughput, jitter, and packet loss are
derived under different LTE scenarios.
2.3 Security and Identification

As to the WebRTC security aspects, the WebRTC stack includes Datagram
Transport Layer protocol that is designed to prevent eavesdropping and
information tampering and to support real-time data encryption and
association management service to Secure Real-time Transport Protocol.
An issue with a traditional desktop SW is whether a user can trust
the application itself [11]. The installation of a new SW or a plug-in can
potentially scrumptiously install a malware or other undesirable SW.
Typically, users have no idea where the SW was created or from whom they
are downloading the SW. Thus, a malicious third party has possibility in
repackaging perfectly safe and trusted SW to include malware, and offering
this package on free SW websites. Advantageously, WebRTC is not a
plug-in, nor is there any installation process for any of its components.
The WebRTC platform is simply installed as part of WebRTC compatible
browser (e.g., Chrome or Firefox). Therefore, there is no risk of installing
malware or virus through the use of an appropriate
Technology
WebRTC application. Anyway, WebRTC applications should still be
accessed via a HTTPS website signed by a valid authenticator.
Moreover, the WebRTC technology is done in such way that a user is
explic- itly asked a permission for the camera or microphone to be used. As
the permission can be done on one-time or permanent access base, a
WebRTC application cannot arbitrarily gain access or operate any device.
Furthermore, if the microphone or camera is being used, the user interface
is required to expressly show the user that the microphone or camera are
being operated.
It is desirable that a user is able to verify the identity of their
communicating partners, i.e. users naturally wants to be certain that they
are communicating with users they believe that they are speaking to, and
not to an imposter. The authentication of peers has to be performed
independently from the signaling server as the signaling server itself cannot
be trusted. To do this, web-based identity providers can be utilized, such as
Facebook Connect, BrowserID, OAuth, etc. The role of these providers is to
verify the user identity to other users, based on the authority of identity
provider itself. This allows users to tie their authentication on other services
to their main account on a trusted service. The implementations of each
identity provider may differ due to independent development by different
companies rather than being based on an open-source standard, but the
underlying principle and functions remain the same.
3 Comparison of Videoconferencing Platforms

In this section, we provide a brief comparison of selected real-time
communication platforms and we highlight differences between them and the
WebRTC technology.
3.1 Skype
Comparing to WebRTC, Skype is an application. Skype is available for
different platforms and provides video and audio services between two or
more users. Additionally, users may exchange digital documents (images,
text, video and any others), and text and video messages.
3.2 Hangouts
Hangouts is a communication platform, developed by Google, supporting
instant messaging, video, audio, and SMS between two or more users. The
service can be accessed online (Gmail or Google+ websites) using either PCs
or mobile phones. Comparing to the WebRTC technology, Hangout uses a
proprietary protocol, although with WebRTC elements.
3.3 WebEx
WebEx Meeting Center is a part of commercial solution developed by Cisco.
It is a web conferencing platform delivered through the Cisco Collaboration
Cloud.
The stream processing is done in the cloud and the multiplexing of media
streams is not at the charge of end-point device. It uses a proprietary
algorithm named Universal Communication Format that deals with a large
range of media.
3.4 Jitsi
Comparing to previously mentioned solutions, Jitsi is a free and open source
multi-platform supporting video, audio, and instant messaging. It is based
on the OSGi framework (Collection of Java Library). Jitsi is not a web-
oriented, a user is asked to download and install specific SW to run it. In
terms of functions, libraries are not as rich as the WebRTC ones but the
main abilities are effective. Similarly to WebEx, the media stream
processing could be done at a server instead of at the end-point device.
The Jitsi videobridge, a videoconferencing solution supporting the
WebRTC. It’s a Selective Forwarding Unit, a centralized server, which is
forwarding selected streams to other participating users in the
videoconference call.
3.5 AnyMeeting
AnyMeeting is a web conferencing and webinar services allowing users to
host and to attend web based conferences and meetings and to share their
desktop screens with other users via web browser.
The above mentioned platforms are summarized and compared in Table
1. The comparison takes into account aspects such as the place of stream
multiplexing, set of functions (text chat, screen sharing, etc.), web-browser
compatibility, and multi-connection support.
Table 1. Comparison of videoconferencing platforms.
Skype Hangouts WebEx Jitsi Jitsi Any WebRTC

video Meeting
bridge
Multiplexing Client Server Server Server Client Server Client
Rich. functionalities Middle No Yes Yes No Midle Yes
Requirements App Plugin App App None App None
Multi-call setup Limited Yes Yes Yes Yes Limited Yes
4 Measurements and Results

In this section, we investigate impact on the multiplexing server’s CPU load
and memory requirements for different numbers of used end-point devices;
distributed processing issues are discussed in more details for example in
[12, 13]. We set up a testbed that considers 3 PCs with various HW and
SW configurations as indicated in Table 2. The configuration of the server is
indicated in the table
Technology 1
as well. Each participant receives a return video with multiplexed streams
of each participant in one video stream, i.e. the end-point device has to
decode one flow instead of one flow per end-point device. The multiplexing
is provided by Kurento media server [14, 15].
Table 2. Parameters of PCs.
PC1 I5 2.5 GHz dual-core, Nvidia 610 M 1 GB, 4 GB RAM, Ubuntu 14.04
LTS (NO 720P)
PC2 I3 1.7 GHz dual-core, Nvidia 820 M 2 GB, 4 GB RAM, Win 10 (NO
120P)
PC3 I7 2.5 GHz dual-core, Nvidia GTX970 3 GB, 8 GB RAM, Win 7
Server Intel(R) Xeon(R) CPU 5160 @ 3 GHz, 8 GB RAM, Ubuntu 16.04.1 LTS
In total, there have been tested 28 scenarios, which differs in the number
of involved end-point devices, PCs (1, 2, and 3), and display resolution
configuration (160 × 120, 320 × 240, 640 × 480, and 1280 × 720), see
Table 3. Notice that the configuration id is constructed in such way that
the first digit indicates PC
(using Table 2) and the second digit is associated to the display resolution;
for example the configuration 1222, refers to two interconnected end-point
devices
(PC1 and PC2) where the display resolution is set to 640 × 480. The
resolution of the return video, for mono user configuration is used 800 × 600
and for the other cases 200 × 150 video per user is sent back to the end-
point device. The PC1 and PC2 use all the available bandwidth, while the
data rate of PC3 is
limited to 600 kB/s.
All measurements are done by using Firefox browser, and the values are
given as an arithmetic average of ten consequent measurements (which is a
common way to evaluate a CPU load).
Figure 1 shows the server’s CPU load for different testing scenarios
(please notice that the server’s CPU has 4vCPU, therefore the load can
theoretically reach up to 4 × 100%). As can be observed, the server’s CPU
load is the lowest for the PC3, where the bit rate is the lowest. As expected,
the bit rate and there-
fore the video quality has a considerable impact on the server’s CPU load.
For increasing number of involved end-point devices, the server’s CPU load
increase as well (scenarios 102030–132333). From the point of display
resolution, the server’s CPU load strongly depends on the resolution and bit
rate, often given by the HW/SW configuration of involved end-point
devices. Figure 1 indicates that about one CPU per end-point device is
needed.
The impact on server’s RAM is illustrated in Fig. 2. As we can expect,
the requirements on the RAM are directly proportional to the increasing
number of communicating end-point devices, more or less no matter what is
the HW/SW configuration of these end-point devices. Whereas, the display
resolution has a negligible impact on the memory requirement.
Scenarios with one PC are special cases that illustrate the memory and
CPU consumption when no multiplexing is performed, i.e. the scenarios
represent the load needed for decoding and encoding of the video flow.
Table 3. Testing configurations.
Configuration PC Sent resolution Received Bit rate [kB/s]

resolution by
end-point
10 PC1 160 × 120 800 × 600 373
11 PC1 320 × 240 800 × 600 658
12 PC1 640 × 480 800 × 600 1470
13 PC1 1280 × 720 800 × 600 1470
20 PC2 160 × 120 800 × 600 514
21 PC2 320 × 240 800 × 600 514
22 PC2 640 × 480 800 × 600 1430
23 PC2 1280 × 720 800 × 600 2150
30 PC3 160 × 120 800 × 600 486
31 PC3 320 × 240 800 × 600 601
32 PC3 640 × 480 800 × 600 608
33 PC3 1280 × 720 800 × 600 590
1020 PC1+PC2 160 × 120 200 × 150 832
1121 PC1+PC2 320 × 240 200 × 150 1040
1222 PC1+PC2 640 × 480 200 × 150 2490
1323 PC1+PC2 1280 × 720 200 × 150 2050
1030 PC1+PC3 160 × 120 200 × 150 750
1131 PC1+PC3 320 × 240 200 × 150 1110
1232 PC1+PC3 640 × 480 200 × 150 1250
1333 PC1+PC3 1280 × 720 200 × 150 1220
2030 PC2+PC3 160 × 120 200 × 150 900
2131 PC2+PC3 320 × 240 200 × 150 1060
2232 PC2+PC3 640 × 480 200 × 150 1130
2333 PC2+PC3 1280 × 720 200 × 150 2620
102030 PC1+PC2+PC3 160 × 120 200 × 150 1220
112131 PC1+PC2+PC3 320 × 240 200 × 150 1870
122232 PC1+PC2+PC3 640 × 480 200 × 150 1900
132333 PC1+PC2+PC3 1280 × 720 200 × 150 2670
Technology
Fig. 1. Server’s CPU load.
Fig. 2. Requirements on the size of server’s RAM.

5 Conclusion
In this paper, we have discussed basic characteristic of the WebRTC
technology, and compare the technology with some other exiting real-time
videoconferencing systems. Additionally, we have evaluated impact on the
multiplexing server’s CPU load and requirements on RAM for different
numbers of end-point devices running WebRTC sessions. In total, we have
tested 28 different configurations. The obtained measurements illustrate a
strong relation between the video resolution and bit rate of the involved
end-point devices and the server’s CPU load. The requirements on server’s
RAM are than directly proportional to the number of involved end-point
devices, no matter what is the HW/SW configurations and considered
display resolutions. Obtained results illustrate that the server has to be
dimensioned in such way that about one CPU has to be considered per one
end-point device.
The current scenarios only include PCs. In our next work, we plan to
take into account a more heterogonous environment and to consider as end-
points mobile devices as well.
Acknowledgments. This research work was supported by the Grant Agency of CES-
NET grant no. 542/2014.
References
1. Segec, P., Paluch, P., Papan, J., Kubina, M.: The integration of WebRTC and
SIP: way of enhancing real-time, interactive multimedia communication. In:
12th IEEE International Conference on Emerging eLearning Technologies and
Applications (ICETA), pp. 437–442 (2014)
2. Amirante, A., Castaldi, T., Miniero, L., Romano, S.P.: On the seamless
interaction between WebRTC browsers and SIP-based conferencing systems.
IEEE Commun. Mag. 51(4), 42–47 (2013)
3. Mandyam, G., Ehsan, N.: HTML5 connectivity methods and mobile power con-
sumption. Technical report, Qualcomm (2012)
4. Fund, F., Wang, C., Liu, Y., Korakis, T., Zink, M., Panwar, S: Performance of
dash and WebRTC video services for mobile users. In: 20th International Packet
Video Workshop (PV), p. 18. (2013)
5. Nurminen, J., Meyn, A., Jalonen, E., Raivio, Y., Marrero, R.: P2P media
streaming with HTML5 and WebRTC. In: IEEE Conference on Computer
Communications Workshops, pp. 63–64 (2013)
6. Singh, V., Lozano, A., Ott, J.: Performance analysis of receive-side real-time
congestion control for WebRTC. In: 20th International Packet Video Workshop
(PV),
p. 18 (2013)
7. Rhinow, F., Veloso, P., Puyelo, C., Barrett, S., Nuallain, E.: P2P live video
streaming in WebRTC. In: World Congress on Computer Applications and
Information Systems, p. 15 (2014)
8. Heikkinen, A., Koskela, T., Ylianttila, M.: Performance evaluation of distributed
data delivery on mobile devices using WebRTC. In: IEEE International Wireless
Communications and Mobile Computing Conference (2015)
Technology
9. Taheri, S., Beni, L.A., Veidenbaum, A.V., Nicolau, A., Cammarota, R., Qiu, J.,
Lu, Q., Haghighat, M.R.: WebRTCBench: a benchmark for performance
assessment of WebRTC implementations. In: 13th IEEE Symposium on
Embedded Systems For Real-time Multimedia (2015)
10. Carullo, G., Tambasco, M., Di Mauro, M., Longo, M.: A performance evaluation
of WebRTC over LTE. In: 12th Annual Conference on Wireless On-Demand
Network Systems and Services (2016)
11. Pa-lka, D., Zachara, M., Wójcik, K.: Evolutionary scanner of web application
vulnerabilities. In: Gaj, P., Kwiecien´, A., Stera, P. (eds.) CN 2016. CCIS,
vol. 608,
pp. 384–396. Springer, Cham (2016). doi:10.1007/978-3-319-39207-3 33
12. Piórkowski, A., Szemla, P.: Client-side processing environment based on com-
ponent platforms and web browsers. In: Kwiecien´, A., Gaj, P., Stera, P. (eds.)
CN 2013. CCIS, vol. 370, pp. 21–30. Springer, Heidelberg (2013). doi:10.1007/
978-3-642-38865-1 3
13. Amirante, A., Castaldi, T., Miniero, L., Romano, S.P.: Janus: a general purpose
WebRTC gateway. In: 14th Proceedings of the Conference on Principles,
Systems and Applications of IP Telecommunications (2014)
14. Kurento documentation, release 6.6.0 (2016). https://www.kurento.org/
15. Fernandez, L.L., Diaz, M.P., Mejias, R.B., Lpez, F.J., Santos, J.A.: Kurento: a
media server technology for convergent WWW/mobile real-time multimedia
communications supporting WebRTC. In: 14th IEEE International Symposium
on A World of Wireless, Mobile and Multimedia Networks (WoWMoM), pp. 1–6
(2013)
Analytical Modelling of Multi-tier Cellular
Networks with Traffic Overflow
(✉)
Mariusz Gl-abowski
֒ , Adam Kaliszan, and Maciej Stasiak
Chair of Communication and Computer Networks,

Faculty of Electronics and Telecommunications, Poznan University of
Technology, Poznan´, Poland
{mariusz.glabowski,adam.kaliszan,maciej.stasiak}@put.poznan.pl,
http://nss.et.put.poznan.pl/
Abstract. In the paper, a new analytical model of multi-service com-

munication networks with traffic overflow is elaborated. It is assumed
that both: primary resources and alternative resources, forming
networks with traffic overflow, are distributed, as, e.g. in mobile
networks in which traffic from a group of 4G cells is overflowed to a
group of 3G cells. The paper proposes also a new method for
determining parameters of multi- service traffic that overflows from
the distributed primary resources to the distributed alternative
resources, as well as a method for occupancy distribution calculation
and blocking probability calculation in the alternative resources
servicing overflow traffic. The results of the analytical calculations are
compared with the results of the simulation experiments of some
selected structures of overflow systems with the distributed primary
and alternative resources.
1 Introduction
The very dynamic increase in the volume of data traffic in mobile
telecommunications networks (HSPA, LTE, CDMA2000 EV-DO,
TDSCDMA and Mobile WiMax) seems to be posing a real challenge for
mobile operators. The operators of these networks are compelled to
introduce mechanisms that would increase the effectiveness the usage of
network resources. One of such mechanisms is optimization of cell loads,
that can be obtained, among others, due to a concept of self-organising (self-
optimizing, self-configuring) networks (SON) [1]. The execu- tion of this
concept is possible, i.a., due to the well-known connection handover
mechanism that transfers calls from cells with high load to neighbouring
cells. Such mechanism is usually executed between cells operating within the
same technology only, e.g., GSM.
Over time, the increasing diversity of wireless cellular network
technologies led to the emergence of hierarchical (multi-tier) overlay wireless
systems, deploy- ing different technologies, e.g. 2G (GSM), 3G (UMTS), and
4G (LTE) in the same area. The availability of different network
technologies that co-exist in the same area allowed mobile network
operators to apply the well-known optimization technique, the so-called
traffic overflow. Applying the concept of traffic
Technology
DOI: 10.1007/978-3-319-59767-6 21
Analytical Modelling of Multi-tier Cellular Networks with Traffic 257
Overflow
overflow, the mobile operators usually overflow traffic that cannot be
serviced in a group of cells with lower range (e.g., picocells) to a group of
cells with higher range (macrocells) [2–5]. In this case, traffic directed to
macrocells (upper tier, treated as alternative resources) from picocells (lower
tier, treated as primary resources) is the overflow traffic with different
characteristics than the traffic that is directly offered to a network by its
users [2, 4]. As a consequence, the literature distinguishes between two
types of handover: horizontal – when handover of connections takes place
within the same technology – and vertical – when handover of connections
is executed between different technologies (for example, between 2G and
3G, or WiMAX and WiFi).
The traffic overflow mechanism is known in the literature for about 70
years. Its first applications were limited to the single-service (single-rate)
hierarchical telecommunications networks with alternate routing [6–15].
Subsequently, this effective optimization technique was applied to the multi-
service (multi-rate) networks in order to ensure required quality of service
parameters for various traffic classes [16–22], as well as to increase the
performance and capability of networked cloud data centers [23].
In all the models presented in literature the assumption was that each of
primary resources, to which calls are directly offered, can be considered as a
single resource with complete sharing policy and modelled using the so-
called full- availability group model [24, 25]. Such an assumption is justified
in the case of the systems without load balancing between neighbouring cells
of the same tier [17]. However, the modern load-balancing mechanisms (self-
optimization techniques) treat a group of neighbouring cells, covering a
given area, as a single subsystem, and execute horizontal connection
handover (within the same tier/technology) from cells with high load to
neighbouring cells of the group [26]. Consequently, the area covered by a
given group of cells can be treated – from the point of view of traffic
engineering – as a multi-service system in which the service of multi- rate
calls is performed with the use of “separated” (distributed) resources (cells)
that are component parts of a given network and – in consequence – should
be modelled using the so-called limited-availability group model [27, 28].
The present article considers for the first time a system in which both
primary resources and alternative resources of multi-service networks with
traffic overflow are composed of a number of separated (distributed)
component resources. The term “separation” means that a call will be
serviced only when at least one of the primary resources (one of a cell of
the group of cells at the lower tier, e.g., picocells) and at least one of the
alternative resources (one of a cell of the group of cells at the upper tier,
e.g., macrocells) have the appropriate amount of resources that is required
for the call to be serviced.
The paper is organized as follows. Section 2 presents a description of a
multi- service overflow system with distributed primary and alternative
resources. In Sect. 3, the model of the primary and alternative resources of
the multi-service overflow system is presented. Section 4 presents a
comparison of the results of the analytical calculations with the results of
the simulation experiments for some selected structures of systems with
distributed primary and alternative resources. Section 5 sums up the paper.
258 M. G-labowski
֒ et al.
2 Multi-service System with Traffic Overflow

Let us consider a general diagram of an overflow system with distributed
primary and alternative resources. In the system, there are r primary
resources. Each of
A1,1
Am,1 A1,r Am,r
t1,1 ... tm,1 ... tm,r
t1,r
f1 f1 . . . f1 fr fr . . . f r
...
1 2 . . . v1 1 2 . . . vr
R1,1 Rm,1 R1,r Rm,r
2
σ21,1 σm,1 σ21,r 2
σm,r
t1,1 ... t1,r ...
tm,1 tm,r
f0 f0 f0 .. f0
.
1 2 3 .. v0
.
Fig. 1. Traffic overflow system with distributed primary and alternative resources
resource k (0 < k ≤ vr) has the capacity of fk BBUs. The total capacity
Vj of r primary resources are composed of vr component resources. Each
component the primary resources No. j, expressed in the so-called basic
bandwidth units –
BBUs [29], is then:
Vj = vjfj (1)
The given primary resources can admit a call of a given class for service
only when it can be entirely serviced by BBUs of one of the distributed
component resources. This means that ti,j BBUs that are necessary to
service a call of class i in resource j, cannot be “divided” between different
component resources and must be serviced exclusively by the BBUs that
belong to one, randomly selected
component resource. Each of the primary resources j (0 < j ≤ r) is
offered a set of m Erlang traffic streams. The intensity of class i (0 < i ≤
m), offered to resource j (0 < j ≤ r), is denoted as Ai,j (Fig. 1).
The traffic that cannot be serviced by the primary resources is not rejected
but it is offered to the alternative resources. The alternative resources are
composed of v0 components resources. Each component resource of the
alternative resources has the capacity f0. The total capacity V0 of the
alternative resources system:
V0 = v0f0 (2)
As in the case of the primary resources, the alternative resources system can
admit a call of a given class for service only when it can be entirely serviced
by BBUs of one of the distributed component resources.
The alternative resources are offered traffic streams that overflow from
the primary resources. The class i traffic stream offered to the alternative
resources,
traffic of class i (0 < i ≤ m) that overflows from resource j (0 < j ≤ r),
and the is characterised by two parameters: the average value Ri,j of the
intensity of
Overflow
variance
i,j of the intensity of traffic of class i that overflows from resource j
σ2 of
the primary resources system. The variance of overflow traffic is
characterized by values that exceed the average value of this traffic.
3 Analytical Model of Overflow System with Distributed

Resources
3.1 Analytical Model of Primary Resources
In the paper it is assumed that the primary resources as well as the alterna-
tive resources can be analysed using the model of the limited-availability
group (LAG) [27]. The LAG is composed of a number of separated resources
with different capacities. In the case of the LAG, ti,j BBUs, required to set
up a class i connection in primary group j, cannot be “divided” between
different component resources and must be serviced exclusively by the
BBUs that belong to one, randomly selected component resource. Thus, the
LAG is an example of the so-called state-dependent system in which
admission for service of a new call depends on the structure of the system
[30–32].
The occupancy distribution in LAG, modelling the primary resources
with the capacity Vj, can be approximated by the generalised Kaufman-
Roberts distribution [27]:
n[P ] Σm
= )[P (3)
A t ξ (n −
]
n i,j i,j i,j i,j n−ti,j
Vj i=1 t Vj
where ξi,j(n) is the so-called conditional transition probability that defines

which
part of class i stream offered the primary resources j transfers the service
process between the neighbouring states. The conditional transition
probability ξi,j(n) is determined in a combinatorial way as follows:
1) F (Vj − n, v, ti,j −
ξi,j (n) = 1 − (4)
F (Vj − n, v, f )
The combinatorial function F (x, s, f ) determines the number of
arrangements
of x free BBUs in s resources with the capacity of f BBUs each:
⌊ x
F (x, v, f ) ⌋ (−1)u x + v − u(f + 1) −
= Σf (5)
+1
v
1
u
u=0
v− 1
258 M. G-labowski
֒ et al.
Having the occupancy distribution (3), the blocking probability for
individual
multi-service traffic classes offered to the primary resources No. j [27] can be
determined:
Σ
[Ei,j ]Vj Vj
n ] Vj (6)
= n=0
{1 − ξi,j (n)}[P
260 M. G-labowski
֒ et al.
3.2 Parameters of Overflow Traffic that Overflows from the

Primary Resources
Overflow traffic of class i that overflows from the primary resource j is
characterised be the three following parameters: the average value Ri,ji,j ,
variance σ2 and the volume of demands ti,j. The average value of this
traffic is the part of the traffic Ai,j that cannot be serviced in primary by
resources. Having the blocking probability [Ei,j]Vj (Formula (6)), the
average value Ri,j for traffic of class i that overflows from the resource j
will be determined by the following formula:
Ri,j = Ai,j · [Ei,j ]Vj (7)
To determine the variance of overflow traffic we use the approximate method
proposed in [17]. According to this method, a decomposition of each of the
primary resources Vj into m fictitious resources with the capacity i,j is to
V∗
be carried out first. The assumption is that each fictitious group will service
exclusively calls of one class, which provides an opportunity to apply
Riordan formula [7] to determine the variance of traffic of class i that
overflows from
the resource j. The capacity of the fictitious resource i,j is defined as this part
V∗
of the real resource Vj that is not occupied by calls of the remaining classes
(different from the class i) [17]. Eventually, we get:
V∗ = V
i,j j Σm l,j l,j (8)
l=1,l/=i
— Y t
where Yl,j is the average value of traffic of class l serviced in the resource j, i.e.,
the average number of calls of class l serviced in the resource j. According to
the definition of traffic intensity, we have:
Yl,j = Al,j(1 − [El,j]Vj ) (9)

Having the parameters Ai,j, Ri,j and Vi,j we can determine the variancei,jσ2
∗
for individual call streams that overflow from the primary resource j to the
system of alternative resources, using the Riordan’s formula [7]:
2 Ai,j
σi,j = Ri,j ⎞
i,j
⎝V∗
, (10)
ti,j
i,j
+ 1 − Ai,j + +1−R ⎠
Ri,j
In Formula (10), the quotient i,j
V ∗ /ti,j normalizes the system to a single-service
case. The operation of this kind is necessary because Riordan Formulas (7) and
(10) determine the parameters of overflow traffic in single-service systems.
At this point we already have all the parameters that characterize traffic
streams offered to the system of alternative resources and we are in position
to determine their peakedness coefficients:
Overflow 1
Zi,j = σ2 /Ri,j (11)
i,j
260 M. G-labowski
֒ et al.
3.3 Analytical Model of Alternative Resources

In order to model the alternative resources with overflow traffic, we can
apply the same approach as in [21]. In [21], the overflow system with state-
independent non-separated primary resources and with distributed
(separated) alternative resources was considered. To model the state-
dependent alternative resources, in [21] the Hayward approach was applied
[33]. As a result, the occupancy distribution [Pn] and the blocking
probability Ei,j for the alternative distributed resources with the capacity of
V0 BBS, can be written as follows [21]:
r m
Σ
n[P ] Σ Ri,j
n V0 /Z0 = i,j ξi,j i,j )[P n−ti,j ]V0 /Z0 (12)
t
Z i,j (n − t
j=1 i=1
[E ]
ΣV0 /Z0 (13)
(n)}[P
= {1 −
i,j V0 /Z0 i,j n V /Z
n=0 ] 0 0
ξ
while the the transition probability ξi,j(n) is determined on the basis of Eq.
(5),
as follows:
F V0
Z0
− n, v0, ti,j − 1
F Z00 − n, v0, Z00

ξi,j(n) = 1 − (14)
V f
The parameter Z0 in Formulas (12)–(14) is the so-called aggregated
peaked-
ness factor and is calculated as a weighted average of peakedness factors for
individual traffic classes Zi,j offered to secondary resources. A weight related
to the share of the coefficient Zi,j in the aggregated peakedness coefficient Z0
is directly proportional to this number of BBUs that is required to service
offered traffic Ri,j:
Z Σr Σm Ri,jti,j
0
j=1
i=1
Zi,j Σr Σm Rl,ut
l,u
(15)
= u=1 l=1
Formulas (12)–(14) make it possible to determine all important

characteristics of the overflow system with distribution of alternative
resources, i.e., the occupancy distribution [P ]V0 /Z0 (Formula (12)) and the
blocking probabilities [Ei,j]V0 /Z0 (Formula (13)) for particular traffic classes.
4 Numerical Results
The proposed method of modelling the overflow systems with distributed
both primary and alternative resources is an approximate one. In order to
evaluate the accuracy of the proposed method, the results of the analytical
calculations were compared with the results of the simulation experiments
for a number of selected overflow system.
To evaluate the accuracy of the proposed model the results of the
analytical calculations were compared with the data provided by the
Overflow experiments. For this particular purpose, a dedicated simulator
simulation 3
of the considered networks at the call level was constructed [34]. The
approach to modelling of
262 M. G-labowski
֒ et al.
multiservice systems with traffic overflow at the call level adopted in this
article makes it possible to use it to determine traffic characteristics, e.g.
hierarchical wireless networks [31]. The results of the simulation are
presented in Figs. 2, 3, 4, 5 and 6 in the from of appropriately marked
points with 99-percent confidence interval calculated after the t-Student
distribution for the number of ten series.
Fig. 2. Blocking probability in System 1
The parameters of the considered systems were as follows.
Parameters of the analyzed systems No. 1

1. Primary resources
–
– The number
Resources 1, of primary
2 and 3 (allresources r=3
primary resources have the same structure):
• f1 = 20 BBUs, v1 = 3, V1 = V2 = V3 = 3 × 20 = 60 BBUs,
• offered traffic: t1,∗ = 1 BBUs, t2,∗ = 2 BBUs, t3,∗ = 4 BBUs,
• proportion of offered traffic: A1,1t1,1 : A2,1t2,1 : A3,1t3,1 = 1 : 1 : 1;

A1,2t1,2 : A2,2t2,2 : A3,2t3,2 = 1 : 1 : 1; A1,3t1,3 : A2,3t2,3 : A3,3t3,3 =
1 : 1 : 1;
2. Secondary resources
– f0 = 20 BBUs, v0 = 6, V0 = 120 BBUs.

–
– The number
Resources 1: of primary resources r = 3
• f1 = 20 BBUs, v1 = 3, V1 = 3 × 20 = 60 BBUs,
Overflow

264 M. G-labowski
֒ et al.

Overflow
• offered traffic: t1,1 = 1 BBUs, t2,1 = 2 BBUs, t3,1 = 4 BBUs,
– Resources 2:
• f2 = 30 BBUs, v2 = 2, V2 = 2 × 30 = 60 BBUs,
– Resources 3:
• f3 = 16 BBUs, v3 = 4, V3 = 4 × 16 = 64 BBUs,

– f0 = 20 BBUs, v0 = 6, V0 = 120 BBUs.

– Resources
– The number1, of primary
• f1 = 20 BBUs, v1 = 3, V1 = V2 = V3 = 3 × 20 = 60 BBUs,

A1,2t1,2 : A2,2t2,2 : A3,2t3,2 = 1 : 1 : 1; A1,3t1,3 : A2,3t2,3 : A3,3t3,3 =
1 : 1resources
2. Secondary : 1;
– f0,1 = 20 BBUs, v0,1 = 3, f0,2 = 30 BBUs, v0,2 = 2, V0 = 3 · 20 + 2 · 30
=
120 BBUs.

–
– The number
Resources 1, of primary
• f1 = 20 BBUs, v1 = 2, V1 = V2 = V3 = 2 × 20 = 40 BBUs,

A1,2t1,2 : A2,2t2,2 : A3,2t3,2 = 1 : 1 : 1; A1,3t1,3 : A2,3t2,3 : A3,3t3,3 =
1 : 1 : 1;
– f0 = 30 BBUs, v0 = 4, V0 = 120 BBUs.

–
– The number
Resources 1: of primary resources r = 3
• f1 = 20 BBUs, v1 = 2, V1 = 2 × 20 = 40 BBUs,

266 M. G-labowski
֒ et al.

– Resources 2:
• f2 = 30 BBUs, v2 = 2, V2 = 2 × 30 = 60 BBUs
– Resources 3:
• f3 = 20 BBUs, v3 = 3, V3 = 3 × 20 = 60 BBUs,

– f0 = 30 BBUs, v0 = 4, V0 = 120 BBUs.
The analytical and simulation results of the blocking probability in the
alternative resources of the overflow system with distributed primary
resources are presented in Figs. 2, 3, 4, 5 and 6.
The results of the present study clearly demonstrate that the accuracy of
the method is not dependent on either the number and structure of LAGs
that compose primary groups or LAGs that form secondary groups. In the
same way, a differentiation in demands of individual classes (the parameter
t) has no influence on the accuracy of the proposed method which is
comparable with the accuracy of calculations for a single LAG.
The results of the blocking probability for the traffic classes with the
highest demands obtained on the basis of the simulation are marginally
higher than the results obtained on the basis of the proposed analytical
method. For classes with the lowest demands, the situation is reversed – the
results of the simulation slightly exceed the results of the calculations.
This particular dependence between the results of the analytical modeling
and the results of simulation experiments is, however, convenient from the
practical point of view. The same degree of overestimation applies to and
involves those traffic classes that are characterized by the highest blocking
probability and, in consequence have the highest influence on the operation
of a system. Engineering rules for designing network systems are consistent
with the principle of dimensioning that is based on the “worst case”
scenario. Within this context, a slight overestimation of the QoS parameters
for the traffic classes that have the highest impact on the operation of a
system fits well into the principle.
5 Conclusion
The paper proposes a new analytical model of multi-service overflow
systems with distributed primary and alternative resources. The results of
analytical modelling have been compared with the simulation results. The
accuracy of the proposed method is comparable with the accuracy of the
classical models elaborated for modelling the overflow systems with fully
available primary resources. The model can be applied for analytical
modelling of multi-service multi-tier cellular network.
Overflow
References
1. Nagesh, C.B., Jayaramaiah, D.: A survey of self-organizing networks (son).
Int. J. Innov. Res. Comput. Commun. Eng. 3(4), 2916–2923 (2015).
http://www.rroij.com/abstract.php?abstract id=56994
2. Tripathi, N., Reed, J., VanLandinoham, H.: Handoff in cellular systems. IEEE
Pers. Commun. 5(6), 26–37 (1998)
3. Fernandes, S., Karmouch, A.: Vertical mobility management architectures in
wireless networks: a comprehensive survey and future directions. IEEE
Commun. Surv. Tutorials 14(1), 45–63 (2012)
4. Lin, Y.-B., Chang, L.-F., Noerpel, A.: Modeling hierarchical microcell/macrocell
PCS architecture. In: 1995 IEEE International Conference on Communications,
ICC 1995, Seattle. Gateway to Globalization, vol. 1, pp. 405–409, June 1995
5. Li, S., Grace, D., Wei, J., Ma, D.: Guaranteed handover schemes for a
multilayer cellular system. In: 7th International Symposium on Wireless
Communication Sys- tems, pp. 300–304, September 2010
6. Bretschneider, G.: Die Berechnung von Leitungsgruppen f u¨ r berfließenden
Verkehr in Fernsprechwählanlagen. Nachrichtentechnische Zeitung (NTZ) (11),
533–540 (1956)
7. Wilkinson, R.: Theories of toll traffic engineering in the USA. Bell Syst. Techn. J.
40, 421–514 (1956)
8. Matsumoto, J., Watanabe, Y.: Theoretical method for the analysis of queueing
system with overflow traffic. Electron. Commun. Japan (Part I: Commun.)
64(6), 74–83 (1981). http://dx.doi.org/10.1002/ecja.4410640610
9. Rapp, Y.: Planning of junction network in a multi-exchange area. In:
Proceedings of 4th International Teletraffic Congress, London, p. 4 (1964)
10. Herzog, U.: Die exakte Berechnung des Streuwertes von u¨berlaufverkehr
hinter Koppelanordnungen beliebiger Stufenzahl mit vollkommener bzw.
unvollkommener Erreichbarkeit. A E U¨ 20(3) (1966)
11. Schehrer, R.: On the exact calculation of overflow systems. In: Proceedings of
6th International Teletraffic Congress, Munich, pp. 147/1–147/8, September
1970. https://itc-conference.org/
Resources/Persistent/21594392c496c4c0193b4864ec86
0502ebdec2bc/schehrer70.pdf
12. Bretschneider, G.: Extension of the equivalent random method to smooth
traffics. In: Proceedings of 7th International Teletraffic Congress, Stockholm
(1973)
13. Schehrer, R.: On the calculation of overflow systems with a finite number of
sources and full availiable groups. IEEE Trans. Commun. 26(1), 75–82 (1978)
14. Fredericks, A.: Congestion in blocking systems - a simple approximation
technique. Bell Syst. Techn. J. 59(6), 805–827 (1980).
http://onlinelibrary.wiley.com/doi/ 10.1002/j.1538-7305.1980.tb03034.x/abstract
15. Shortle, J.F.: An equivalent random method with hyper-exponential service. J.
Perform. Eval. 57(3), 409–422 (2004)
16. G-la֒bowski, M., Hanczewski, S., Stasiak, M.: Erlang’s ideal grading in diffserv
modelling. In: Proceedings of IEEE Africon 2011, pp. 1–6. IEEE, Livingstone,
Septem- ber 2011. http://dx.doi.org/10.1109/AFRCON.2011.6072139
17. G-labowski,
֒ M., Kubasik, K., Stasiak, M.: Modeling of systems with overflow
multi- rate traffic. Telecommun. Syst. 37(1–3), 85–96 (2008)
268 M. G-labowski
֒ et al.
18. G-la֒bowski, M., Kubasik, K., Stasiak, M.: Modelling of systems with overflow
multi- rate traffic and finite number of traffic sources. In: Proceedings of 6th
International Symposium on Communication Systems, Networks and Digital
Signal Process- ing 2008, Graz, pp. 196–199, July 2008.
http://dx.doi.org/10.1109/CSNDSP.2008.
4610800
19. Huang, Q., Ko, K.-T., Iversen, V.B.: Approximation of loss calculation for
hierarchical networks with multiservice overflows. IEEE Trans. Commun. 56(3),
466–473 (2008)
20. G-labowski,
֒ M., Kaliszan, A., Stasiak, M.: Two-dimensional convolution algorithm
for modelling multiservice networks with overflow traffic. Math. Prob. Eng.
2013,
18 (2013). Article ID 852082. http://dx.doi.org/10.1155/2013/852082
21. G-la֒bowski, M., Kaliszan, A., Stasiak, M.: Modelling overflow systems with dis-
tributed secondary resources. Comput. Netw. 108, 171–183, (2016). http://www.
sciencedirect.com/science/article/pii/S1389128616302675
22. G-labowski,֒ M., Hanczewski, S., Stasiak, M.: Modelling of cellular networks
with traffic overflow. Math. Prob. Eng. 2015, 15 (2015). Article ID 286490.
http://dx.doi.org/10.1155/2015/286490
23. Ku¨hn, P., Mashaly, M.E.: Multi-server, finite capacity queuing system with mutual
overflow. In: Fiedler, M. (ed.) Proceedings of 2nd European Teletraffic Seminar,
pp. 1–6. Karlskrona, September 2013. paper S1 2
24. Kaufman, J.: Blocking in a shared resource environment. IEEE Trans. Commun.
29(10), 1474–1481 (1981)
25. Roberts, J.: A service system with heterogeneous user requirements –
application to multi-service telecommunications systems. In: Pujolle, G. (ed.)
Proceedings of Performance of Data Communications Systems and their
Applications, pp. 423–
431. North Holland, Amsterdam (1981)
26. G-labowski,
֒ M., Hanczewski, S., Stasiak, M.: Modelling load balancing
mechanisms in self-optimising 4G mobile networks with elastic and adaptive
traffic. IEICE Trans. Commun. E99-B(8), August 2016
27. Stasiak, M.: Blocking probability in a limited-availability group carrying mixture
of different multichannel traffic streams. Ann. Télécommun. 48(1–2), 71–76 (1993)
28. G-labowski, ֒ M., Stasiak, M.: Multi-rate model of the group of separated
transmission links of various capacities. In: Souza, J.N., Dini, P., Lorenz, P.
(eds.) ICT 2004. LNCS, vol. 3124, pp. 1101–1106. Springer, Heidelberg (2004).
doi:10.1007/
978-3-540-27824-5 143
29. Roberts, J., Mocci, V., Virtamo, I. (eds.): Broadband Network Teletraffic, Final
Report of Action COST 242. Commission of the European Communities.
Springer, Berlin (1996)
30. G-labowski,
֒ M., Kaliszan, A., Stasiak, M.: Modeling product-form state-dependent
systems with BPP traffic. Perform. Eval. 67, 174–197 (2010)
31. Stasiak, M., G-labowski,
֒ M., Wi´sniewski, A., Zwierzykowski, P.: Modeling and
Dimensioning of Mobile Networks. Wiley, Chichester (2011)
32. G-labowski, M., Hanczewski, S., Stasiak, M., Weissenberg, J.: Modeling Erlang’s
ideal grading with multi-rate BPP traffic. Math. Prob. Eng. 2012, 35 (2012).
Article ID 456910
33. Ershova, E., Ershov, V.: Digital Systems for Information Distribution. Radio and
Communications, Moscow (1983). in Russian
34. G-la֒bowski, M., Kaliszan, A.: Simulator of full-availability group with
bandwidth reservation and multi-rate Bernoulli-Poisson-Pascal traffic streams. In:
Proceedings of Eurocon 2007, Warszawa, pp. 2271–2277, September 2007.
http://dx.doi.org/ 10.1109/EURCON.2007.4400605
New Technologies
Multi-level Stateful Firewall Mechanism
for Software Defined Networks
Fahad Nife1,2(✉) and Zbigniew Kotulski1

1
Faculty of Electronics and Information Technology,
Warsaw University of Technology, Warsaw, Poland
2
Faculty of Science, Al-Muthanna University, Muthanna, Iraq
Abstract. Traditional networks are often quite static, slow to modify,

dedicated for a single service and very difficult to scale, what is typi-
cal for a large number of different network devices (such as switches,
routers, firewalls, and so on), with many complex protocols
implemented or embedded on them. Software Defined Network (SDN)
is a new technology in communication industry that promises to
provide new approach attempting to overcome this weakness of the
current network paradigm. The SDN provides a highly scalable and
centralized control architecture in which the data plane is decoupled
from the control plane; this abstraction gives more flexible,
programmable and innovative network architecture. However,
centralization of the control plane and ability of programming the
network are very critical and challenging tasks causing security
problems. In this paper we propose a framework for securing the SDN
by introducing an application as an extension to the controller to
make it able to check every specific flow in the network and to push
the security instructions in real-time down to the network. We also
compare our proposal with other existing SDN-based security
solutions.
Keywords: Software Defined Network · Stateful-Packet filtering ·

Flow table · OpenFlow protocol
1 Introduction
In computer networking, the emerging network architecture that decouples
the control plane from the data plane, logically centralizes the network
intelligence and state, and abstracts the underlying network infrastructure
from the applications is known as Software Defined Network [1]. SDN is a
new framework, where the control plane of all devices in the network is
migrated from the forwarding plane and consolidated into a software-based
device called the controller. This controller is responsible for making and
forwarding decisions for the entire network, while the switches and routers
just perform basic packets forwarding. The controllers have full view of the
network and consider the totality of the network before making any
decision. Since SDN relies more on software and, because of
DOI: 10.1007/978-3-319-59767-6 22
272 F. Nife and Z. Kotulski
its network intelligence, SDN is logically centralized [2], it supports rapid

innovation and provides new ways and opportunities to enhance and
protect the network.
In this paper we introduce a stateful firewall solution for securing an
open network using SDN technology. This proof of concept proposal aims to
build a reactive Stateful SDN-Based Traffic Filtering framework for SDN
networks by getting the advantages of using the OpenFlow protocol
messages, which are exchanged between the controller and the OpenFlow-
enabled switches. It extends the controller using a software which makes use
of some tables used to keep track of each flow in the network to provide
stateful packet filtering. The method relies on dividing the network into
planes to create a multi-plane system of defense and to make it easy
protecting the network from external attacks as well as from internal
malicious users. The network security policy will be centralized in the
controller, where there is the place of making decision regarding how the
switches should handle the packets, while a switch will only enforce this
decision. The controller can at any time reconfigure the security rules and
redeploy them at any device under its control. Thus, we have more efficient,
flexible and centralized point of security management represented by the
controller, which in turn spreads the network security policy dynamically
across the distributed checkpoints.
The rest of this paper is organized as follows. Section 2 describes the
theoretical background of SDN, OpenFlow and firewalls. Section 3
presents state of the art of current SDN security solutions. The overview of
the proposed system’s architecture is specified in Sect. 4. The analysis of the
new prototype solution and its comparison to other known SDN-based
firewalls are presented in Sect. 5, while Sect. 6 concludes the paper with a
summary of its content and future enhancements.
2 Background
To relate our proposal to other existing network solutions and to give the
background of the suggested method along with its functionality and
components, we start from a brief description of the SDN architecture, the
OpenFlow protocol and firewalls.
2.1 SDN Architecture

The OpenFlow-based SDN architecture mainly consists of two planes, see
Fig. 1, the control plane and the data plane [3]. The control plane strives to
act as a network operating system and it is responsible for programming
and managing the data plane. It makes use of the information provided by
the data plane and defines network operation and routing [4]. The other
plane, which is the data plane, comprises the forwarding network elements
that are connected to each other. It is responsible for data forwarding, as
well as monitoring the network
Multi-level Stateful Firewall Mechanism for 273
SDN
by gathering local information and statistics [5]. For more eligibility and
scala- bility in SDN networks, network services provided by the network
applications are abstracted as a separate plane, which may introduce new
network features, such as security, QoS, load balance, bandwidth
management, and so on. Two interfaces that must be implemented by the
control plane are the Northbound interface (NBI) and the Southbound
Interface (SBI). The Northbound Inter- face presents a programmable API
used by the application plane to receive an abstracted and global view of
the network from the controllers and then use that information to provide
an appropriate guidance back to the control plane. Typically, control
software provides its own API to applications as there is no standardized
API today [6]. The Southbound Interface (e.g. OpenFlow) is used for the
communication between the controller and switches. The data plane ele-
ments use the SBI to reveal their capabilities toward the controller (Control
Plane). It is used to provide grammatic control, statistics reporting and
events notifications.
Fig. 1. Software-Defined Network abstraction
2.2 OpenFlow
The most commonly used protocol for the Southbound interface of SDN is
the OpenFlow protocol [7]. It is standardized by the Open Networking
Foundation
(ONF) and it is typically responsible for establishing the communication

through secure control channel between the controller and switches under
its control. An OpenFlow-enable switch is a basic forwarding device that
forwards packets according to its flow table. The flow table contains a set of
flow entries (also called flow rules) each of which consists of headers fields,
activity counters, priority, timeout, actions, cookie and flags, see Fig. 2.
Fig. 2. OpenFlow table
The header fields (Match fields) is 15-tuples packet header to be

matched against the incoming packet’s header. A Priority field is to specify
which flow should be selected in case of more than one flow entry match the
incoming packet. Counters fields are updated only when packets are matched
and used to keep track of each flow, table and port and provide statistics to
the Controller. The action fields contain data, where there is a set of zero or
more actions that should be applied to the matched packets. The possible
actions include forwarding the incoming packet to specific port(s),
encapsulating and sending the packet to the controller, dropping the packet,
or checking it with the next flow table or with specific table in the pipeline.
Timeout fields specify the validity of the flow entry before expiration. The
Cookie field is an opaque data value chosen by the controller. It can be used
by the controller to filter the flow entries that affected by flow statistics,
modification and deletion requests. Cookie is not used when processing
packets. Flags fields alter the way flow entries are managed. The flow tables
of the switch are manipulated by the controller, where it can modify the
behavior of the switches with regard to forwarding by inserting, modifying
and removing the switch flow entries.
Every packet received by the switch, the lookup process begin with the
first flow table entries till the last flow table in the pipeline, the process end
either with match and execute the corresponding action for the matched
entry or with miss (no rule match the packet) and execute the default rule
which is either send
SDN
the packet to the controller or drop the packet. If matching occurs, the
action field for that entry is applied on that packet, While in case of no
matching, the default rule will applied which may encapsulate the packet
(or part of the packet, it is common the packet header and the number of
bytes defaults to 128 if the switch is capable of buffering) with Packet-in
message and send it to the controller. The controller according to its policy
rules will take a decision of how should that switch handle the packet and
forward it back to the requested switch. According to that decision, the
switch will apply it on the packet and modify its flow table by adding new
entry to help handling next similar packets [8].
2.3 Firewall
Firewalls are fundamental security mechanisms for any network defense
systems. Firewall is usually placed between an authorized system and
external environments to act as a gateway through which all inbound and
outbound network traffic must pass [9]. It is checkpoint that compares each
incoming packet with its policy rules to decide whether to allow or deny the
packet from pass through it. Each firewall have its policy rules database
which define how an firewall should handle incoming and outgoing network
traffic for some specific IP addresses and range of addresses, protocols,
applications, and packet content types based on the organization’s security
policies [10]. Firewalls can be software or hardware-based and can be
categorized according to deferent criteria such as their functionality or the
layer (ISO/OSI Reference Model and TCP/IP model) they operate: the
Network, Session, Transport or Application layer. The most common fire-
walls that work on the network and transport layers are Stateless Firewalls
and Stateful Firewalls.
Stateless Firewall. It enforces security policies within a network by employ

a filtering process for incoming and outgoing packets. It is looks at the IP
address of the packet and compare it against predetermined Rule set defined
by the network administrator to decide whether it should be allowed
through or not [11]. Each packet maintained separately, so it does not need
to save the state or any information about the packet to be used for
processing the next packet in the same flow, see Fig. 3.
Stateful Firewall. It attempts to track the state of network connections

when filtering packets. It often makes use of a table called state table to
store information about each traffic flow passing through. This state table
holds entries that represent all the communication sessions of which the
device is aware. Every entry holds specific information (e.g. source and
destination IP addresses, Port numbers, flags, sequence and
acknowledgment numbers) that uniquely identifies the communication
session it represents [12]. For each connection started out through the
stateful device, a state table entry is created. When the inspected packet
matches an existing firewall rule that permits it, the packet is passed and
new entry is added to the state table, see Fig. 4. Then, when traffic returns,
the
Fig. 3. Stateless firewall
packet’s information compared against the state table information to

determine whether it is part of a currently logged communication session.
The packet is allowed to pass only if it is related to a current table entry.
3 Related Works
Several recent efforts in the field of SDN security have been presented in
many previous contexts, such as scanning attack prevention [13], DDoS
attack detection [14], IDS [15], IPS [16], Stateless firewall and Stateful
firewall.
FortNox [17] is an extension to the open source NOX controller. The
security policy enforcement kernel implement a novel analysis algorithm to
prevent any of application attempts to insert flow rules that may change
flow rules enforced by security policy. FortNox is a real-time rule conflict
detection engine, where after detecting any conflict, it is either accept or
refuse the new rule based on rule-based authorization. FRESCO [18]
introduce security application framework consists of several reusable
modular libraries that can be connected together to build more sophisticated
security applications. FRESCO developers can easily connect the necessary
modules including its libraries and assigning values to its interfaces to
produce the required security functions, like firewalls, IDS, or scan
detections. FLOWGUARD [19], represents a comprehensive SDN framework
for facilitate accurate detection and flexible resolution of firewall policy
violations in dynamic OpenFlow networks. FLOWGUARD provide an
automatic and real- time violation resolution mechanism, when the network
states are updated or the configurations are dynamically changed, it detect
firewall policy violations by checks the flow path spaces and compare it
against the specified authorized space in the firewall.
SDN
Fig. 4. Stateful firewall
FleXam [20, 21], proposes a sampling method that make the controller
able to access packet-level information that is necessary for different
security applications. With FleXam, the controller can decide which part
of the packet (e.g. only headers or, maybe, only payload) and where they
should be sent. More- over, the packet can be sampled either stochastically
based on predetermined probability or deterministically based on pattern.
The authors of [22] propose a prototype in which the security rules are
specified in the flow table in both controller and switches. The switch acts
as simple packet-pusher based on its flow table and any unknown traffics
are sent to the controller for inspection. The prototype proposed in [23]
provides orchestration services for security policy management, reactive
application that processing the state of the connection, the security are
expressed with OpenFlow rules and in the data plane and the generic
algorithm is introduced for processing the Finite State Machine (FSM) of
the TCP protocol. Proactive Security Mechanism has been introduced in
[24]. In that work, fuzzy based system was designed to evaluate the threat
level of identified threats. The authors of [25] introduce an SDN-based
mitigation system for ransomware threats. Their solution is based on
CryptWall findings and the two designed SDN1 and SDN2 mitigation
ransomware methods, which rely on utilizing a dynamic blacklist database
of up-to-date list of known proxy servers that used for the purpose of
ransomware. Both methods try to break the communication between the
victim and the infected C&C server, so no public key will retrieve and as a
result the encryption process will not happen.
4 System Architecture
4.1 Overview
The diagram presented in Fig. 5 shows that the system’s architecture is
compatible with the SDN structure: it consists of three planes distributed
into the SDN planes. The highest plane involves the system functionality,
which is executed by different modules, responsible for maintaining the
tables in the other planes and providing the required security functions to
protect the SDN network. The second firewall plane consists of the set of
tables that are built in the control plane. These tables are used to improves
the overall performance by reducing the lookup process time needed to
handle a packet. The most down plane provides the flow state-aware using
tables placed in the OpenFlow-enabled switches and holds entries used to
keep track for each specific flow.
Fig. 5. System architecture overview
The functionality of the firewall application that run on a top of the

controller are executed by 7 modules, which in turn maintain tables in,
both, the controller and the OpenFlow-enabled switches. The first module
is Main Module. It is the control unit that play the coordinator and
manager roles for other units. The other module is Event Listener which is
triggered by some predefined events and which responds with an action
corresponding to that event. The State Extractor module is responsible for
extracting some information from the
SDN
packet. The SecPol Div module is initiating SecPolTable for each switch
joins to the controller. Each entry in the SecPolTable table represents the
security policy for that switch prepared by the network administrator. The
other table is STable. It is maintained in the switch and used to keep track
of each traffic flow pass through the switch. SecPol Lookup module is
responsible for checking out the SecPolTable for all switches in the whole
path for that packet from the source to the destination if it is inside the
network or to the gate way switch if the final destination is outside the
network. Con State module is used to install new entry in STable of all
switches in the path using S PROTOCOLE protocol, this approach is to
avoid additional delays (switch-controller communication) when the first
packet arrives for next switches in the path. The last module is Mode Selec,
which is responsible of switching between the tow control behavior modes.
Now, before explaining the main components and tables of the proposed
stateful application, let us discuss interactions among OpenFlow switches,
controllers and the stateful firewall application. Every time an OpenFlow-
enabled switch receives a packet, it parses the packet’s header, which is
matched against the STable entries before the flow table pipeline for any
match. If the match occurs, the packet will forwarded meaning that the
packet is part of previous flow, otherwise no matching occurs, The matching
will be against the flow table entries to do the action related with matched
entry, or if no matching entry, the packet will encapsulated with Packet-in
message and forwarded for the controller to decide how to handle the packet,
see Fig. 6.
Fig. 6. Packet flow in stable enabled open flow switch

Once the controller receives the Packet-in notification, it identifies the

correct action for the packet according to the security policy stored in the
SecPolTable corresponding for that switch and then using Packet-out
messages installs one or more appropriate entries in the flow table of
requesting switch. The event of the controller sends the Packet-out message
that will trigger the Event Listener object to Call other three functions,
State Extractore object, which in turn extracts some information and
attributes that make up the communication session (e.g. IP addresses,
Port numbers, Sequence and Acknowledgment numbers, and flags) which
can all combined to be used as a fingerprint for an individual connection.
The other object is SecPol Lookup, which is responsible for checking out
the SecPolTable for all switches in the whole path for that packet to be sure
it does not violate the security policy of the network, if so, and finally the
Con State module to install entry in STable of corresponding switches using
S PROTOCOLE Add message and all switches in the path, this approach is
to avoid additional delays (switch-controller communication) when the first
packet arrives for next switches in the path.
One more important introduced mechanism done by the SecPol Div,
which for each data plane device join to the controller, this SecPol Div
object will ini- tiate SecPolTable for that new connected switch, this
mechanism will improve the overall performance by reducing time needed
by the controller to decide how should that switch handle the packet if it
does not match any entry in its flow table. When security policy updated
by the network administrator, the SecPol Div will rebuild all the
SecPolTable and reinforce and dynamically propagate the security policies
for every data plane.
To specify the controller behavior on the flow and STable entries, 2
modes Norm Mode and Critic Mode are implemented. The Mode Selec
module is responsible of switching between this modes. The Norm Mode is
the default mode in which the Timeout field in the STable is started with
60 s and there is no limit on the number of entries of STable, once the S
PROTOCOLE enabled switch exposure to onslaught or its Stable nearly to
full (exceed the threshold) it switch will send S PROTOCOLE Critical
message to the controller and accord- ingly the Mode Selec module will
implement the Critic Mode in which rather than start with 60 s as timeout
for establishing the flow, it will to be 15 s, and put limit on the number of
entries of the STable per source and destination IP, per source and
destination ports, or even per rule policy to help reducing the risk of fulling
the STable. When the STable entries decreased to some reasonable size the
switch will send S PROTOCOLE Normal message to the controller to
inform the Mode Selec module to implement the Norm Mode.
4.2 STable Table and SecPolTable Table

STable is the table generally implemented as a hash table, which typically
resides in the OpenFlow-enabled Switch RAM and is used to keep track of
each valid connection. STable offers stateful inspection capability by records
the state of a connection based on protocol specific information. Each new
connection requires adding new entry in the STable for the connection. The
STable entry includes
SDN 1
the following information: Protocols, Source IP addresses and Destination
IP addresses, Source port and Destination port, Reversed Source and
destination IP addresses and ports (to represent response traffic), the
protocol being used for the connection, the time specifies the validity of rule
before removing it, the TCP state of the connection (in a case TCP is
used), the sequence number, and the acknowledgment number, see Table 1.
The stateful firewall application maintains this table using Add, Remove,
and Clear messages of the designated protocol called S PROTOCOLE, see
Sect. 4.3.
The other table is SecPolTable table. It is simple table consists of rows
and columns initiated by the SecPol Div module for each specific switch join
to the controller and used to holds Security Policy related for that switch.
This table will improve the overall performance by reducing time needed by
the controller to decide how that switch should handle the packet if it is not
match any entry in its flow table. Any change of the network security by
the network administrator will enforce the SecPol Div to rebuild all the
SecPolTable tables and dynamically propagate the security policies for every
data plane.
Table 1. Example of STable Table’s entries
Prtcl Src IP Dst IP SrcPrt DstPrt State SeqNo AckNo Timout

TCP 192.168.10.9 172.16.2.23 1054 21 SYN SENT 0 0 93
TCP 172.16.2.23 192.168.10.9 21 1054 ESTABLISHED 1 726 2845
... ....... ....... .. ... ....... .. ... ...
4.3 S PROTOCOLE
It is designated communication protocol between the controller and the
OpenFlow-enabled switch. The protocol S PROTOCOLE defines five mes-
sages. Three of them: S PROTOCOLE Add, S PROTOCOLE Remove
and S PROTOCOLE Clear are to enable state-based connection monitoring
and to help the ConState module to maintain the STable by adding,
removing and clearing the STable entries of the data plane. The two other
messages: S PROTOCOLE Normal and S PROTOCOLE Critical are to
help the Mode Selec restricting the controller behavior. S PROTOCOLE
with its messages is used only by the ConState module and it is not
affected by other SDN Modules and applications. Table 2 shows the direction
of the messages flow along with descriptions of their functionality.
5 Analysis
We now estimate effectiveness of the presented solution, such performance
analysis calculated for memory consumption, processing times and
throughputs and vulnerabilities mitigations.
Table 2. S PROTOCOLE messages
Message name Direction Description

S PROTOCOLE Add Controller to Switch Add new STable entry for the new
traffic
S PROTOCOLE Remove Controller to Switch Delete specific entry from the
STable
S PROTOCOLE Clear Controller to Switch Empty the STable
S PROTOCOLE Normal Switch to Controller Enforce the Mode Selec module to
implement the Norm Mode
S PROTOCOLE Critical Switch to Controller Enforce the Mode Selec module to
implement the Critic Mode
In order to compute memory consumption required by our approach, we

need to focus on the size of STable which run in OpenFlow-enabled switch.
As explained in Sect. 4.2 each STable entry is 26-byte, so for STable with
10000 entries the required amount of memory is 26 MB, while for STable
with 100000 entries the required amount of memory is 260 MB, which can
be considered a reasonable amount of memory. The other data structure
used are SecPolTable tables does not require large memory as it divide the
rule table into small tables, one main table for shared rules and other tables
for every joined switch.
Optimization in our proposed prototype happens in different places: (1)
using the STable, that will help to increase performance and security by
associating the individual packets with their respective flows, so any match
between the incoming packet with any Stable entry lead to forward the
packet to its destination through the appropriate port(s) without requiring
any extra computation or controller communication overhead; (2) by rule
base checking, where in SDN, every Packet-in arrives some algorithm must
be used to check the rules and decide how to handle the packet. An essayist
approach is to scan every rule in priority order until a match is found or the
rule base is exhausted. This linear search through the rule base is
convenient for small rule policies, but it is inef- ficient for larger policies,
particularly when match done with the default rule or the last rule entry of
the policy. So the worst-case, per packet running time (for each Packet-in
from each switch) is O(n), where n is the number of entries in the rule base.
In contrast, let us calculate it for our proposed solution, for ease of mea-
surement we break up the process into two parts: the first part searches the
rule policy and builds up SecPolTable and the second part is scanning the
SecPolTable policy rules to find the match. Let j be the number of switches
connected to the SDN controller and n the number of policy rules. Building
each SecPolTable table requires scanning through the policy file or just the
remaining of it to determine which entry belongs to it. By doing so, the first
joined switch requires scanning the whole policy file with O(n) processing
time, producing first SecPolTable with, say, n1 entries. Now for the next
joined switch,
SDN
Table 3. Comparison of the proposed approach with other SDN security solutions
Publication Problem Proposed Behavior Extra Capabilities Performance

approached solution memory exploited
Collings, J. Dynamic Stateful H/W Reactive No Centralized High com-
et al. [22] packet firewall control munication
filtering overhead
Zerkane, S. DoS attack Stateful/ Reactive Yes Centralized High com-
et al. [23] mitigation Stateless control munication
overhead
Suh, M. Block Stateless Reactive No Centralized High
et al. [26] unnecessary firewall control performance
mitigation
Pena, J. G. Packet Stateless Proactive Yes Distributed High
et al. [27] filtering firewall control performance
Vasudevan, Prevent Stateless Reactive Yes Distributed -
S. et al. unauthorized firewall control
[28] access
Jeong C. Malicious IDS Proactive Yes Centralized High
et al. [15] traffic control performance
inspection
Proposed Dynamic Stateful Reactive Yes Centralized High
solution packet firewall control performance
filtering
building up the SecPolTable requires only to search the remaining (n−n1)

entries with O(n − n1) processing time. We can observe that each time a
new switch joins the network, the processing time needed to build
SecPolTable is decreasing. Fortunately this process is done only one time for
each switch or, additionally, when the policy file is updated. Now on, each
Packet-in arrives (for example from switch 1) only we need to search the
SecPolTable related to that switch with O(n1) processing time that is
lease than O(n), which in result will increase the overall throughputs.
Regarding the protection level provided by a network security solution, a
simple stateful firewall, which is a good security system to protect the
network still is vulnerable to some attacks like DDoS or DoS. Such attacks
adapt at exploiting the heart of most stateful firewalls STable and try to fall
it to capacity and block new incoming connections. Our suggested prototype
uses mechanisms and implements modes that restrict the timeout for
establishing connection and limits the number of the entries in the STable,
which in turn can prevent or at least mitigate possibilities of such attacks in
the network.
To summarize this draft analysis of security functions and performance let
us look at Table 3. We compare the proposed approach with SDN-dedicated
security solutions known from the literature. It is seen that some of the
systems are only simple packet filtering firewalls while some other
additionally provide dynamic packet checkup. This requires new tables in
the controller side which in turn leads to high communication overhead. To
overcome this limitations of
existing approaches, we make use of a new table in the data plan side that
provides the state-awareness and decreases the switch controller
communication overhead. Another advantage of our approach is
application of a multi-level structure of the firewall with the levels
associated with SDN planes, distributed sensing and modular structure.
Such an approach makes it possible easy exten- sions of the firewall
functions and integration with other security systems. For instance, our
firewall is compatible with the ISO/OSI Layer 2 firewall solution proposed
in [29], which consists of two basic elements: Ethernet frames filtering in
Layer 2 and a three-tier reputation system of critical data acquisition and
analysis for the firewall rules configuration. Such a distributed multi-tier
reputation system modulus could slightly improve threats identification
and enforce SDN protection.
6 Conclusion
In this paper we introduced a reactive stateful firewall approach for securing
SDN-based networks. Our under construction suggested solution does not
enforce any change to the current SDN paradigm in terms of its design and
behavior. Moreover, it attempts to enhance the security of the
programmable network and simplify security management. This proof-of-
concept solution tries to overcome the limitations of the OpenFlow
implementation that relies on a simple “match- action” paradigm. Such an
OpenFlow approach causes lacks of stateful processing function in the
SDN data plane, where it makes use of specified tables in the data plane
devices to keep track of TCP connection or pseudo-connection. The security
policy of SDN proposed in our approach is enforced by the firewall appli-
cation that is centralized on the top of the controller, while the data plane
device is considered as a distributed checkpoint. Thus, instead of one
firewall placed at the edge of the network’s boundary and providing a single
point of failure, we can have a plane-like firewall that creates additional
boundaries within the network and provides “defense-in-depth”, which is
considered as a good solution for networks with different levels of trust. The
behavior of our solution is reactive, which gives an advantage of keeping the
flow tables in the data plane small and which helps not to fill them with
unmatched rules. For the next enhancements, we strive to improve the
expected performance of the proposed prototype. We will focus on
implementing the application with proactive approach to improve the
system’s overall performance by avoiding redundant times caused by com-
munication between the switch and the controller for undefined packets.
Also we plan to deploy the prototype in a real network environment to
confirm its strength against active attackers. As a next step we will
integrate the system with a soft methods-based security tool (a multi-tier
trust and reputation system) what is a reasonable enhancement of a
firewall in widespread and untrusted networks.
SDN
References
1. Adrian, L., Kolasani, A., Ramamurthy, B.: Network innovation using openflow:
a survey. IEEE Commun. Surv. Tutorials 16, 493–512 (2013)
2. Azodolmolky, S.: Software Defined Networking with OpenFlow Get Hands-on
with the Platforms and Development Tools Used to Build OpenFlow Network
Applica- tions. Packt Publishing Ltd., Birmingham (2013)
3. Pujolle, G.: Software Networks Virtualization, SDN, 5G and Security. ISTE Ltd.
and Wiley, Great Britain, United States (2015)
4. Kreutz, D., Ramos, F.M.V., Verissimo, P.E., Rothenberg, C.E., Azodolmolky,
S., Uhlig, S.: Software-defined networking: a comprehensive survey. Proc. IEEE
103, 14–76 (2015)
5. Underdahl, B., Kinghorn, G.: Software Defined Networking for Dummies. Wiley,
Hoboken (2015)
6. Jain, R., Paul, S.: Network virtualization and software defined networking for
cloud computing: a survey. IEEE Commun. Mag. 51, 24–31 (2013)
7. The Open Networking Foundation, OpenFlow Switch Specification (2014)
8. Astuto, B.N., Mendonca, M., Nguyen, X.N., Obraczka, K., Turletti, T.: A survey
of software-defined networking: past, present, and future of programmable
networks. IEEE Commun. Surv. Tutorials 16, 1617–1634 (2014)
9. Sharma, R.K., Kalita, H.K., Issac, B.: Different firewall techniques: a survey. In:
5th IEEE International Conference on Computing. Communications and
Networking Technologies (ICCCNT), Hefei, Anhui, China, pp. 1–6 (2014)
10. Scarfone, K., Hoffman, P.: Guidelines on Firewalls and Firewall Policy,
Gaithers- burg (2009)
11. Duan, Q., Al-Shaer, E.: Traffic-aware dynamic firewall policy management: tech-
niques and applications. IEEE Commun. Mag. 51, 73–79 (2013)
12. Trabelsi, Z.: Teaching stateless and stateful firewall packet filtering: a hands-on
approach. In: 16th Colloquium for Information Systems Security Education,
Lake Buena Vista, Florida, pp. 95–102 (2012)
13. Mehdi, S.A., Khalid, J., Khayam, S.A.: Revisiting traffic anomaly detection
using software defined networking. In: Sommer, R., Balzarotti, D., Maier, G.
(eds.) RAID 2011. LNCS, vol. 6961, pp. 161–180. Springer, Heidelberg (2011).
doi:10.1007/
978-3-642-23644-0 9
14. Braga, R., Mota, E., Passito, A.: Lightweight DDoS flooding attack detection
using NOX/OpenFlow. In: 35th Annual IEEE Conference on Local Computer
Networks, LCN, Denver, Colorado, pp. 408–415 (2010)
15. Jeong, C., Ha, T., Narantuya, J., Lim, H., Kim, J.: Scalable network intrusion
detection on virtual SDN environment. In: 2014 IEEE 3rd International
Conference on Cloud Networking (CloudNet), Luxembourg, pp. 264–265 (2014)
16. Francois, J., Aib, I., Boutaba, R.: Firecol: a collaborative protection network for
the detection of flooding DDoS attacks. IEEE/ACM Trans. Networking (TON)
20, 1828–1841 (2012)
17. Porras, Ph., Shin, S., Yegneswaran, V., Fong, M., Tyson, M., Gu, G.: A security
enforcement kernel for openflow networks. In: HotSDN 2012 ACM Special
Interest Group on Data Communication SIGCOMM, pp. 121–126. ACM,
Helsinki (2012)
18. Shin, S., Porras, P., Yegneswaran, V., Fong, M., Gu, G., Tyson, M.: FRESCO:
Modular composable security services for software-defined networks. In: NDSS
2013 Network and Distributed System Security Symposium, San Diego, CA, pp.
1–16 (2013)
19. Hu, H., Han, W., Ahn, G.J., Zhao, Z.: FLOWGUARD: building robust firewalls
for software-defined networks. In: HotSDN 2014, pp. 97–102. ACM, Chicago
(2014)
20. Shirali-Shahreza, S., Ganjali, Y.: Flexam: flexible sampling extension for
monitoring and security applications in OpenFlow. In: HotSDN 2013, Hong
Kong, China,
pp. 167–168. ACM, New York (2013)
21. Shirali-Shahreza, S., Ganjali, Y.: Efficient implementation of security
applications in openflow controller with flexam. In: IEEE 21st Annual
Symposium on High- Performance Interconnects, San Jose, CA, USA, pp. 49–54
(2013)
22. Collings, J., Liu, J.: An openflow-based prototype of SDN-oriented stateful hard-
ware firewalls. In: IEEE 22nd International Conference on Network Protocols,
Raleigh, NC, USA, pp. 525–528 (2014)
23. Zerkane, S., Espes, D., Le Parc, P., Cuppens, F.: Software defined network-
ing reactive stateful firewall. In: Hoepman, J.-H., Katzenbeisser, S. (eds.) SEC
2016. IFIP AICT, vol. 471, pp. 119–132. Springer, Cham (2016). doi:10.1007/
978-3-319-33630-5 9
24. Lar, S., Liao, X., ur Rehman, A., Ma, Q.: Proactive security mechanism and design.
J. Inf. Secur. 2, 122–130 (2011)
25. Cabaj, K., Mazurczyk, W.: Using software-defined networking for ransomware
mitigation: the case of cryptowall. IEEE Netw. Mag. Global Internetworking 30,
14–20 (2016)
26. Suh, M., Park, S.H., Lee, B., Yang, S.: Building firewall over the software-
defined network controller. In: The 16th IEEE International Conference on
Advanced Com- munications Technology, Daejeon, South Korea, pp. 744–748
(2014)
27. Pena, J.G., Yu, W.E.: Development of a distributed firewall using software
defined networking technology. In: 4th IEEE International Conference on
Information Sci- ence and Technology, Shenzhen, China, pp. 449–452 (2014)
28. Vasudevan, S.: Firewall a new approach to solve issues in software define net-
working. In: 6th International Conference on Emerging Trends in Engineering
and Technology, pp. 14–19 (2016)
29. Konorski, J., Pacyna, P., Kolaczek, G., Kotulski, Z., Cabaj, K., Szalachowski,
P.: A virtualization-level future internet defense-in-depth architecture. In:
Thampi, S.M., Zomaya, A.Y., Strufe, T., Alcaraz Calero, J.M., Thomas, T.
(eds.) SNDS 2012. CCIS, vol. 335, pp. 283–292. Springer, Heidelberg (2012).
doi:10.1007/
978-3-642-34135-9 29

Method of The Management of Garbage Coll-201-300

Uploaded by

Copyright:

Available Formats

Method of The Management of Garbage Coll-201-300

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Method of The Management of Garbage Coll-201-300

Uploaded by

Copyright:

Available Formats

Utilization of Redundant Communication Network 181

4.1 General Idea

Fig. 1. Communication buses in multiple network interface node MuNetIN

In systems with redundancy, communication with every Slave would be

4.2 Triple Network Interface Communication System

4.3 Query Scheduler

Fig. 2. Query distribution

5 Formal Description of Distribution Algorithm

be stored in vector Q = q1 · · · qr , where:

Hence, Distributor for given i should choose such j, that:

tj = min {tj : bti,j = 1 ∧ qj = 0}.

5.2 Aperiodic Transactions

6 Algorithm Realization Example

P = 50 50 50 100 100 150 150 300

Fig. 3. Deadlines and queries realization

In the second cycle cf equals 34.5 which is still less than S − m.

7 Conclusions and Future Work

Slawomir Przylucki1(✉), Artur Sierszen2, and Dariusz Czerwinski1

Abstract. With the spread of Dynamic Adaptive Streaming over

Keywords: Home network · Software Defined Network (SDN) · Scal-

2 Background and Related Works

contained in these files, can switch between different representations

– Multiple DASH clients are connected to an external network (typically,

– Network resources module (NRM).

First two modules are associated with a few fundamental functionalities.

Traffic shapping module (TSM)

Fig. 1. The architecture of the proposed SDHN solution.

3.1 Network Resources Module

3.2 QoS Controller

Fairness Distribution Block. The algorithm implemented in the FDB is

Yoptimal = max[min(UFi(xi))] Σ(xi) ≤ Blimit (1)

4 Test Scenario, Evaluation and Results

4.1 The Testbed Configuration

client DASH client DASH

720p 720p NRM container

Fig. 2. The structure of the testbed.

– three temporal layers (T0 - 6 f/s, T1 12 f/s and T2 24 f/s),

4.2 Evaluation Metrics

Fairness index. Publications of recent years bring many proposal of metrics

to respective demands. As the result, we propose the following definition of the

Fluctuation index. Based on our experience [14], the bitrate fluctuations

Link Utilization Index. It is obvious that the limited capacity of outgoing

Here, Blimit is the bandwidth of the bottleneck and Bn is the summary

4.3 Results of SDHN Evaluation

In the most of tested cases, the legacy DASH was sensitive to

calculate near-optimal bitrate values what in combination with playout

Table 1. The mean values of the FlucI and LUI indexes.

Scenario – system Indexes

In all the tests conducted, the proposed SDHN solution demonstrated

Fig. 5. The fairness Index for the test B.

For the standard DASH distribution system (legacy DASH), the

5 Conclusion and Future Work

10. ISO/IEC FDIS 23009-5:2016: Information Technology - Dynamic adaptive

Bartl-omiej Pl-aczek(✉) and Marcin Bernas

Institute of Computer Science, University of Silesia,

Abstract. In this paper the connectivity is analysed of vehicular ad

Keywords: Vehicular ad hoc networks · Connectivity · Minimum

2 Related Works and Contribution

3 A Method for Minimum Transmission Range