GPON Service Level Agreement based

Dynamic Bandwidth Assignment Protocol

Ching-Hung Chang, P. Kourtessis and J. M. Senior

A dynamic bandwidth assignment protocol is proposed that demonstrates

impartial and highly efficient bandwidth arrangement for gigabit-capable

passive optical networks. In particular the novel algorithm automatically

modifies the assigned bandwidth per optical network unit to support quality of

service and service level agreement according to maximum network capacity

and users’ queuing status. Network simulation results have demonstrated

superior mean packet delay performance achieving a tenfold reduction of

packet delay at high network load when compared to other dynamic

bandwidth assignment algorithms.

Introduction: Time division multiplexing-passive optical networks have been

adopted to provide a cost-effective solution for the implementation of point to

multipoint (P2MP) future broadband access networks [1]. gigabit-capable

passive optical networks (GPONs) in particular support the development of

full service access networks (FSANs) by providing multiple network rates

downstream and upstream, and guaranteed quality of service (QoS) over

distances of up to 60km [2].

According to the standard [2], the optical line terminal (OLT) assigns available

time slots to each optical network unit (ONU) for transmission, by means of
grant packets, according to the report packets send by each ONU to notify the

OLT about their buffer queuing status. To avoid the accumulation of “waiting

time” associated with the report and grant packets polling times, an

interleaved polling scheme with adaptive cycle time (IPACT) [3] has been

proposed for ethernet-PONs (EPONs) to overlap multiple polling requests in

time. Although intended for EPON applications the interleaved polling scheme

can be directly applied to allocate bandwidth in GPON architectures, being

also of tree topology, by readjusting the time to transmit the grant packets

from the OLT to each ONU and redesigning the frame format.

Consequently IPACT can utilize bandwidth more efficiently than constant

time-slot PONs [3], and therefore it has been used to develop other EPON

dynamic bandwidth assignment (DBA) algorithms, some examples of which

are the dynamic credit distribution [4], guaranteed minimum bandwidth [5],

two-layer bandwidth allocation [6] and Intra-ONU bandwidth allocation

schemes [7]. What all these approaches have in common is that ONUs are

assigned to a single service level and same aggregate data rates. However,

the GPON standard [2] defines multiple data rates and the internet service

providers (ISPs) in practice, assign multi-service levels according to

customers’ requirements.

This paper proposes a novel DBA scheme for GPON which achieves QoS at

three different types of service levels and diverse network throughputs.

Initially it dynamically assigns to each ONU a guaranteed minimum bandwidth

from the overall network capacity to satisfy their basic service requirements.
Consequently, the OLT will apportion any unused bandwidth to ONUs

according to their buffer queuing status. Therefore, following probable

variations in network capacity, the OLT will be capable of readjusting the

guaranteed minimum and unused bandwidths among ONUs to comply with

subscriber contracts.

Proposed algorithm: The OLT is designed to provide ONUs with three service

levels, denoted by t, at different weights Wt to represent the priority of

accessing the network. As a result service level one for example will acquire

the lowest weight which will reflect the amount of time slots it will occupy in

one polling cycle. To simulate practical network conditions, these weights are

selected in order to comply with the NTT VDSL service plans corresponding

to 50, 70 and 100Mbit/s for service levels with weights 2, 3, and 4 respectively

[8]. The transmission time-slots for each user are assigned in two stages. In

the first stage, a guaranteed minimum bandwidth, Btmin, is assigned to each

ONU according to its service level. In particular to provide diverse service

levels and being able to adapt to variable data rates, Btmin should be defined

as the supplement of two parameters; a basic bandwidth, Bbasic, and an extra

guaranteed bandwidth, Btex. Henceforth, the guaranteed minimum bandwidth

for service level t will be given by:

t
Bmin = Bbasic + Bext (1)

In eq.1, the basic bandwidth, Bbasic, describes a constant bandwidth available

to all ONUs independently of service level. In order for the OLT to dynamically

assign more bandwidth to higher service level ONUs at variable data rates,
the extra bandwidth parameter, Btex, is included in the calculation of Btmin. For

service level t, the extra bandwidth, Btex, is assigned by considering the

maximum network capacity Btotal and is therefore given by:

Wt
Bext = ( Btotal − k × Bbasic ) 3
(2)
∑W N
t =1
t t

where k is the number of ONUs comprising the network and Nt is the number

of ONUs subscribed to service level t.

In the probability that in each polling cycle only a fraction of ONUs use their

entire guaranteed minimum bandwidth, the algorithm in the second stage will

proportionally assign the unused bandwidth to the ONUs whose requirement

exceed Btmin, as calculated before. Therefore for ONUi the extra assigned

bandwidth, Biex_assigned, is given by the ratio of its extra required bandwidth,

Biex_required, determined for each user according to its required bandwidth and

relative Btmin, over the total extra required bandwidth multiplied by the unused

bandwidth following:

B i ex _ requried
Bexi _ assigned = ∑ Bunused × (3)
∑ Bex _ required

The unused bandwidth for ONUi from stage one can be calculated by

subtracting the corresponding queuing length Qi, representing the bandwidth

requirement, from the total minimum guaranteed bandwidth.

In summary, the maximum allocated bandwidth Bmax_allocated for ONUi will be

equal to the summation of Btmin from stage one and Bex_assigned from stage two.
Otherwise if ONUi’s request for bandwidth is smaller than the total,

Bmax_allocated will be equal to Qi as given below:

 Bmin
t
+ Bexi _ assigned 
i
B max_ allocated = min   (4)
 Qi 
 

The new algorithm which is designed to grant network users with diverse

service levels improved performance is called the dynamic minimum

bandwidth (DMB) scheme.

Network modeling and simulation results: To evaluate the performance of the

proposed scheme, a GPON network was devised in OPNET modeler

comprising a single OLT and 16 ONUs with varying weights (1*W3 + 5*W2 +

10*W1) to represent different service levels. Since the fundamental

differentiation in implementing EPON and GPON DBA algorithms is in the

frame format, a simple GPON packet header was modeled to simulate the

upstream transmission of 1Gbit/s for link lengths of up to 25km [2]. In addition,

a 2ms maximum cycle period and 1µs guard time were chosen to establish

data transfer between the OLT and ONUs in order to offer a platform for

comparison with published algorithms [4-7]. Network traffic was generated

based on a self-similar model with typical Hurst parameter of 0.8, maximum

ONU channel capacity of 100 Mbit/s and basic bandwidth of 34 Mbit/s to

effectively simulate the requirement in bandwidth of representative network

services.
The simulated channel throughput versus network load characteristics for

IPACT and the proposed DMB scheme are shown in Fig. 1. To be able to

provide a valid evaluation of the proposed algorithm’s performance against

IPACT, taking into consideration that the latter was intended for EPON

applications, preliminary work investigated the successful adaptation of

IPACT to the GPON model. It may be observed that IPACT can be equally

applied to EPON and GPON architectures establishing its validity as a

reference for assessing the proposed scheme. In addition, the two algorithms

exhibit similar performance when modeled as GPON as far as the achieved

network throughput is concerned indicating identical bandwidth utilisation.

Fig. 2, however, which displays the simulated mean packet delay against

network load demonstrates a diversion in performance between the two

algorithms. In this case the DMB exhibits around a factor of 10 less packet

delay in comparison with IPACT under high network load, suggesting that the

proposed algorithm is much more efficient in utilizing the limited network

capacity. This is due to the fact that when the maximum network capacity has

been reached, the ONUs with higher requirement in bandwidth are allocated

most of the available network bandwidth in priority.

Finally, Fig. 3 shows the simulated network packet delay response for the

assigned service levels. It becomes apparent from the plot that when each

ONU’s offered load is greater than half its maximum capacity, total user

requirement in bandwidth exceed the network capacity. Therefore for

increasing values of network load, the OLT will proportionally allocate most of
the unused bandwidth among the higher service level ONUs, resulting to less

packet delay for extended load figures, given by 0.54, 0.61 and 0.74 for

service levels 1, 2, and 3 respectively.

Conclusion: This paper has demonstrated a novel DBA algorithm exhibiting

impartial and highly efficient bandwidth arrangement for TDM-GPONs. In

particular the maximum bandwidth allocated per ONU is dynamically assigned,

based on network capacity, subscriber service levels and queuing status

aiming to assign more bandwidth to higher service level subscribers and

consequently to reduce their packet delay compared to their low service

counterparts. Contrasting the proposed scheme with alternative reported

EPON algorithms, simulation results have demonstrated reduced mean

packet delay accomplishing up to a tenfold decrease at high network load. In

addition, the demonstrated network performance in packet delay of high

service level users is sustained for increased traffic demonstrating network

integrity and QoS according to subscriber service levels.

References

1 X.-Z. Qiu, P. Ossieur, J. Bauwelinck, Y. Yi, D. Verhulst, J. Vandewege,

B. DeVos, and P. Solina, "Development of GPON Upstream Physical-
Media-Dependent Prototypes," Journal of Lightwave Technology, vol.
22, pp. 2498-2508, 2004.

2 ITU-T Recommendation. G.984.2, "Gigabit-capable passive optical

networks (GPON): Physcal media dependent (PMD) layer
specification," 2003.

3 G. Kramer, B. Mukherjee, and G. Pesavento, "IPACT a dynamic

protocol for an Ethernet PON (EPON)," Communications Magazine,
IEEE, vol. 40, pp. 74-80, Feb 2002.
4 H. Miyoshi, T. Inoue, and K. Yamashita, "QoS-aware dynamic
bandwidth allocation scheme in Gigabit-Ethernet passive optical
networks," IEEE International Conference on Communications, vol. 1,
pp. 90-94, 2004.

5 D. Nowak, P. Perry, and J. Murphy, "Bandwidth Scheduling

Techniques for Differentiated Services Support in Ethernet Passive
Optical Networks," Proc. of 9th International Workshop on Component
Oriented Programming (WCOP), in conjunction with 18th European
Conference on Object-Oriented Programming (ECOOP), June 2004.

6 J. Xie, S. Jiang, and Y. Jiang, "A dynamic bandwidth allocation scheme

for differentiated services in EPONs," Communications Magazine,
IEEE, vol. 42, pp. S32-S39, Aug 2004.

7 N. Ghani, A. Shami, C. Assi, and M. Y. A. Raja, "Intra-ONU bandwidth

scheduling in Ethernet passive optical networks," Communications
Letters, IEEE, vol. 8, pp. 683-685, 2004.

8 NTT, "NTT VDSL service plan,"

http://www.asist.co.jp/jensspinnet/bflets.html, Feb 2006.

Authors’ affiliations:
Optical Networks Group, Science and Technology Research Institute (STRI),
University of Hertfordshire, College Lane Campus, Hatfield, Herts AL10 9AB,
UK.

E-mail addresses:
[email protected]
[email protected]

Figure captions:

Fig. 1 Channel throughput for IPACT and Dynamic Minimum Bandwidth in

GPON

Fig. 2 Mean packet delay for IPACT and Dynamic Minimum Bandwidth in
GPON

Fig. 3 Packet delay for three service levels in Dynamic Minimum Bandwidth in
GPON

Figure 1
 1000
900
800
Throughput (Mbit/s)

700
600
500
400
300 IPACT in GPON
200 DMB in GPON
100 IPACT in EPON
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
Netowrk offered load

Figure 2

10

0.1
Delay (s)

0.01

0.001 IPACT
DMB
0.0001
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
Network offered load

Figure 3

0.1
Delay (s)

0.01

DMB_S3
0.001
DMB_S2
DMB_S1
DMB_mean
0.0001
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1
ONU offered load