EE555: "Broadband Networks Architecture"

Professor A. Zahid, Summer Semester 2003

HW#2: Solution

1. Chapter 8, Stallings Book, Problems 8.10, 8.11 and 8.17

Answer:

8.10 Ts = (1)(0.7) + (3)(0.2) + 10(0.1) = 2.3 ms

Define S as the random variable that represents service time. Then

[ ]
σ T2 s = Var [S] = E (S − T s ) = (1 − 2.3)2 (0.7) + (3 − 2.3 )2 (0.2) + (10 − 2.3 )2 (0.1) = 7.21 ms
2

Using the equations in Figure 8.6a:

A = 0.5 (1 + (7.21/5.29)) = 1.18

a. ρ = λTs = 0.33 × 2.3 = 0.767

Tr = Ts + (ρTsA)/(1 – ρ) = 2.3 + (0.767 × 2.3 × 1.18)/(0.233) = 11.23 ms

r = ρ + (ρ2A)/(1 – ρ) = 3.73 messages

b. ρ = λTs = 0.25 × 2.3 = 0.575

Tr = 2.3 + (0.575 × 2.3 × 1.18)/(0.425) = 5.97 ms

r = 0.575 + (0.575 × 0.575 × 1.18)/(0.425) = 1.49 messages

c. ρ = λTs = 0.2 × 2.3 = 0.46

Tr = 2.3 + (0.46 × 2.3 × 1.18)/(0.54) = 4.61 ms

r = 0.46 + (0.46 × 0.46 × 1.18)/(0.54) = 0.92 messages

8.11 λ = 0.05 msg/sec; Ts = (14,400 × 8)/9600 = 12 sec; ρ = 0.6

a. Tw = ρTs/(1 – ρ) = 18 sec

b. w = ρ2/(1 – p) = 0.9

8.17 a. System throughput: Λ = λ + PΛ; therefore Λ = λ/(1 – P)

Server utilization: ρ = ΛTs = λTs/(1 – P)
T (1− P )T s
Tq = s =
1 − ρ 1 − P − λT s
 b. A single customer may cycle around multiple times before leaving the system.
The mean number of passes J is given by:
∞ 1
J = 1(1− P ) + 2(1− P )P + 3(1 − P )P +K = (1 − P ) ∑ iP =
2 i−1
i =1 1−P

Because the number of passes is independent of the queuing time, the total time
in the system is given by the product of the two means, JTr.

2. Suppose a company has a router with 10 interfaces (each at 10 Mbps) to its ISP. Outgoing
packets arrive to the router according to a Poisson distribution with mean value, λ= 50,000
packets/sec. The packet lengths are exponentially distributed with mean 1000 bits. When the
outgoing packets arrive at the router, they are "randomly" dispatched to one of the 10 interfaces.
Suppose the router is capable of handling 100,000 packets/s without any packet loss (i.e. it has
high throughput and large buffers). Total packet delay is the sum of the waiting time and the
service time
• Calculate the mean packet delay (Hint: This is an M/M/N Queuing model, where N is the
number of servers that is 10 in this case)
• Repeat if the company decides to replace the ten interfaces with a single high-speed
interface at 100 Mbps (Hint : This is an M/M/1 Queuing model)
Answer:

a.
M/M/N Queuing Model
Packet delay = queuing delay = Tq
N = number of servers = 10
Ts = time to service for one packet = 1000/10Mbps = 100 ms
p = l.Ts, l= mean arrival rate into each queue = 50000/10 = 5000 packets/sec per interface
p = 5000x100m = 0.5
Tq = C.Ts/(N.(1-p))+Ts, where C is the Erlang C function = probability that all servers are busy.
C = (1-K)/(1-pK), where K is the Poisson ratio function
K = K1/K2
N-1 N
Where K1 = S (Np)I / I! , K2 = S (Np)I / I!
I=0 I=0
For N = 10, p =0.5, we got
Tq = 101 us

I I! Np (Np)^I (Np)^I / I!
0 1 5 1 1
1 1 5 5 5
2 2 5 25 12.5
3 6 5 125 20.83333
4 24 5 625 26.04167
5 120 5 3125 26.04167
6 720 5 15625 21.70139
7 5040 5 78125 15.50099
8 40320 5 390625 9.68812
9 362880 5 1953125 5.382289
10 3628800 5 9765625 2.691144
K1 143.6895
K2 146.3806
K 0.981615
C = (1-K)/(1-pK) = 0.036105
Ts = 0.0001
Tq = C.Ts/(N.(1-p))+Ts 0.000101

b.
One fast interface
M = 100 Mbps
Ts = 1000/100Mbps = 10 ms
p = l.Ts, l= mean arrival rate into each queue = 50000 packets/sec per interface
p = 50000x10m = 0.5
Tq = Ts / (1-p) = 20 us

3. What kind of queuing model may be suitable to model an ATM switch? What kind of queuing
model is suitable to model an IP router? Justify your answer.
Answer:

ATM uses fixed size cells, and so the service time is deterministic. If we assume the arrival
process to be random (or Poisson) then we can use M/D/1 model. (more generally, we can use
G/D/N for a general arrival process and N servers).

IP packets vary in size, and depending on their length distributions they may incur different
service times. If we assume a exponential distribution of packet sizes then we can use the M/M/1
model (or more generally, G/M/N).

4. William Stalling's Book: #5.4, #5.5, #10.2

Answer:

5.4
a.
We reason as follows: A total of X octets are to be transmitted. This will require a total of
 X / L cells. Each cell consists of (L+H) octets. Thus,
X
N=
 X / L( L + H )
The efficiency is optimal for all values of X which are integer multiples of the cell information
size. In the optimal case, the efficiency becomes
L
N _ opt =
L+H
For the case of ATM, with L = 48 and H = 5, we have N_opt = .91

b.
Assume that the entire X octets to be transmitted can fit into a single variable-length cell. Then,
X
N=
X + H + Hv
5.5
a. The transmission time for one cell through one switch is
t = (53 × 8)/(43 × 106) = 9.86µs.
b. The maximum time from when a typical video cell arrives at the first switch (and possibly
waits) until it is finished being transmitted by the 5th and last one is
2 × 5 × 9.86µs = 98.6µs.
c. The average time from the input of the first switch to clearing the fifth is
(5 + 0.6 × 5 × 0.5) × 9.86µs = 64.09µs.
d. The transmission time is always incurred so the jitter is due only to the waiting for switches to
clear. In the first case the maximum jitter is 49.3µs. In the second case the average jitter is
64.09 – 49.3 = 14.79µs.
10.2 Throughput is as follows:

0≤λ≤1 1 ≤ λ ≤ 10 10 ≤ λ
A-A' traffic 0.8 (1.8 × 0.8)/(0.8 + λ) 1.44/10.8 = 0.13
B-B' traffic λ 1.8λ/(0.8 + λ) 18/10.8 = 1.67
Total throughput 0.8 + λ 1.8 1.8

The plot:
2.0

1.5 Total Throughput

B-B'
Throughput

1.0

A-A'
0.5

0.0
0 1 2 3 4 5 6 7 8 9 10
λ
5. A user is connected to a FR network through a T-1 line. The Committed Information
Rate (CIR) is 1 Mbps with a Committed Burst Size, Bc, of 5 Mbits each 5 seconds.
The Excess Burst Size, Be, is 1 Mbits every 5 seconds. Answer the following questions:
a) What is the access rate?
b) What is the maximum data rate the user can use all of the time without worrying about frames
being discarded?
c) If the user wants to take a risk, what is the maximum data rate the user can be used with no
chance of frames being discarded if there is no congestion?
d) Can the user send data at a constant bit rate of 1.4 Mbps all the time?
e) The user sends data at a rate of 1.4 Mbps for 2 seconds and nothing for the next three seconds.
Is there a danger of discarding if there is no congestion? Is there a danger of discarding if there
is congestion? Explain
Answer:
a) 1.544 Mbps
b) 1 Mbps (CIR)
c) 1.2 Mbps = (Bc+Be)/Tc
d) No. Some frames will be discarded since the rate exceeds CIR
e) There is no risk of discarding frames (whether there is congestion or not) since the average
data rate is 0.56 Mbps which is less than the CIR

6. The purpose of HEC field is to protect against errors in the header of ATM cells.
Assume bit errors occur at random and that the HEC can correct all single errors and can detect
all double errors only. Assume that the bit error rate p = 10-6
• What is the probability that the cell will be delivered to the right destination?
• What is the probability that the cell is missinserted?
• What is the probability that the cell is dropped?
Answer:

Assume Bit Error Rate = p, Header length = L = 40 bits, Number of errors in Header = X.
Probability (cell is delivered to right destination) = P (X = 0) + P (X = 1) = (1-p)40 + 40p(1-p)39
Probability (cell is misinserted) = P (X ≥ 3) = 1 - ((1-p)40 + 40p(1-p)39 +780p2(1-p)38)
Probability (cell is dropped) = P (X = 2) = 780p2(1-p)38

7. The IP datagram for a TCP ACK message is 40 bytes long (20 bytes of TCP header, 20 bytes
of IP header). Assume this ACK is traversing an ATM network that uses AAL5 to encapsulate IP
packets. How many ATM cells are needed to carry the TCP ACK?
Answer:

The length of the AAL/5 CS-PDU into which the ACK is encapsulated is 48 bytes (no padding is
needed, just 8 bytes of trailer) and this fits in a single ATM cell.

8. In your opinion, what are the main problems in deploying ATM in the Internet?
Suppose you have a connection between hosts "A" and host "B". IP packets pass through an IP
network, then through an ATM network and then back through an IP network. What happens
during link failures (consider a link failure inside an IP network, a failure in the link connecting
the IP network to the ATM network and finally a link failure inside the core ATM network) core
ATM network
Answer:

(a)
The main problems is that ATM is connection-oriented (meaning that it uses virtual circuit with
locally significant labels, needs a setup phase, admission control, where packets follow same
path). Whereas IP (which is likely to cross ATM clouds), is connection-less datagram service
(meaning that it uses globally unique addresses to make route decisions, does not perform setup
or admission control, and packets may follow different paths).
Hence, there are problems in mapping the IP addresses into ATM addresses (through the address
resolution protocol ARP).
Also, there are problems in establishing the paths in the ATM cloud. Packets arrive at the ATM
cloud and need to be buffered before the connection is established. Also, packets may arrive at
different entry points to the ATM clouds, which may require establishing multiple virtual paths
for the same IP flow.
Furthermore, there is no tear-down message in IP flows in general (for TCP we may use the FIN
flag for example). But in general we need to rely on timer mechanisms to tear down the ATM
connections.
In addition, QoS at the IP level (which is expressed using RSVP for example) needs to be mapped
into ATM QoS, type of service and GFC fields.
Finally, routing in ATM must inter-operate with routing in the Internet (they use different routing
algorithms that rely on potentially different metrics to make routing decisions. ATM runs its own
routing algorithm although it is considered a data-link protocol).

(b)
If the failure occurs in the Internet (outside of the ATM cloud), then the dynamic Internet routing
will re-route the packets around the failure. This occurs transparently (i.e., without the end points
being notified). If the route change leads to a change of path such that the entry point to the ATM
cloud is changed, then a new path must be setup in the ATM cloud.

If the failure occurs in the ATM cloud, then the path must be torn down and re-established, which
may lead to a lot of loss end-to-end.

9. Voice and Video signals are to be carried over IPv6. What is the efficiency of an IPv6 packet
that carries 10 ms of 64 Kbps voice? What is the efficiency of an IPv6 packet that carries 1 frame
of 4 Mbps MPEG2 video, assuming 30 frames/s?
Answer:

a) 10msec of voice at 64 Kbps = 640 bits = 80 Bytes

IPv.6 header = 40 Bytes
Efficiency = 80/120 = 2/3 = 66.7%
b) one frame of video = 4 * 106 / 30 = 133333 bits = 16666 Bytes
Efficiency = 16666 / ( 16666+40) = 99.8%