Download as pdf or txt
Download as pdf or txt
You are on page 1of 3

SONGSRIBOONSIT Norranat Student ID: C3IM4505

Lecture 06 Report: Tahoe

In this lecture, the mechanism of the Tahoe which is a window control protocol in TCP has
been discussed. The experiment was also conducted using NS2 program and the result is explained
as follows.

Figure 1. The topology of simulated network

The network setup is similar to the previous experiments. However, in this experiment, there
are two variables to be adjusted which are bandwidth (BW) and delay (d) between node 2 and 3
according to figure 1. The different levels of these two parameters are related to the changing of
window size during the transmission. This report discusses the effect of bandwidth first, then the
influence of delay is explained after that.

a.) b.)

c.) d.)
Figure 2. the results with the bandwidth of a.) 1 Mbps b.) 2 Mbps c.) 4 Mbps d.) 5 Mbps and delay of 20 ms
SONGSRIBOONSIT Norranat Student ID: C3IM4505

The first parameter of interest is bandwidth. In figure 2.a, the bandwidth is adjusted to 1 Mbps
which is considered very low. At the beginning, the slow start threshold (ssthresh) is initially set to 64
based on the buffer size of the receiver (maximum window size). Then, every node starts their
transmission in the slow start phase and the congestion window (cwnd) is increased exponentially
until it exceeds the ssthresh. Right after that, the transmission enters the congestion avoidance phase
and can detect the packet drops which triggers the cwnd reset algorithm. The algorithm resets the
cwnd to 1 and reduces ssthresh by half. Then, the second slow start phase starts. However,
immediately after it surpasses the sshthresh at 32, the system still detects the packet loss which
triggers the cwnd reset algorithm again. After that this process which contains slow start phase and
congestion avoidance repeats. It may be noticed that the ssthresh sequence is 64, 32, 16, and steady
at 19. This is because the bandwidth in figure 2.a is intentionally adjusted to be very low which causes
the packet loss when packets exponentially flow in the system. Nevertheless, in the third loop, the
exponential increase occurs until only 16 packets, and the system can handle this amount of packets.
Therefore, the congestion avoidance phase starts and the cwnd increases linearly until around the
size of 40 where loss happens and the ssthresh becomes 19. In the following loops, the system can
support all packets in the exponential phase, and the loss occurs only in the congestion avoidance
phase. Because the system environment is stable, the following loops can be observed to have the
same shape.
In figure 2.b, the bandwidth was increased to 2 Mbps, and this change increases the channel
capacity and reduces the packet drops. Similar to the bandwidth of 1 Mbps, 2 Mbps also cannot
support the first exponential rise of cwnd to 64 Mbps, and the ssthresh is reduced to half after that.
However, different from 1 Mbps bandwidth, there is no packet loss right after entering the congestion
avoidance phase in the second loop. The cwnd continues linearly increasing for a while until the
packet loss happens. Then the process repeats and cwnd is reset to 1 every time it linearly increases
to the size of around 47.
Figure 2.c shows very similar results to figure 2.b, except that the characteristics of cwnd
resetting occur when the cwnd linearly increases and touches the level of around 62. This is because
the bandwidth has been increased to 4 Mbps and higher bandwidth can support more packets flowing
in the system.
Finally figure 2.d illustrates the event that the cwnd continues its rising and the transmission
occurs only once within the simulation of 20 seconds. The reason is that the bandwidth is further
increased to 5 Mbps. It is worth noting that even though the increasing congestion avoidance phase is
called linear. The result from the simulation does not show the linear line of increasing. This is
because as the cwnd increases, it takes more time for all packets to come back. Therefore, it
becomes slower for cwnd to increase at the higher cwnd level.
SONGSRIBOONSIT Norranat Student ID: C3IM4505

a.) b.)

c.) d.)
Figure 3. the results with the bandwidth of 4 Mbps and delay of a.) 1 ms b.) 10 ms c.) 20 ms d.) 30 ms

In this part, the effect of different levels of delay is explained. Figure 3. a, b, c and d are
results obtained by varying the delay at 1, 10, 20 and 30 ms respectively. The first notice is that as the
delay increases, it takes longer for the system to reach the optimal sending rate both in slow start
phase and congestion avoidance phase. The second point is that the level of cwnd that packet loss
occurs is higher as the delay is higher. To explain this behavior, normally, the packet losses happen
when there is congestion of packets in the networking system. Reducing the delay means reducing
the round trip time. Then, the sender receives the ACKs message faster which makes the sender
sending the next packets faster. As a result, the reducing delay can accelerate the speed of the
system while the system has the same capacity. Because of this reason, the packet drops occurs
faster when the delay is reduced. Finally, if the delay is larger, the picture 3. d shows that there is a
packet drop only once during the simulation time similar to picture 2. d. However, the difference is that
it takes longer in figure 3. d to reach the optimal value in the first loop because of the higher delay.

You might also like