In the above combo scenarios, couple of competing and concurrent TCP
flows/connections set up. Flow1 runs between Sender1 and Receiver1 and the Flow2 runs
between Sender2 and Receiver2 as depicted in the above Topology diagram (Figure 2-1).
Sender2 has been setup with constant 100ms delay along with invariable TCP SACK and
Receiver2 is also setup with fixed TCP SACK in all the flow2 scenarios. On the other
hand, Sender1 and Receiver1 have been setup with different TCP variants as well as
different Delays in each scenario, thus the RTT varies in Flow1 from 20ms, 100ms,
200ms to 400ms and TCP variants change to CUBIC, HIGHSPEED, and WESTWOOD.
As can be seen from the above table-1, unlike Baseline scenarios, in Combo scenarios the
performance of the TCP variants have been impacted by the active multiple TCP flows
and the performance degraded by increasing the end-to-end delay. For moderate delay of
20ms and 100ms all the TCP variants in Flow1 yield a throughput of 46.79 Mbps on
average whereas TCP variants in Flow2 yield around 10.36 Mbps on average. Thus
CUBIC, HIGH SPEED, WESTWOOD take over greater bandwidth capacity as compare
to the Flow2 TCP SACK.
For flows with larger delays (long distances), such as 200ms, the tremendous change can
be seen from Flow2 with TCP SACK which captured greater part of the bandwidth,
however, in Flow1, HIGHSPEED attained a lower throughput and WESTWOOD
attained the lowest of only 9.4Mbps.
The performance of CUBIC, HIGHSPEED and WESTWOOD suffered from the
implementation of greater delay of 400ms as flow2 with TCP SACK captured a greater
part of the bandwidth about 59.74 on average, however, in Flow1, CUBIC yielded
22.97Mbps, HIGHSPEED attained lower throughput of 17.78Mbps and WESTWOOD
attained the lowest of merely 4.42Mbps.
In the above scenarios, the inconsistent average round trip time (RTT) did not provide a
basis for deciding a conclusion, however, fair treatment of different TCP flows should
hold regardless of differences of round trip time.