Traffic Management for the Available Bit Rate (ABR) Service in ...

Number of Throughput
Sources TBE Bu er T1 T2 T3 T4 T5 Total % CLR
2A+V 128 256 3.1 3.1 6.2 10.6 1.2
2A+V 128 1024 10.5 4.1 14.6 24.9 2.0
2A+V 512 1024 5.7 5.9 11.6 19.8 2.7
2A+V 512 2048 8.0 8.0 16.0 27.4 1.0
5A+V 128 640 1.5 1.4 3.0 1.6 1.6 9.1 15.6 4.8
5A+V 128 1280 2.7 2.4 2.6 2.5 2.6 12.8 21.8 1.0
5A+V 512 2560 4.0 4.0 4.0 3.9 4.1 19.9 34.1 0.3
5A+V 512 5720 11.7 11.8 11.6 11.8 11.6 58.4 100.0 0.0
Table 8.1: Simulation Results: Summary
number of cells are lost but there is only one timeout, the throughput degrada-
tion may not be that severe. On the other hand, even if a few cells are lost but
the losses happen far apart triggering multiple timeouts, the throughput will
be severely degraded. Hence, the cell level metric CLR is not a good indicator
of the TCP level performance.
2. E ect of Bu ering: Larger bu ers always give higher TCP throughput for our
in nite TCP applications.
We study the e ect of bu ers on latency in section 8.14. Brie y, since the ABR
control can drain out the queues after the initial transient, the average latency
should be low. The e ect of new TCP sources starting up does not increase the
latency signi cantly since they start at a window of one MSS, irrespective of
their ICRs.
275