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

issues and e ect of bursty applications like World Wide Web running on TCP is
examined.
8.4 TCP Performance With Cell Loss
Cell loss will occur in the network if the ATM switches do not have su cient
bu ers to accomodate this queue buildup. In this section, we will show simulations
to demonstrate the problem of cell loss on TCP performance and identify the factors
which a ect the performance under such conditions. Speci cally, we show that TCP
achieves peak throughput over ABR without the necessity of very large bu ers (we
quantify this requirement in section 8.14). Then we limit the bu er size based on
the Transient Bu er Exposure (TBE) ABR SES parameter and the number of TCP
sources. Though the TBE parameter was initially intended to allow some control over
bu er allocation, we nd that it is ine ective in preventing cell loss. We then study
the e ect of cell loss on TCP level packet throughput, and the various parameters
a ecting the performance (bu er size, number of sources, TCP timer granularity
parameter, cell drop policy etc).
Speci cally, when cell loss does occur, the cell loss ratio (CLR) metric, which
quanti es cell loss, is a poor indicator of loss in TCP throughput. This is because
TCP loses time (through timeouts) rather than cells (cell loss). Smaller TCP timer
granularity (which controls timeout durations) can help improve throughput. Due
to fragmentation, a single cell loss results in a packet loss. This further obscures
the meaning of the CLR metric. If the ABR rates do not converge to optimum
values before the cell loss occurs, the e ect of the switch congestion scheme may be
264