We study the performance of bidirectional TCP/IP connections over a network
that uses rate-based dow and congestion control. An example of such a netw
ork is an asynchronous transfer mode (ATM) network using the available bit
rate (ABR) service, The sharing of a common buffer by TCP packets and ackno
wledgment tacks) has been known to result in an effect called ack compressi
on, where acks of a connection arrive at the source bunched together, resul
ting in unfairness and degraded throughput. It has been the expectation tha
t maintaining a smooth flow of data using rate-based how control would miti
gate, if not eliminate, the various forms of burstiness experienced with th
e TCP window flow control. However, we show that the problem of TCP ack com
pression appears even while operating over a rate-controlled channel, altho
ugh the queues are primarily at the end systems now. The degradation in thr
oughput due to bidirectional traffic can be significant. For example, even
in the simple case of symmetrical connections with adequate window sizes, t
he throughput of each connection is only 66.67% of that under one-way traff
ic.
By analyzing the periodic bursty behavior of the source LP queue, we derive
estimates for the maximum queue size and arrive at a simple predictor for
the degraded throughput, applicable for relatively general situations. We v
alidate our analysis using simulation on an ATM network using the explicit
rate option of the ABR service. We show that the analysis predicts the beha
vior of the queue and the throughput degradation not only in simple configu
rations, but also in more general situations, such as multiple synchronized
connections between a pair of end systems, multiple end systems having dif
ferent round-trip delays, and in configurations with significant amounts of
cross-traffic.