PCP Design

Table 2: A Summary of PCP Mechanisms

Mechanism	Description	Goal	Section
probes	Senders use short transmission bursts with limited payload to ``prove'' the existence of available bandwidth, while minimizing any long term effects of failed tests.	low loss	Section 3.1
direct jump	Given a successful test, senders increase their base rate to the rate that test.	min response time	Section 3.1
probabilistic accept	Accept tests taking into account the variance observed in the available bandwidth measurements.	fairness	Section 3.1
rate compensation	When existing senders detect increasing queueing, they reduce their rates to drain the queue.	low loss, low queues	Section 3.2
periodic probes	Senders periodically issue new probes to try to acquire additional bandwidth.	work-conserving	Section 3.3
binary search	Senders use binary search to allocate the available bandwidth.	min response time, work conserving	Section 3.3
exponential backoff	Senders adjust the frequency of tests to avoid test collisions and failures.	stability	Section 3.3
history	Senders use heuristics to choose the initial probe rate.	min response time	Section 3.4
tit for tat	Reduce speed of rate compensation if past compensation was ineffective.	TCP compatibility	Section 3.5

In this section, we sketch the various elements of the PCP design. Table 2 provides a road map. For this discussion, we assume that PCP is used by all endpoints in the system to manage network resources; we defer to the end of this section a discussion of how to make PCP backwardly compatible with TCP end hosts. PCP represents a clean-slate redesign to endpoint congestion control; we do however retain the TCP mechanisms for connection management and flow control.