Early regulation of aggressive flows.

摘要

In this thesis work, we propose router mechanisms to regulate aggressive best-effort traffic. By aggressive traffic, we mean flows that do not reduce their sending rate in response to congestion or consume much more bandwidth than a TCP connection with the same parameters. The problems that our mechanisms address range from unfairness against competing traffic that employ congestion control to congestion collapse. The goal of the proposed mechanisms is to drop undeliverable packets—packets that are dropped somewhere in the network before they reach their destination—as close to the periphery of the network as possible. The key ideas of our approach are: (1) routers at the edge of the network keep track of incoming flows and their arrival rates; (2) routers use RED (Random Early Detection) for queue management and generate rate-limited source quenches on packet drops to advise sources to reduce their sending rates; and (3) routers at the edge of the network snoop on source quenches passing through them and use these to control per-flow regulators. Regulators adjust their maximum sending rate using a multiplicative-decrease, additive-increase discipline. A decrease is triggered by the arrival of a source quench; an increase is triggered for every estimated round-trip time period. We named our mechanism ERAF - Early Regulation of Aggressive Flows.; In order to better understand the behavior of ERAF we developed an analytical performance model which we validated. We also examined the interactions between Source Quench congestion control and TCP congestion control. The evaluation and validation of our mechanism was done through simulations using the ns-2 network simulator. We considered a large variety of network topologies and traffic patterns.; Our results indicate that ERAF is able to shield responsive traffic from the impact of unresponsive and aggressive traffic and prevent congestion collapse, while consuming a reasonable amount of network resources. The bandwidth requirements for ERAF signaling traffic are low; ERAF is able to operate efficiently using less than 1% of the available link bandwidth in most cases. (Abstract shortened by UMI.)