Why flow-completion time is the right metric for congestion control

Abstract

Abstract-Users typically want their flows to complete as quickly as possible: They want a web-page to download quickly, or a file transfer to complete as rapidly as possible. In other words, Flow Completion Time (FCT) is an important -arguably the most important -performance metric for the user. Yet research on congestion control focuses almost entirely on maximizing flow throughput, link utilization and fairness, which matter more to the operator than the user. In this paper we show that existing (TCP Reno) and proposed (XCP) congestion control algorithms make flows last much longer than necessary -often one or two orders of magnitude longer than they need to; and we argue that over time, as the network bandwidth-delay product increases, the discrepancy will get worse. In contrast, we show how a new and practical algorithm -RCP (Rate Control Protocol) -enables flows to complete close to the minimum possible.

Loss-resilient window-based congestion control

“…Moreover, our simulations have revealed that the behavior of the loss-resilience system is not sensitive to the actual value of the retransmission timeout. 3 Hence, we conclude that defining the retransmission timeout as a roughly proportional but larger version of the recovery timeout prevents unstable behavior of the system. At the same time, it permits efficient recovery of multiple losses of the same packet.…”

“…Two families of approaches have been proposed to implement congestion control. Rate-based algorithms directly control the transmission rate of the connection [2,3], whilst window-based algorithms force the connection to obey a 'packet conservation' principle, which means that a new packet is not pushed into the network until an old packet leaves [4]. Typically, window control injects bursts of data into the network, whilst rate control pushes data at regular time interval so as to meet a target average rate.…”

A robust adaptive congestion control strategy for large scale networks with differentiated services traffic

“…The queueing state of the entire network after applying the controller (15) is now given bẏ (20) where τ l (t) denotes the time-varying delay, l = 1, ..., M ; M is the number of time-varying delays in the network; A 0 , A l , B l , and B c are the system matrices that are defined as follows…”

“…Motivated from the robust adaptive control techniques in [24], the time-varying gain k pi (t) is now designed according to the modified parameter projection method and is applied to system (20) to estimate the unknown but bounded incoming traffic λ pi (t) as followṡ [13] is a special case of (22) when…”

“…FLUID FLOW MODEL OF A LARGE SCALE DIFF-SERV NETWORK An "ideal" congestion control mechanism should be able to simultaneously satisfy the QoS specifications of the aggregate traffics in addition to congestion avoidance. The main QoS specifications of the aggregate flows and the performance metrics of any congestion control scheme should include both the node throughput as well as the delay and the packet loss rates [19], [20]. Given the trade-offs among the performance metrics it is clearly important to consider them all together.…”