Congestion Control for High Band Width Delay Network

Abstract

newlineTheory and simulations suggest that current Internet congestion control mechanisms are likely to run into difficulty in the long term as the per-flow bandwidth-delay product increases. This motivated us to step back and re-evaluate both control law and signalling for congestion control. Motivated by CSFQ, we chose to convey control information between the end-systems and the routers using a few bytes in the packet header. The most important consequence of this explicit control is that it permits a decoupling of congestion control from fairness control. In turn, this decoupling allows more efficient use of network resources and more flexible bandwidth allocation schemes. Based on these ideas, we devised XCP, an explicit congestion control protocol and architecture that can control the dynamics of the aggregate traffic independently from the relative throughput of the individual flows in the aggregate. Controlling congestion is done using an analytically tractable method that matches the aggregate traffic rate to the link capacity, while preventing persistent queues from forming. The decoupling then permits XCP to reallocate bandwidth between individual flows without worrying about being too aggressive in dropping packets or too slow in utilizing spare bandwidth. We demonstrated a fairness mechanism based on bandwidth shuffling that converges much faster than TCP does, and showed how to use this to implement both min-max fairness and the differential bandwidth allocation. Our extensive simulations demonstrate that XCP maintains good utilization and fairness, has low queuing delay, and drops very few packets. We evaluated XCP in comparison with TCP over RED, REM, AVQ, and CSFQ queues, in both steady-state and dynamic environments with web-like traffic and with impulse loads. We found no case where XCP performs significantly worse than TCP. In fact, when the per-flow delay-bandwidth product becomes large, XCP s performance remains excellent whereas TCP suffers significantly. We believe that XCP is viable and practical as a congestion control scheme. It operates the network with almost no drops, and substantially increases the efficiency in high bandwidth-delay product environments. By putting the control state in the packets, XCP needs no per flow state in routers and can scale to any number of flows. Further, our implementation requires only a few CPU cycles per packet, making it practical even for high-speed routers. Using a control theory framework that is a fluid model of the protocol is stable for any link capacity, feedback delay, or number of sources. In contrast to the various AQM schemes where parameter values depend on the capacity, delay, or number of sources, our analysis shows how to set the parameters of the new protocol to constant values that are effective independent of the environment. Our extensive packet-level simulations are show that, regardless of the queuing newlinescheme, TCP s performance degrades significantly as either capacity or delay increases. In contrast, the new protocol achieves high utilization, small queues, and almost no drops, independent of capacity or delay. Even in conventional environments, the simulations show that our protocol exhibits better fairness, higher utilization, and smaller queue size, with almost no packet drops. Further, it maintains good performance in dynamic environments with many short web-like flows, and has no bias against long RTT flows. A unique characteristic of the new protocol is its ability to operate with almost zero drops. Although we started with the goal of solving TCP s limitations in high-bandwidth large-delay environments, our design has several additional advantages. newlinePopular myths that cheaper memory, high-speed links, and high-speed processors will solve the problem of congestion in computer networks are shown to be false. A simple definition for congestion based on supply and demand of resources is proposed and is then used to classify various congestion schemes. The issues that make the congestion problem a difficult one are discussed, and the architectural decisions that affect the design of a congestion scheme are presented. It is argued that long-, medium-, and short-term congestion problems require different solutions. Some of the recent schemes are briefly surveyed, and areas for further research are suggested.