Paper Review : Dynamic Behaviour of Slowly-Responsive Congestion Control Algorithms
Reviewer : Seh Leng Lim
This paper is an attempt to investigate the behaviour of slowly-responsive, TCP-compatible congestion control algorithms under more realistic dynamic network conditions, with the aim of answering the fundamental question of whether these algorithms are safe to deploy in the Internet. This is because currently, TCP compatibility condition refers only to static conditions, and not dynamic ones like the Internet.
The main contribution of the paper is its observation that in return for smoother transmission rates, slowly-responsive congestion control (SlowCC) algorithms lose throughput to faster ones (like TCP) under dynamic network conditions, thereby not achieving long-term fairness. In fact, the TCP-compatible algorithms all appeared to be safe for deployment, as the more slowly responsive ones can be made to avoid causing the network to go into persistent overload persistent loss rates on sudden bandwidth reductions by incorporating a self-clocking mechanism based on packet conservation. The research from this paper is useful to encourage support for alternative congestion control algorithms which are TCP compatible and which respond more slowly to congestion, thereby producing a smoother bandwidth usage profile. These slowly-responsive TCP compatible algorithms are needed to provide smoother bandwidth usage for streaming video and audio applications, which current TCP is very poor at.
The key main ideas expounded are:
(a) TCP compatible whereby a congestion control mechanism is TCP compatible if its bandwidth usage, in the presence of a constant loss is the same as TCP.
(b) The idea of moving from a TCP controlled world to one where there is no single dominant congestion control mechanism and instead there is a wide variety of mechanisms tailored to different application requirements.
I think that the paper has a significant contribution (rating of 4) to the study of alternative congestion control algorithms which are TCP compatible and slowly-responsive in a dynamic environment. The authors made use of a single-bottleneck “dumbbell” configuration for their experiments. The authors ran stress tests which involve “square-wave” and “saw-toothed” patterns of available bandwidth to benchmark the behaviour of SlowCC under these conditions. They studied characteristics of the SlowCC such as stabilization time (defined as the time for the packet loss rate to stabilize after the start of a sustained period of high congestion, where the packet loss rate has stabilized if it is within 1.5 times its steady-state value), long-term fairness bandwidth fairness, transient fairness, bottleneck link utilization and smoothness of transmission rates. Although the experiments are based on simulation, they are actually stress tests which means that their settings may be worst than in reality. Therefore, I see that their observation of the behaviour of SlowCC in their experiments may provide the lower bound for the behaviour of SlowCC in the Internet.
The fact that SlowCC will not disrupt the TCP dominated Internet is not sufficient to attract users. I feel that the lack of aggressiveness of SlowCC in utilizing new available bandwidth is a huge drawback, and may discourage their adoption, unless absolutely necessary. Therefore, to realize the author’s vision of a TCP-compatible Internet world, much more research needs to be done to get SlowCC to achieve bandwidth fairness.
Researchers and builders who build streaming audio and video applications will have a better appreciations from this paper of the technical difficulties faced in using alternative TCP-compatible congestion control algorithms over the Internet.