Multiple Time Scale Congestion Control for Self-Similar Network Traffic

Analytical and empirical studies have shown that self-similar tra c can have a detrimental impact on network performance including ampli ed queuing delay and packet loss rate. Given the ubiquity of scale-invariant burstiness observed across diverse networking contexts, nding e ective tra c control algorithms capable of detecting and managing self-similar tra c has become an important problem. In this paper, we study congestion control algorithms for improving network performance|in particular, throughput|under self-similar tra c conditions. Although scale-invariant burstiness implies the existence of concentrated periods of contention and idleness, the long-range dependence associated with self-similar trafc leaves open the possibility that correlation structure at larger time scales may be exploited for performance enhancement purposes. We construct a 2-level multiple time scale congestion control protocol that exercises congestion control concurrently across two time scales an order of magnitude apart. The rst component|acting at the smaller time scale|is a generic linear increase/exponential decrease feedback congestion control that uses implicit prediction a orded by feedback to a ect rate control sensitive to changes in network state at 20{200ms time scales. The second component|acting at 2{5s time scales|uses explicit prediction to detect persistent shifts in overall network contention and uses this information to modulate the aggressiveness exhibited by the rst component. We show that cooperative interaction between the two congestion control modules acting on information at di erent time scales leads to improved performance visa-vis the case when the large time scale component is absent. We show that the improvement factor increases with long-range dependence and we show that as the number of ows engaging in multiple time scale congestion control (MTSC) increases, both fairness and e ciency are preserved.