Improved Competitive Ratios of Online Buffer Management Algorithms for Multi-Queue Switches in QoS Networks
浩二 小林, 修一 宮崎, 寿男 岡部 · Kyoto University Research Information Repository (Kyoto University) · 2008
The online buffer management problem formulates the problem of queuing policies of network switches supporting QoS (Quality of Service) guarantee. For this problem, a lot of models have been considered. Among others, we focus on multi-queue switches in QoS Networks proposed by Azar et~al. Azar et~al introduced the relaxed model in order to achieve a good upper bound on the competitive ratio for this model. In this paper, we improve the competitive ratios of several multi-queue models by improving an upper bound for the relaxed model. We propose an online algorithm $DS$ (Dual Scheduling) for the relaxed model. This algorithm works for (either preemptive or non-preemptive) 2-value model, but it uses as subroutines online algorithms for the non-preemptive unit-value model, which has been extensively studied. The performance of $DS$ depends on the performance of the algorithms used as subroutines. The followings are a couple of examples of the improvement on the competitive ratios of multi-queue models using our result: (i) We improved the competitive ratio of deterministic algorithms for the non-preemptive 2-value model from $4$ to $3.177$ for large enough $B$. a switch can store up to $B$ packets simultaneously. (ii) We proved that the competitive ratio of randomized algorithms for the non-preemptive 2-value model is at most $frac{17}{2} - sqrt{30} simeq 3.023$ for large enough $B$.