Instant Service Policy and Its Application to Deficit Round Robin

Abstract

Many of scheduling algorithms that provide a predefined bandwidth to a traffic flow fall into a category of Latency-rate ($\mathcal{LR}$) server. A series of $\mathcal{LR}$ servers can be viewed as a virtual node with an $\mathcal{LR}$ server of the latency which is simply the summation of individual latencies of actual $\mathcal{LR}$ servers. Deficit Round Robin (DRR) is such an $\mathcal{LR}$ server and the simplest one to implement, so that it is adopted in many real systems. In this research we suggest a novel policy of Instant Service, which is applicable to any round-robin schedulers. We then apply this policy to DRR, make a variation called DRR with Instant Service (DRR-IS), and analyze it. We prove that the DRR-IS is still an $\mathcal{LR}$ server. We calculate its latency and investigate its fairness characteristics. We demonstrate the DRR-IS, compared with DRR, provides about 30% better latency while have the same complexity and the same fairness.