Approximate sorting of packet-scheduling in high-speed networks

Abstract

Fairness, latency and computational complexity are three important factors in evaluating the performance of a scheduling algorithm. Fairness must be satisfied so that service can be distributed according to the reserved rate. Only when latency is irrelevant to the number of connections, is it possible to minimize the end-to-end delay through controlling the reserved rate. Among existing scheduling algorithms, Round Robin is the least complex. However, conventional Round Robin is unable to ensure fairness, and the improved round robin algorithms like Deficit Round Robin, Weighted Round Robin and Virtual Round Robin are unable to ensure that their latencies are irrelevant to the number of connections although they guarantee fairness. Potential Round Robin developed for analysis of fairness and latency reduction is thus proposed. It is based on the introduction of a new concept, Round Potential Function. The function splits service time into a number of service round periods to guarantee fairness regardless of the serving process used in the period. In the analysis of latency, service round periods are re-split into multiple scanning cycles for further service distribution with approximate sorting between scanning cycles. As a result, latency is no longer relevant to the number of connections while the low complexity of round robin is kept.