Randomized on-line matching (ROLM): randomized scheduling algorithms for high-speed cell-based switches

Abstract

Scheduling algorithms for virtual output queuing (VOQ) switches aim to provide high performance (low latency and high throughput), fairness as well as easy and low cost implementation. These goals have led to the development of a wide range of schedulers that are characterized by deterministic decision making. Randomness in decision making enables high performance scheduling as well as fair treatment of requests, when used appropriately. However, it has been proven to be costly in switching environments, which require high speed decisions, as the experience with the first randomized algorithm, namely PIM, is shown. In this paper, we introduce a class of practical scheduling algorithms that employ randomness in a fashion that leads to fast, low cost implementations. The class of algorithms, named random on-line matching (ROLM), employs randomness effectively, calculating maximal matchings and thus, leading to high performance schedulers with optimal mean service time that outperform significantly the competitive deterministic schedulers; furthermore, it employs randomness efficiently, resulting to schedulers that can accommodate effectively high-speed links