Fair scheduling in an optical interconnection network

Abstract

Existing fair scheduling schemes have focused primarily on scheduling multiple ows to a single output. The limited work that has focused o n s c h e duling multiple ows to multiple outputs has assumed a non-blocking, slotted-time, cell-based network with a centralized c ontroller. This paper presents a fair scheduler suitable for use in bu erless circuit-switched blocking networks operating with distributed, asynchronous controllers and variable length messages. We begin by describing the potential for starvation in the Gemini interconnect network, an optical, circuit-switched network. A proposed distributed fair scheduler is presented and shown to solve this problem. The tradeo s and limitations of performing many-to-many fair scheduling in general, and that of our fair scheduler in particular, are discussed.

A Scheduling Algorithm in a Core Optical Router With Heterogeneous Traffic

“…Since it deals with the best-effort services only, the service differentiation issues are not addressed in [4,5]. A fair scheduler presented in [6] is suitable in buffer-less circuit-switched blocking networks operating with distributed, asynchronous controllers and variable length messages. The tradeoffs and performance limitations of the fair scheduler are discussed.…”

Gemini: an optical interconnection network for parallel processing

“…The Gemini interconnect is a novel processor-to-processor interconnection network for tightly-coupled multicomputers [7], [8], [9]. It includes an end-to-end optical data path (including switching of the optical signals) for highbandwidth, large data volume message delivery.…”