Maximizing TCP throughput in OBS networks

Abstract

A key goal of designing memoryless OBS networks is the minimization of burst loss. Loss is due mainly to restraints that can occur at core nodes, such as contention. Such events occurring at OBS layer are misinterpreted by TCP congestion control mechanisms that reacts on (i) burst drops, (ii) end-to-end delay variation and (iii) throughput changes. This paper proposes a destination-ordering mechanism over a load balanced OBS network, that minimizes TCP misinterpretation and maximizes TCP throughput.