Performance of a distributed scheduling protocol for TWIN

Abstract

This paper discusses a scheduling mechanism for a new network architecture (TWIN) that provides arbitrary capacity up to a wavelength to any source-destination pair as needed, without optical-to-electronic conversion. The network emulates ultra-fast switching in the passive network core through the use of ultra-fast wavelength tunable lasers at the network edge. This architecture is suitable for any end-to-end traffic load, from static or quasi-static load (Sonet), to highly dynamic (IP) load. The key enabler of this architecture is a scheduling mechanism that schedules transmissions for maximal throughput. We propose a distributed scheduling scheme that is randomized for highly dynamic load and can learn to adjust for quasi-static load. We derive analytical formulae for the performance of the proposed scheme when load is highly dynamic, show that it outperforms standard protocols (such as <i>aloha</i>) and illustrate the effect of learning for quasi-static load through simulation.