AbstractsComputer Science

Performance analysis of a proposed hybrid optical network

by Natthapong Liamcharoen




Institution: University of Pittsburgh
Department:
Year: 2016
Posted: 02/05/2017
Record ID: 2066309
Full text PDF: http://d-scholarship.pitt.edu/29019/1/Liamcharoen_ETD2016_final.pdf


Abstract

This dissertation discusses a novel Hybrid Optical Network (HON) that can provide service differentiation based on traffic characteristics (i.e., packet, burst, and long-lived flow) with QoS guarantee not only in network layer, but also in physical layer. The DHON consists of sophisticated edge-nodes, which can classify, monitor, and dynamically adjust optical channels in the core layer as traffic variation. The edge nodes aggregate traffic, identifying end-to-end delay by ingress queuing delay or burst timeout. The network can estimate number of channels by arriving traffic intensity and distribution with estimated upper-bound delay. The core layer employs two parallel optical switches (OCS, OBS) in the same platform. Thanks to the overflow system, the proposed network enhances utilization with fewer long distance premium channels. The premium channel can quickly handle burst traffic without new channel assignment. With less overprovisioning capacity design, the premium channel enhances utilization and decrease number of costly premium channels. This research also proposes mathematic models to represent particular DHON channels (i.e., circuit, packet, and burst). We employ method of moments based on overflow theory to forecast irregular traffic pattern from circuit-based channel (i.e., M/M/c/c) to overflow channel, in which G/G/1 model based on Ph/Ph/1 matrix can represent the overflow channel. Moreover, secondary channel supports packet-based traffic over wavelength channel with two service classes: Class I based on delay sensitive traffic (i.e., long flow) and Class II for non-delay sensitive traffic (e.g., best effort). In addition, mixture of traffic in the wavelength channels is investigated based on M/G/1 and M/G/2 with specific service time distribution for particular class. Finally, we show our DHON based on (O-O-O) switching paradigm has improved the performance over typical (O-E-O) switching network architecture based on NSF topology.