Analysis and synthesis of optical burst switched networks

光突發交換網絡的分析和綜合

Student thesis: Doctoral Thesis

View graph of relations

Author(s)

  • Shuo LI

Related Research Unit(s)

Detail(s)

Awarding Institution
Supervisors/Advisors
Award date15 Jul 2014

Abstract

Optical burst switching (OBS) is an optical networking technology for transporting data from any edge router (ingress node) to another edge router (egress node) through optical cross connects (OXC). It combines the benefits of optical packet switching and optical circuit switching (OCS), while avoiding their limitations. In an OBS network, data packets from the access network destined to the same egress node are aggregated at the ingress node to form a burst which is sent to the core network with a separate burst control packet. This burst control packet is transmitted in the optical domain through a separate control channel but processed electronically at each core node to reserve bandwidth for the burst. The burst is transmitted all-optically after some time without waiting for any acknowledgment from the egress node. When the burst reaches the egress node, it is segregated into packets which are forwarded to their destinations in the access network. Since OBS uses one-way reservation, there is no guarantee that a transmitted burst will reach its destination, and bursts may be dropped at intermediate nodes due to contention. Thus, burst blocking probability is an important performance measurement in OBS networks. Deflection routing is one of the key contention resolution strategies. Erlang Fixed Point Approximation (EFPA) and Overflow Priority Classification Approximation (OPCA) are two algorithms that have been used to evaluate the blocking probability in OBS with deflection routing. Both EFPA and OPCA require fixed-point iterations which may require a long running time. In the first part of this thesis, we study the properties of OPCA, which is based on a hierarchical structure with a finite number of layers, and prove that the OPCA iterations alternately produce upper and lower bounds which consistently become closer to each other as more fixed-point iterations in each layer are used. We also demonstrate numerically that only a small number of iterations per layer are required for the bounds to be sufficiently close to each other. This behavior is demonstrated for various system parameters such as offered load, number of links per trunk, and maximum allowable number of deflections. Next, we study two new performance measurements in OBS networks. The bursts dropped at intermediate nodes have already utilized the bandwidth before they are dropped, but do not contribute to the network throughput. Thus the utilization is sometimes very high but the throughput is quite low in the network. We classify trunk utilization into effective utilization and ineffective utilization to distinguish the utilization caused by bursts that reach and do not reach their destinations, respectively. As a benchmark for OBS, we consider an idealized version of OCS, designated I-OCS, which does not incur ineffective utilization. We study the efficiency of OBS versus I-OCS networks for different network models to facilitate the understanding of performance implications of effective and ineffective utilizations. Based on the idea to reduce ineffective utilization by increasing the likelihood that bursts which have already used a lot of network resources reach their destinations, thereby increasing network throughput, we combine contention resolution strategies - burst segmentation and least remaining hop-count first (LRHF), with an Emulated-OBS wavelength reservation scheme, designated as EBSL, to solve the burst contention problem. We shall use I-OCS and I-OCS with alternative routing as benchmarks. Through discrete event simulations, the performance of EBSL is compared with that of original OBS/JET, Emulated-OBS, Emulated-OBS with segmentation (tail dropping) and Emulated-OBS with deflection routing and trunk reservation. The results show that our proposed strategy can eliminate the congestion collapse of the goodput and effective utilization, reduce the blocking probability under heavy load conditions. Using the fairer version of LRHF instead of LRHF in EBSL also provides insight into fairness efficiency tradeoffs. Furthermore, adding deflection routing to EBSL can provide better performance under light and medium traffic loads.

    Research areas

  • Packet switching (Data transmission), Computer networks, Optical communications