This paper studies the performance of TCP transport protocol over an optical burst-switched (OBS) wavelength division multiplexed (WDM) wide-area network. Typically, an OBS network consists of optical core routers and electronic edge routers connected by WDM links. At the network ingress nodes, IP packets are assembled into bursts that are routed through the core network and disassembled at the network egress nodes. This paper studies the effects of OBS network characteristics and parameters on TCP's delay and throughput performance: (i) burstification (burst-assembly and disassembly) delays, (ii) data-burst scheduling and (iii) variation of burst packet parameters (i.e. burst size, burst time-outs), and (iv) burst drop probability. Detailed results based on an ns2-based simulator, that has been extended to incorporate WDM and OBS networking, are presented.

This paper studies the problem of designing survivable optical wavelength division multiplexed (WDM) networks. A wavelength-routed wide area backbone network supporting circuit-switched traffic is considered. This paper also considers the use of optical wavelength conversion technology which has been shown to help improve network performance. However, wavelength conversion is still an expensive technology and using optical conversion could potentially result in signal quality degradation. In survivable networks, protection against failures is provided using backup paths that are determined when a session is established. In this paper, we present three primary and backup route computation mechanisms that attempt to improve overall network performance compared to existing solutions. One of the key design goals is to reduce the number of required converters per node. First, we present a routing algorithm, termed conversion free primary routing (CFPR) that computes primary paths without wavelength conversion, as far as possible. Next, we present a converter multiplexing technique that is used to share wavelength converters among multiple backup paths. This significantly reduces the number of connections blocked due to wavelength converter unavailability and reduces the number of wavelength converters required at each node, thus reducing system cost. Finally, we propose a backup path relocation scheme that migrates existing backup paths, whenever needed, to accommodate more primary paths and also to obtain primary routes with fewer hops. This is done to improve network utilization and reduce blocking probability. The proposed techniques are analyzed in detail using a discrete-event simulation model. The results show that significant reduction in blocking probability is possible with the proposed mechanisms. The number of converters required at each node to achieve a given blocking probability is also seen to be four times lower, compared to existing architectures based on static shortest path routing.