Multiple TCP Flow Performance Study over OBS Networks

Optical burst assembly is a key function for the internetworking between TCP/IP networks and optical burst switched networks. TCP behaviour is influenced by the employment of the Optical Burst Switching (OBS) transfer mode in wide area networks for several reasons, mainly related to the aggregation of segments in bursts. The burst assembly/disassembly process takes place at network edges. Ingress edge routers perform the electrical-to-optical interfacing and collect incoming IP packets to form optical bursts. In this context a class is defined by a common egress edge router and by service requirements, when quality of service is supported. The assembly process proceeds as follows: when an IP packet arrives at the edge node, it is processed and then queued, in relation to the assigned class, to be included in a burst. In order to preserve optical network time transparency, the assembly procedure is equipped with an assembly time out or a limit on maximum burst size that makes the payload ready to leave. Then, the prepared payload is sent to an E/O conversion unit to be transmitted into the proper fibre and wavelength. A control packet is also generated for each burst in order to reserve network resources, such as wavelengths, at each node along the path from the ingress to the egress node, for a proper time interval. As far as TCP segment aggregation in optical bursts, two main different strategies can be adopted: - per flow assembly: an optical burst contains data of the same class and of the same flow. Per flow queuing is needed at the ingress of the assembly unit; - mixed flow assembly: a burst may contain information from different flows of the same class. Per class queuing is needed at the ingress of the assembly unit. The mixed flow solution leads to a simpler implementation of the assembly unit, which mainly depends on the number of O-SAPs and not on the number of flows. Burst losses represent the aspect that mainly influences end to end performance. They are a consequence of contention in OBS nodes and depend on node architecture, scheduling and contention resolution algorithms. Routing strategies and network dimensioning which avoid traffic concentration on few links allow to reduce contention occurrence. Burst losses cause, in their turn, multiple segment losses which lead to correlation in TCP segment losses. The effect of these losses depends on the level of segment aggregation in a burst and on the assembly strategy. In this paper end to end performance are studied in a multiple flow scenario and recent results on the influence of different assembly strategies is investigated by means of simulation.