Control-Plane Congestion in Optical-Burst-Switched Networks

We examine optical burst switching (OBS) networks with out-of-band header transmission and electronic processing. We present the first detailed analysis of the potential effects of control- plane processing limitations on the overall throughput and latency performance of OBS networks. We present an accurate analytical model that explicitly accounts for the header queuing process in core nodes, and we provide a set of design guidelines for provisioning OBS networks such that the control- plane does not become the throughput bottleneck of the system. We also estimate the minimum end-to-end latency associated with offsets and burst assembly that is required to ensure proper OBS operation. We find that ultrafast header-processing speeds (<100 ns per header) are not required for efficient OBS operation. We show that provisioning a header-offset size that corresponds to a header-processing-queue length of 50 is sufficient for a wide range of practical OBS systems. For a fully meshed network topology, a total end-to-end control latency that is 10 times longer than the duration of a single header- processing duration is required for proper network operation. By contrast, more sparsely connected networks, such as those deployed in the long haul and metro today, may require an average end-to-end control latency that is hundreds of times as large as the header-processing duration.