Unfairness problem of Optical Flow Switching network under self-similar traffic

We show that OFS suffers from the long-flow short-flow (LFSF) unfairness problem especially for single wavelength network, and this problem becomes even more severe for the real-world self-similar Internet traffic. By segmenting the long flows into many short flows and scheduling the transmission dynamically, the LFSF unfairness can be reduced significantly. By combining the segmentation strategy with the previously proposed virtual output queuing (VOQ) scheduling approach, it is shown that the throughput of OFS network under self-similar traffic can be improved from 25% to 85% for the single wavelength case, and from 55% to over 90% for the multi-wavelength case.