Are independent traffic sources suitable for the evaluation of interconnection networks

Many simulation-based performance studies of interconnection networks are carried out using synthetic workloads under the assumption of independent traffic sources. Although this assumption may be useful for some traffic patterns, it may lead to erroneous conclusions about the usefulness of some design proposals for heavy loads. In this work we show that some traffic patterns generated by independent sources introduce an undesired uneven utilization of resources that privilege some nodes (compared to others), and that affects performance measurements in unexpected ways. The limitation of these privileges, for example using burst-synchronized traffic that forces fast nodes to wait for slower ones, produces more realistic performance results. Just as an example, in this work we show that the utilization of a congestion control mechanisms looks