FBICM: Efficient Congestion Management for High-Performance Networks Using Distributed Deterministic Routing

Abstract. As the number of components in cluster-based systems increases, cost and power consumption also increase. One way to reduce both problems is us-ing smaller networks with adequate congestion management mechanisms. Re-cent successful proposals (RECN) eliminate the negative effects of congestion, the Head-of-Line (HOL) blocking, leaving congestion harmless. RECN relies on source-based networks architectures, where the entire route is placed at packet headers before injection. Unfortunately, distributed table-based routing is also common in cluster-based networks, being InfiniBand the most prominent exam-ple. We propose a novel congestion management technique for distributed table-based routing. The mechanism relies on additional congestion information lo-cated at routing tables. With this information HOL blocking is minimized by smartly using switch queues. Detailed memory organization and the way conges-tion information is updated/propagated is described. Preliminary results indicate that with modest resource requirements maximum network performance is kept regardless of congestion