Imperial College London
Computer clusters equipped with reconfigurable accelerators have shown promise in high performance computing. This paper explores novel ways of customizing data communication between accelerator nodes, which is often a bottleneck when scaling up the cluster size. Based on the direct connection of high speed serial links between advanced reconfigurable devices, the authors develop and evaluate CusComNet, a scalable, flexible and efficient communication framework. The CusComNet framework is built around customizable, packet-based communication and supports three main types of customization: packet protocol customization, system-level customization, and prioritized communication customization.