Hardware Implementation of an Advanced RDMA Engine

Modern computing clusters consist of many heterogeneous computing units that work collectively in order to fulfill high computing tasks. Low latency communication between the remote processes that run on these servers is a critical factor for achieving high performance. This paper describes the hardware implementation of an advanced RDMA engine which is used in the ExaNeSt prototype. The RDMA engine provides advanced resiliency and Quality-of-Service (QoS) features, supports multipath, and allows user-level transfer initiation