Low-latency Java communication devices on RDMA-enabled networks

摘要

Providing high-performance inter-node communication is a key capability for running high performancerncomputing applications efficiently on parallel architectures. In fact, current systems deployments are aggregatingrna significant number of cores interconnected via advanced networking hardware with Remote DirectrnMemory Access (RDMA) mechanisms, that enable zero-copy and kernel-bypass features. The use of Javarnfor parallel programming is becoming more promising thanks to some useful characteristics of this language,rnparticularly its built-in multithreading support, portability, easy-to-learn properties, and high productivity,rnalong with the continuous increase in the performance of the Java virtual machine. However, current parallelrnJava applications generally suffer from inefficient communication middleware, mainly based on protocolsrnwith high communication overhead that do not take full advantage of RDMA-enabled networks. This paperrnpresents efficient low-level Java communication devices that overcome these constraints by fully exploitingrnthe underlying RDMA hardware, providing low-latency and high-bandwidth communications for parallelrnJava applications. The performance evaluation conducted on representative RDMA networks and parallelrnsystems has shown significant point-to-point performance increases compared with previous Java communicationrnmiddleware, allowing to obtain up to 40% improvement in application-level performance on 4096rncores of a Cray XE6 supercomputer.