A High-Performance DSM Leveraging Software Defined Network

Distributed shared memory(DSM) is an implementation of the shared memory model in distributed systems. Software-defined networking (SDN), can be effectively exploited to improve the performance of DSM. In this research, we propose an SDN-based distributed shared memory architecture. This architecture improves distributed applications by tuning network transmission and offloading functionality to the underlying network. We present a dynamic SDN multicast mechanism to reduce the number of consistency-related messages without interfering the unrelated processes in the same multicast group. In addition to these, we design a state-aware transmission mechanism that reduces the contention delays by using multicast communication. Furthermore, a novel write-update mechanism based on SDN orderly multicast is presented. The experiments on the prototype system show that the distributed applications using SDN-based DSM yield remarkable performance improvements. Moreover, this type of prototype system is feasible to implement on existing hardware.