Abstract

AbstractDistributed Shared Memory (DSM) architectures are becoming popular to exploit parallelism of architectures while offering flexibility of using both shared and distributed memory paradigms to application developers. At the same time, Networks on Chip (NoC) have become reality to address communication bottlenecks in massively parallel tile-based processor architectures. In NoC-based DSM architectures, the synchronization overhead for spawning a task on a remote network node may lead to high performance penalties. In order to reduce the synchronization delays during remote task spawning, the design of Network Interface (NI) becomes important. In this paper, we present a network interface architecture which supports task spawning between network nodes by employing efficient synchronization mechanisms. The proposed NI internal hardware support offloads the software from handling the synchronization during remote task spawning and hence results in achieving better overall performance. Simulation results highlight that the proposed hardware architecture improves the performance by up to 42 % in comparison to existing state of the art approaches. The FPGA prototype is also used to depict the benefits of the proposed approach for real world applications. Implementation results show the low area footprint of the proposed hardware.KeywordsShared MemoryNetwork InterfaceDistribute Shared MemoryMessage Sequence ChartRemote Direct Memory AccessThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call