Prioritized multiprocessor networks: design and performance

摘要

This paper proposes and evaluates prioritized direct shared-memorynmultiprocessor networks. We use three components to implementnprioritized networks, namely, priority-based link arbitration, priorityninheritance, and dynamic virtual channels. The two major results fromnour study are: (i) adding priorities to direct shared-memorynmultiprocessor networks can lead to reduced average transactionnlatencies and increased system throughput when running traditionalnparallel applications, and (ii) a prioritized multiprocessor network cannbe used to reduce the worst-case latencies of time-constrained trafficnwhen it co-exists with best-effort traffic, without penalizing thenaverage performance of best-effort traffic