Communication strategies for shared-bus embedded multiprocessors

13 years 10 months ago

Download www.ece.umd.edu

Abstract— This paper explores the problem of efﬁciently ordering interprocessor communication operations in both statically and dynamically-scheduled multiprocessors for iterative dataﬂow graphs with probabilistic execution times. In most digital signal processing applications, the throughput of the system is signiﬁcantly affected by communication costs. We explicitly model these costs within an effective graph-theoretic analysis framework. We show that ordered transaction schedules can signiﬁcantly outperform both self-timed schedules and dynamic schedules for moderate task execution time variability. As the task execution time variability increases, we show that ﬁrst selftimed and then dynamic scheduling policies are preferred. We perform an extensive experimental comparison on both real and simulated benchmarks to gauge the effect of synchronization and communication overhead costs on these crossover points.

Neal K. Bambha, Shuvra S. Bhattacharyya

Real-time Traffic