Active Messages: a Mechanism for Integrated Communication and

摘要: The design challenge for large-scale multiprocessors is (1) to minimize communication overhead, (2) allow overlap computation, and (3) coordinate the two without sacrificing processor cost/performance. We show that existing message passing have unnecessarily high costs. Research prototypes of messages driven machine demonstrate low but poor introduce a simple mechanism, Active Messages, it intrinsic both architectures, allows cost effective use hardware, offers tremendous flexibility. Implementations on nCUBE/2 CM-5 are described evaluated using split-phase shared-memory extension C, Split-C. further active sufficient implement dynamically scheduled languages which machines were designed. With this latency tolerance becomes programming/compiling concern. Hardware support