作者: Michael Philippsen , Bernhard Haumacher , Christian Nester
DOI: 10.1002/1096-9128(200005)12:7<495::AID-CPE496>3.0.CO;2-W
关键词: Machine code 、 Benchmark (computing) 、 Myrinet 、 Ethernet 、 Fast Ethernet 、 Serialization 、 Java 、 Marshalling 、 Operating system 、 Computer science
摘要: In current Java implementations, Remote Method Invocation (RMI) is too slow, especially for high-performance computing. RMI designed wide-area and high-latency networks, it based on a slow object serialization, does not support communication networks. The paper demonstrates that much faster drop-in an efficient serialization can be implemented completely in without any native code. Moreover, the re-designed supports non-TCP/IP even with heterogeneous transport protocols. We demonstrate computing some of official serialization's generality should traded speed. As by-product, benchmark collection presented. On PCs connected through Ethernet, better improved save median 45% (maximum 71%) runtime set arguments. our Myrinet-based ParaStation network (a cluster DEC Alphas) we 85% 96%), compared to standard RMI, Fast Ethernet; remote method invocation runs as fast 80 μs round trip time, about 1.5 ms. Copyright © 2000 John Wiley & Sons, Ltd.