In current Java implementations, Remote Method Invocation (RMI) is too slow
especially for highperformance computing. RMI is designed for wide-area an
d high-latency networks, it is based on a slow object serialization, and it
does not support high-performance communication networks. The paper demons
trates that a much faster drop-in RMI and an efficient drop-in serializatio
n can be designed and implemented completely in Java without any native cod
e, Moreover, the re-designed RMI supports non-TCP/IP communication networks
, even with heterogeneous transport protocols. We demonstrate that for high
-performance computing some of the official serialization's generality can
and should be traded for speed. As a by-product, a benchmark collection for
RMI is presented, On PCs connected through Ethernet, the better serializat
ion and the improved RMI save a median of 45% (maximum of 71%) of the runti
me for some set of arguments. On our Myrinet-based ParaStation network (a c
luster of DEC Alphas) we save a median of 85 % (maximum of 96 %), compared
to standard RMI, standard serialization, and Fast Ethernet; a remote method
invocation runs as fast as 80 mu s round trip time, compared with about 1.
5 ms. Copyright (C) 2000 John Wiley & Sons, Ltd.