Eecient Java Rmi for Parallel Programming

Java offers interesting opportunities for parallel computing. In particular, Java Remote Method Invocation (RMI) provides a flexible kind of Remote Procedure Call (RPC). Unlike RPC, RMI supports polymorphism, which requires the system to be able to download remote classes into a running application. Sun’s RMI implementation achieves this kind of flexibility at the cost of a major runtime overhead. Using Sun’s JDK 1.2 on a Pentium Pro/Myrinet cluster, for example, the latency for a null RMI (without parameters or a return value) is 1316 μs, 42 times that of a user-level RPC. In this paper, we study a new approach for implementing RMI, based on native compilation. This approach allows for better optimization, reduces the need for processing of type information at runtime, and uses a lightweight communication protocol. We implemented such an RMI protocol in a compiler-based Java system, called Manta. To support polymorphic RMIs and to allow interoperability with other Java Virtual Machines, Manta also implements the Sun protocol and is capable of dynamically compiling and linking bytecode into a running application. The null-latency of our Manta protocol over Myrinet is 37 μs, 35 times faster than the Sun JDK, and only 6 μs slower than a C-based RPC protocol. The cause for this high performance is that almost all of the runtime overhead of RMI has been pushed to compile time. We study the performance differences between the Manta and the Sun RMI protocols in detail. The poor performance of the Sun RMI protocol is in part due to an inefficient implementation of the protocol. We therefore built a more efficient implementation of the Sun protocol by compiling it with the Manta native compiler and reducing various overheads of the protocol. The null-latency for this compiled Sun RMI system is 301 μs, still 8 times slower than Manta. We study various latency and throughput benchmarks and breakdowns for Manta and for the compiled Sun RMI protocol. Finally, we describe the impact of the RMI protocols on application performance. Performance measurements using up to 32 CPUs show that five out of six parallel programs obtain high efficiencies (at least 75%) with Manta. Only two coarse-grained applications perform well with the compiled Sun RMI protocol; on the other four applications, Manta obtains 1.9 to 3.4 times higher speedups.