Discussion

In this chapter, we have experimented with and benchmarked the HPJava language on scientific and engineering applications on a shared memory machine (Sun Solaris 9 with 8 Ultra SPARC III Cu 900 MHz Processors and 16 GB of main memory) and a distributed memory machine (IBM SP3 running with four Power3 375 MHz CPUs and 2 GB of memory on each node). We have explored the performance of the HPJava system on both machines using the efficient node codes we have benchmarked in chapter 6. The speedup of each HPJava application is very satisfactory even with expensive run-time communication libraries such as Adlib.writeHalo() and Adlib.sumDim(). Moreover, performances on both machines shows consistent and similar behaviour as we have seen on the Linux machine. One machine doesn't have a big advantage over others. Performance of HPJava is good on all machines we have benchmarked. When program architects design and build high-performance computing environments, they have to think about what system they should choose to build and deploy the environments. There rarely exists machine-independent software, and performance on each machine is quite inconsistent. HPJava has an advantage over some systems because performance of HPJava on Linux machines, shared memory machines, and distributed memory machines are consistent and promising. Thus, we hope that HPJava has a promising future, and can be used anywhere to achieve high-performance parallel computing.