Early experiences with distributed supercomputing on I-way: First principles materials science and parallel acoustic wave propagation

The authors present experiences relating to two simulations running on a dedicated network of supercomputers that are connected and scheduled with I-WAY. Each code uses PVM (parallel virtual machine) as the message-passing layer so that the heterogeneity of the computers becomes transparent. These computationally intensive codes require hundreds to thousands of processors to achieve acceptable performance. The I-WAY testbed provides access to the needed resources by allowing a researcher to use a virtual I-WAY machine (VIM), which is a collection of supercomputing, visualization, and network resources from around the country. The authors slightly modified the PVM to operate within the I-WAY scheduling and authentication mechanisms and present the user with a familiar interface to run existing programs on this new testbed. New challenges are presented when attempting to write efficient codes for geographically distributed computers. The applications presented here were successfully ported to the current I-WAY environment.