xMPI -- An MPI Implementation over x-Kernel for ATM Networks

High-speed computing experiments using networked multi-computers demonstrate the potential and viability of low-cost alternatives to special purpose parallel computing systems. However, to realize this goal, the application environment must provide support for parallel applications, high-performance computation, low-latency, and highbandwidth communication. To maximize performance, the software systems must be optimized for parallel applications, exploiting the characteristics of the hardware and network. In this position paper, we describe the design philosophy of xMPI , an implementation of MPI based on the x-kernel that is adapted for ATM networks. X-kernel is an embeddable network subsystem which provides a flexible protocol composition framework[3]. The x-kernel abstractions viz. messages, protocol and session objects, events and the uniform protocol interface allow customization of the operating system services