next up previous
Next: Operating Mode Up: No Title Previous: Introduction

Design Decisions

Our protocol and implementation is designed to perform well for a certain hardware configuration and operating mode which we think will be typical for high-performance computing. Applications insensitive to latency, or requiring little communication in the first place, already run well on clusters of workstations communicating via MPI on TCP/IP. We are aiming more at applications that need low latencies and high bandwidths. Those typically require parallel supercomputers with fast custom-made interconnects.





Bernd Pfrommer
Mon May 26 12:18:25 PDT 1997