A year later, we are glad we chose the SP1. Having a full, running Unix OS on each node allowed for easy ports. The large memory on each node provided one-node execution as a useful step in the porting process. Furthermore, the use of portability layers (Chameleon, PCN, and p4) let us port applications quickly and allowed us to become familar with the programming environment on the SP1 in the context of significant applications. We expect MPI to play this role in the future.
We were able to get a wide variety of applications running with impressive performance and with relatively little additional work, demonstrating that it is possible to write portable, efficient parallel programs.
Since the first few months, the pace of moving new applications to the machine has increased. Furthermore, many of the early applications are now in ``production'' mode, no longer functioning as research projects in computing but delivering new scientific results in their application areas. We are now beginning the next phase of our parallel computing research, involving experimentation with a scalable I/O configuration on the SP1. It will be the subject of a separate report.