The Second Pasadena Workshop on System Software and Tools for High Performance Computing Environments was an important step in crystallizing the issues governing the direction of HPC software technology. Its sharp focus on the realities dictating practical constraints was a valuable achievement in itself and produced a high confidence in the workshop findings. While the recommendations may be difficult to accomplish, even if the implementation plan is followed, there is reasonable expectation that its path is valid.
The timing of this workshop was important both in its relation to the HPCC program and with respect to industry directions. The reemergence of shared memory parallel systems such as the CRI T3D and cache coherent systems such as the SGI Power Challenge and Convex Exemplar has shifted the path anticipated for HPC at the beginning of the program. The impact of shared memory in its many forms on the evolution of system software is significant, although there are divergent opinions on the specific ramifications. The juxtaposition of successes in achieving high sustained performance for certain applications on large distributed memory systems on the one hand, and apparent difficulties in programming them on the other provides new understanding about the opportunities and challenges confronting the HPC community, not available at the beginning of the HPCC program. Finally, since the first workshop, the community has benefited from additional workshops in applications and petaflops scaled computing that has contributed to increased understanding of system scalability issues. With these advances since the first workshop, it was clearly time to engage in a major reassessment of the state and direction of HPC system software.
One of the realities dealt with by this ensemble of colleagues was the difficulty in harnessing resources. Simply put, this is because colleagues in all venues of the HPC community are over committed. With shrinking budgets and rapid shifts in priorities resulting in increased planning responsibilities, it is certain that the stress factors will only increase, at least in the near future. The community must find a better way of tasking itself towards common goals. Time constants are too short to permit long lead times typical of top-down driven projects. The ``Circle of Excellence'' model is one approach to gaining maximum benefit from existing capabilities with only small changes. Another component of the solution is to augment the capabilities of government leadership in guiding initiatives by moving the tactical direction of such initiatives closer to the technical contributors. Overall objectives, approach, and project evaluation must remain within the corridors of government, but, if left in the critical path, contention for those conduits of high level decision can become barriers to rather than channels for accomplishment. A means of delegating as much responsibility to the community as can be safely done should be found to expedite and facilitate successful accomplishment of designated initiatives.
If there was a single type of achievement of this workshop, it should be described as clarification. Few final answers were developed. Rather the major result of the workshop was an essential advance in understanding of the dynamics determining the progression in the HPC field and the sharing of that understanding to establish a shared consensus across the community. In a way, it set a new starting point from which to consider the next steps; steps that will lead not only in a somewhat different direction but on more than one course around the obstacles obscuring the final objective. Having pointed to the next mile posts, it is up to our community to make those first steps.