[an error occurred while processing this directive]

NERSC 3 Greenbook

Next: Recommendations Up: Major Findings Previous: Critical Problems and Challenges

Driving Factors

Given the challenges identified above, a number of critical factors have been determined that are contributing to them. These must be addressed by the HPC community in general and those involved in advancing the capability and utility of parallel systems software in particular. Among those issues of most concern are:

1.

The HPC market has been too small for commercial viability to justify ISV R&D investment in sophisticated system software and tools or in large parallel applications packages. If anything, this condition has been aggravated by the functional disparities among HPC vendor offerings.

2.

The multiplicity of architectural models for HPC has been a significant source of difficulty in parallel programming and resource management. System software has been unable to fully compensate for these shortcomings and provide the user with the necessary tools for efficient programming or the means of conveniently achieving efficient execution. One possible exception is SIMD parallel architectures, at least for a limited range of applications.

3.

Tools generally provide inadequate functionality.

4.

Experimental tools have low reliability.

5.

Tools to aid parallel programming often work for small numbers of processors but can not effectively manage the context of a very large number.

6.

Tools of the same type are frequently inconsistent across platforms which hinders acceptance by ISVs and other end-user application developers.

7.

System software and tools for the same platform are often not interoperable.

8.

The path from research experiment to commercial product varies dramatically depending on nature of the tool, community demand, and ease of integration into diverse contexts. No single model of prototyping and commercialization can support and encourage this process (e.g.,. Linux, Mosaic, HPF, Pablo).

9.

HPC system market growth will be driven by commercial applications that can be enabled but not forced by availability of software tools. A successful example is the oil and gas industry who has applied HPC technology to exploratory geophysics through seismic analysis among other means.

10.

HPC hardware and software systems must be positioned as high end of a continuum of parallel system capability from workstations, through SMPs, to scalable HPCs with common programming, environment, and application framework.

11.

Rapid HPCC system software development, evolution, and deployment is critical to the health and aspirations of the HPCC community and the objectives it is addressing.

12.

A key gap in the funding of technology transfer is that agencies sponsor research and prototypes but it is unclear what source of support is available to engage the next step towards broad dissemination of robust systems or even commercialization.

NERSC 3 Greenbook

Next: Recommendations Up: Major Findings Previous: Critical Problems and Challenges