[an error occurred while processing this directive]

NERSC 3 Greenbook

Next: Conclusions Up: Findings of the Second Previous: What Needs To Be

Implementation Initiatives

  This section provides action items that will go a long way toward implementing the recommendations. They involve the formation and sponsorship of a collection of committees with focus charters. Note the use of the term ``circle of excellence.'' This is a notion that achieves the same thing as a center of excellence but has the talent distributed around the country. This allows people in other organizations to be funded to collaborate on specific work. It does not require the lead time to set up a building nor is it as likely to be perceived as in competition with all other research groups. Rather it leverages talent and capabilities already established and partly redirects it in a coordinated manner to specific objectives. Through such a structure, new skills can be easily added permitting new research directions to be exploited quickly and within a given budget. Its last advantage is that it is easier to scale up and down as funding permits/constrains. Implicit in this model is a key change in the process of engaging the research community to address government specified objectives. Technical decisions would be moved out to the field, providing rapid exploitation of new opportunities, while government oversight is maintained.

• Convene an Oversight Committee on HPC System Software and Tools to develop a national software technology plan and to monitor progress, engage the HPC community, and report status on an ongoing basis. The Oversight Committee should include technical representatives from academia, government, and industry.
• Convene a standing committee on HPC software tool set functionality to determine the critical capabilities that need to be available on all parallel platforms from SMPs to HPCs. Conduct a focused workshop on this specific issue to initiate the process.
• Fund development of infrastructure and common interfaces for interoperability of HPC software tools. These efforts should reflect and respond to the findings from the tool set functionality committee. An early workshop should be conducted with strong vendor and SMP community participation to expose practical issues associated with achieving interoperability across systems and software types.
• Develop and deploy a basic tools set compliant with the functionality and interface requirements determined from the above studies.
• Identify, sponsor, and select key applications of commercial economic and national importance and sponsor ISV porting to HPC platforms. As an example, emerging area of synergy is support for massive data assimilation on very large archives through use of HPC resources. Providing access to HPC resources as services on the NII can enable this development.
• Engage the HPC community to determine minimum requirements for hardware support of system software and tools in order to provide a guideline to processor/system manufacturers as to what capabilities need to be included and to procurement agencies to aid in selection of appropriate systems.
• Create a software tools evaluation and test center.
• Support the establishment of a software tool development infrastructure.
• Develop and apply a total cost to solution metrics approach for evaluating progress in the HPC arena.
• Establish a common framework for ease of integrating and sharing computing products to enable HPC system interoperability. To this end develop transparent mechanisms for storing and handling all forms of data to be exchanged among systems and programs.
• Extend application/system evaluation and characterization studies such as JNNIE to provide detailed analysis of hardware impact on application efficiency and scalability as well as programmability and usability.
• Support rapid deployment consortia to produce prototypes/modules and related test suites that serve as an early implementation vehicle.
• Extend the charter of the NHSE (National HPCC Software Exchange) to include support for forming and administering a group representing a broad portion of the community. Empower and fund the NHSE to evaluate and robustify software.
• Create a National Distributed Heterogeneous Testbed as a vehicle for developing and evaluating system software for harnessing the combined capabilities of disparate systems types. Such a testbed should be distributed across the country and may include components at existing HPC centers.

NERSC 3 Greenbook

Next: Conclusions Up: Findings of the Second Previous: What Needs To Be