Web Technologies for Pervasive High Performance Computing

Vision

NPAC has described in an article (Computing on the Web -- New Approaches to Parallel Processing -- Petaop and Exaop Performance in the Year 2007 at http://www.npac.syr.edu/users/gcf/petastuff/petaweb/ ) a vision for using Web technologies to enhance the HPCC programming environment. A modified version of this article will be published soon in the IEEE Internet Computing journal. We describe a three pronged approach with initially Web technologies used for applet based Java user interfaces; then a more general use as base software infrastructure including web frameworks for coarse grain software integration; finally we discuss the use of Java as a basic computing language using optimizing native Java compilers. We believe that HPCC can gain a lot from using the Web distributed computing model as a software base because we see an essential problem with HPCC as not a lack of good ideas but rather a lack of funds and people to build robust highly functional programming environments. The very large creative Web technology base provides the infrastructure that HPCC needs but cannot afford to build ab initio. NPAC accepted a Sun Microsystems invitation to join their new HPCC Consortium because we see that they are critical in ensuring that Java and other Web technologies are developed so that can support computational science and engineering effectively.

Project Description

We suggest that initial actively should focus on the first area described in the vision - namely the use of Web technologies for powerful interoperable user interfaces. We suggest that first year activities in this area would lead onto web based metacomputing activities in the second year. The latter could be important as a cross MSRC PET program but further planning work will be needed before such major distributed computing work can be set up. We do suggest that a planning activity in this area could be important during the coming year but we need to get a broad buy in before starting.

In the area of web based interfaces, we have demonstrated several prototypes including the use of Java graph editors for AVS/Khoros like user interfaces; 1 and 2D scientific visualization Java applets being used in our numerical relativity Grand Challenge; and the "Virtual Programming Laboratory" (VPL) which has built Java and JavaScript interfaces for both HPF and MPI ( http://www.npac.syr.edu/projects/VPL) for use in training at both Syracuse and Cornell. The latter includes performance visualization using Illinois's Pablo system which we linked to the Web. We suggest that the initial project customize the VPL for CEWES use in a similar fashion to our customization of VPL for Cornell's Virtual Workshop which aimed at training needs specific to their SP2. We would use the enhanced security developed for the VPL by Cornell as part of their collaboration with us. We see that this project naturally leads to a an interoperable web interface to the CEWES production facilities and suggest that development of training VPL interface be accompanied by a design study for a similar environment for production HPCC codes.

Deliverables

• Report on planning activity on Web technologies for cross MSRC metacomputing (long term study which must be coordinated with academic government and integrators in all 4 sites). We expect this would only be useful if NPAC was part of a team working on this broad problem.
• Report on Design issues for Interoperable Web Interface for production use of CEWES HPCC facilities
• Customization of Virtual Programming Laboratory (VPL) for use in accompanying NPAC Web based training project. It would also be useable in other CEWES training activities. This would include interoperable support for all the CEWES facilities including IBM SP2, T3E and Origin 2000 platforms. (David check this. I am not certain what HPCC facilities they have and our link to WES server is kaput today). We would also include interfaces to PC clusters - especially those with NT and Linux operating environments.
  • The four platforms above would be delivered incrementally - with all complete by the end of the year.