next up previous contents
Next: Government Intervention Scenarios Up: Vignettes in HPCC Previous: Aerospace Engineering

Structural Engineering

 

(Based on information supplied by Louis Komszik, MacNeal-Schwendler Corp.)

The industrial success of high performance computing requires reliable hardware developed with production considerations, industrial usage and maintenance issues in mind. Systems must be equipped with good analysis, monitoring, debugging and evaluating tools to facilitate software development. They must have reliable compiler, linker and runtime tools, such as MPI. Systems should produce high sustainable performance on production applications. The current form of HPF is not useful in structural analysis with its complex data structures.

The commercial failure of many HPC systems recently must be linked to their being architectural experiments released into the commercial world before their time. There is a difficult economic balance to be reached between funding for new HPC architectures, and resources invested to produce stable commercially useable systems.

Uptake of HPC would be helped by an organized, timely and efficient government funding program to motivate ISVs to port their software onto the new hardware, even when the end-users have not purchased those platforms yet. Current experience with the EUROPORT program shows that many ISV's are willing to port to a machine not in use at the end-users premises yet, if the cost of port and a certain amount of guaranteed revenue or at least advance customer interest can be generated.

ISVs need to have staff well trained in the state-of-the-art computer science and computational mathematics issues. While MacNeal-Schwendler has such staff, this is by no means common amongst ISVs.

ISVs are only going to convert their industrially well established software to computers which are already available or are specifically requested by the end-users (paying customers). These customers are naturally cautious however and only want to buy established software. This situation makes hardware procurement decisions difficult as the choice of system by industrial users is strongly coupled to the established software base gif. Only government intervention in some form may be able to break this deadlock.

The industrial HPC adoption process needs to start with the availablity of reliable HPC platforms; progressing through some form of Government intervention to aid the ports to new HPC systems, and ending up with the stable base of serious end-user customers.

The current process starts at the end-user organisation, who arrives at a procurement decision, followed by the ISV, who then converts the software to the procured hardware system. There is currently no practical government funding programme, although one is being established in connection with an ARPA funding for porting 5 ISV applications onto the IBM SP architecture. Finally, a hardware vendor, who is not selected by the end-user finds it difficult to fund development of the relevant ISV software for their platform, and therefore cannot break out of the cycle.

This scenario is a particular concern in the case of systems that are procured with the express objective of running some set of ISV products. It is very hard to obtain benchmark data in advance to obtain the right procurement decision.

HPC uptake also requires industrial end-users, such as FORD, General Motors, Boeing and Rockwell, to stay in the USA. These end-users are driven by their particular technical area and all they want to do is to solve their ever larger and more difficult problems even faster. Organisations like MacNeal-Schwendler has these customers today and the needs of these users ought to be the dominant driving force behind HPC developments, since it is they who can drive the commercial resource allocation to HPC.



next up previous contents
Next: Government Intervention Scenarios Up: Vignettes in HPCC Previous: Aerospace Engineering



Geoffrey Fox, Northeast Parallel Architectures Center at Syracuse University, gcf@npac.syr.edu