Supercomputer centres need to redefine themselves

The first route is not very interesting. The second is only possible for a small number of centres in Europe. Hence the third route is the best choice for most centres. Then it is the return on invetment for the customer that counts, not the bare performance. This third route is also the direction in which PAC develops.

PAC has much expertise in parallel processing and in metacomputing. The latter is seen by PAC as the most important development that can deliver targeted application performance in a cost effective way to many users.

For many applications tightly coupled MPPs or stand alone work stations are not the right solution. Upstill sees a lot of development on metacomputing software. In some cases, as for instance when high-throughput and good load balance is needed, commercial load sharing software, as LSF or Codine, already provides a good solution.

As an example of multi-national metacomputing project he mentioned Promenvir, in which 50 Gflop/s of computing capacity at 8 sites was linked for a crash simulation.

Issues that could slow down the development of metacomputing applications are firewalls & security, Internet reliability, application models and availability of installed software. More and more firewalls and security measures forbid access to machines from outside a certain domain, making it difficult to built a distributed metacomputer.

Software licensing today is based on the model that you pay for the availability of software, whether it is used or not. According to Upstill, this should be replaced by a pay-per-use-software model. PAC is developing flexible licensing software that can take into account local and global resources.

For more information check in at the Parallel Applications Centre web site.