Hypercomputing and supercomputing in Germany

Mannheim, 21-6-97 In Germany there are several distributed computing projects underway. Two different appproaches are hypercomputing with workstations while they are not in use by their owner, and metacomputing with dedicated large supercomputers. Hypercomputing, a concept for a network-based computer architecture, was presented at the Mannheim Supercomputer seminar by Professor Djamshid Tavangarian, University of Rostock. Metacomputing by Dr. Alexander Reinefeld from the Paderborn Center for Parallel Computing.

Hypercomputing

The Rostoch project wants to connect different workstation clusters via Internet to use free capacity on the workstations for solving a big problem. These workstations are locally distributed. Tavangarian wants to offer an open system that can be used by everybody (who gets the allowance) at any time at any location.

At this moment the universities at Kiel, L¨beck, Rostock, Berlin, Hannover, Magdeburg, Frankfurt, Mannheim and Stuttgart and some others are involved. Tavangarian mentioned that at Rostock there are about 200 workstations available, if one multiplies this with 40 or 50 universities in Germany, this means a big amount of computing power.

On the industrial side Codine from Genias, Neutraubling, and LSF (Load Sharing Facility from Platform Computing) and its distributor science+computing, Tübingen, are involved with their load distribution software.

The major issues in this project are:

communication and response time behaviour of the system
hierachical structuring of the system
process management including migration of applications and checkpointing
management of the hypercomputer
security in heterogeneous environments
accounting

The results of this project will be presented at ARCS '97 (Architectur of computing systems) 8. - 11. September 1997 in Rostock.

Metacomputing

This project is realised in Northrhine-Westfalia and aims at coordinating the use of a pool of geographically distributed high-performance computers. The user's view of an ideal metacomputer is that of a powerful monolithic virtual machine: he has an homogeneous view.

Alexander Reinefeld listed some problems to be solved in this area, the computer centres donot want to give up their local autonomy, there are security issues, distributed program development (debugging, performance monitoring) and a single image of the system.

Their target group of the metacomputing envirnoment does not include program development, no short runs and no real-time applications, but registered users with high computing requirements.

According to Reinefeld the key components of a metacomputer are:

graphical user interface (HTML, Java)
central access point (collects incoming requests, authorization)
configuration database
request analyser
dynamic job distributor
distributed load monitor daemons
cluster management software

Security and confidentiality aspects are very important: one needs regular and high-security levels. Reinefeld showed two industrial applications that benefit from the metacomputer environment: a distributed pharmaceutical applications server and a model for an industrial CFD applications server - both are ESPRIT projects.

Uwe Harms