News

DAS4 Performs Excellent on GRAPH500

At the yearly top conference on Supercomputing in Seattle (12-18 November) it was announced that the DAS-4 cluster computer of the VU Computer Science department scores an honourable 16th place in the new Graph500 list. The DAS-4 cluster is only topped by machines that have much more CPU power and memory available, which indicates that DAS-4 is a very efficient platform for research on graphs.

Traditionally twice a year the so-called TOP500 list is published, listing the latest supercomputers worldwide, ranking them according to their speed while performing operations on very large matrices.  The TOP500 is thus
useful for benchmarking supercomputers built to excell in traditional fields of science and engineering where data is available in a highly structured form.

However, it becomes more and more urgent to support emerging scientific, web and business applications where data is much less well structured, yet the required amount of processing is at least as huge.  Examples are the domains of Medical Informatics, Social Networks, Cybersecurity, and searching the World Wide Web. To model these novel areas, in the new Graph500 list a different style benchmark is now used that stresses the ability to perform many operations on huge graphs as the ranking criterium.

The DAS-4 cluster is used for several novel research projects that use graph algorithms, including a framework for parallel graph algorithms (HipG), distributed reasoning over semantic web data (WebPIE), and distributed model checking. The cluster was built in 2010 by ClusterVision and is one of the six clusters of the Distributed ASCI Supercomputer of the ASCI research school. More information:

drs. Kees Verstoep (c.verstoep@vu.nl, 020 - 598 7765) and prof.dr.ir Henri Bal (bal@cs.vu.nl, 020 - 598 7733)

References:
- http://sc11.supercomputing.org
- http://www.cs.vu.nl/das4
- http://www.graph500.org/nov2011.html