Original URL: https://www.theregister.com/2008/10/09/bull_does_hpc/

Bull juices supers with German engineering

Europe and indigenous HPC

By Timothy Prickett Morgan

Posted in OSes, 9th October 2008 18:40 GMT

French server maker Bull made some noise earlier this week by shaking hands with a ginormous German supercomputing lab, and now, the company has acquired German HPC expert Science + Computing AG.

S+C is based in Stuttgart and has offices in Munich, Berlin, Tuebingen, and Dusseldorf. Its 200 techies and 70 other personnel help build and support HPC clusters based on Linux, Windows, or Unix on behalf of companies in the German automotive and aeronautical industries. (The company counts Audi, BMW, Bosch, Daimler, Porsche, and Infineon Technologies as its key customers).

While the company is small - recording €26m in sales in fiscal 2008 ended in June - it should give Bull a better chance of doing HPC deals outside of its home French market and against stiffer competition.

Supercomputing is and will always be a very nationalistic endeavor, and that is why multinational partnerships are so important. It diffuses the political rhetoric - somewhat. Bull says that it has installed more than 100 supercomputers in the past three years in over 15 countries. And you can expect Bull to play up the fact that it is the only major indigenous supercomputer supplier within the European Union.

IBM, Hewlett-Packard, Sun Microsystems, Fujitsu-Siemens, Dell, Cray, and Silicon Graphics are all located elsewhere. (Well, Fujitsu-Siemens is kinda half European - even though Fujitsu seems to do most of the systems work - but maybe not for long).

The S+C acquisition comes on the heels of Bull's announcement that it is the prime contractor on a 200 teraflops supercomputer being installed at Forschungszentrum Jülich, a government-sponsored research center located in the Germany that has some of the most powerful supers in the world. This is the most recent win, and one of the largest

Juropa, Juropa

In that deal, Bull is using a mix of its own NovaScale x64 rack servers as well as x64 blade servers from Sun Microsystems to create the parallel machine, which is named Juropa (short for Jülich Research on Petaflops Architectures). This 200 teraflops machine will be based on Intel's eight-core "Nehalem" processor, due next year, and the Juropa machine is also slated for an "early 2009" deliver.

Sun is also kicking in its Zettabyte File System for Solaris 10 and Linux and the Lustre cluster file system, an open source file system for HPC clusters that Sun took control of last September. ParTec, a Germany software company that specializes in HPC cluster management, is also in on the Juropa deal.

The interesting bit is that IBM had the contract to build the prototype Juropa box back in June 2007, using System x servers using dual-core Xeon chips from Intel and interconnect from Quadrics. But this time around, Bull prevailed.

While Bull did not give specifics of the Juropa machine, a source familiar with the setup coming to Forschungszentrum Jülich says that it will be comprised of 20 NovaScale servers, which will be used as head nodes and other management nodes, while the compute nodes will be made up of over 2,000 Sun blade servers. If the future Sun blades are two-socket servers, that works out to over 32,000 cores in the compute nodes.

Financial terms of the S+C acquisition were not disclosed, and Bull did not say that the company's experts would be involved in the Juropa installation, but this seems likely.

To date, the two largest publicly known supercomputers built by Bull are in France, and they are based on its Itanium-based NovaScale servers. Both are at the Commissariat a l'Energie Atomique. One has 9,968 cores, using Quadrics interconnect, and it's rated at 52.8 teraflops of sustained performance. The other has 7,680 cores, using InfiniBand interconnect, and it's rated at 42.1 teraflops. Cardiff University also has an InfiniBand cluster made by Bull using 2,048 Xeon cores. It's rated at 20 teraflops. ®