High Performance Computing at Deutsche Telekom AGFrankfurt 11 October 2000 At a press meeting, Willi Seemke from the Telekom daughter DeTeCSM (Deutsche Telekom Computer Service Management) GmbH received a model of an IBM SP computer for their 100th SP system and disclosed the usage for such a lot of supercomputers. Michael Teyssedre, IBM director of RS/6000 Server Sales Worldwide, discussed new developments and perspectives in their high performance computer area, which are discussed in an other article of this issue.
Willi Seemke presented the high performance computing usage. The headquarter is in Darmstadt, 6 service and computer centers are distributed all over Germany, one IT network center, 78 on-site support service centers with about 6200 employees. Their key figures, an installation base with a capacity for 240 000 PC workplaces, integrated office communications package for 137 000 user (end of 2000 about 150 000), integrated services for over 450 different IT applications, operating platforms OS/390, Unix and Windows NT, 3 200 UNIX servers, 177 000 PCs, 135 000 printers and 30 000 terminals, 28 700 MIPS CPU and 200 TeraByte MVS disk space and 110 TeraByte server disc memory. The challenge is to manage this very large, very complex, high volume, leading edge, multivendor and missin critical IT infrastructure. They developed two concepts to make the operation as efficient and economomical as possible:
This means to analyse the existing system standards and systems (hardware, software, tools, procedures) using IBM's server consolidation methods. To work out and present alternative solutions, to 'put together' system packages which cover a defined, strictly limited nurnber of modules to simplify production sequences and to provide optimum last level support, to develop recommendations for pilot configurations, and in particular, the server consolidation approach should be consistently pursued for new applications and systems. The consolidation process for the mainframes started in 1992 with about 110 computer centers and was finished in 1996 with the 6 centers. This is followed by the servers. IBM RS/6000 Supercomputers at Deutsche Telekom The supercomputers are used in for commercial applications, that means in an other context. IBM is one of the major UNIX partners of Deutsche Telekom AG. The 100 systems belonging to the IBM RS/6000 family are used successfully in many projects:
In total the 100 IBM SP supercomputers have about 1100 nodes with 2 to 4 CPUs and 2 to 4 GB RAM. Thus one can count with approximately 3300 CPUs. In the June 2000 Top500 list DeTeCSM has seven IBM SP with an aggregated peak performance of 1.2 TFlop/s using 1600 processors. Thus one can estimate a peak performance of supercomputers of more than 2.5 TFlop/s. All the 100 machines are operated by two people from one central point. DeTeCSM chose the IBM machines because of the high performance, they need in commercial applications. The other reason is the single point of administration. Other issues are availability, reliability, and scalability. But IBM is not the only vendor, DeTeCSM realises a multi-vendor strategy. The IBM SP plays an important role in the server consolidation task, about 1000 servers have been consolidated.
Uwe Harms |
 |