CLiC - Chemnitz Linux Cluster
Heidelberg 23 June 2001 CLiC is a large Beowulf-style cluster, which was installed in autumn 2000 at the Chemnitz University of Technology. It is built from 528 off the shelf commodity PC nodes running Linux, plus some server and management nodes.
Professor Uwe Huebner gave an overview of the current system. The University was supported by the Chemnitz based company Megware computer, which was found in the exhibition of Supercomputer 2001. The system consists of 528 Pentium III, 800 MHz, with 512 MB RAM and 30 GB hard disk as computing nodes. This sums up to 264 GB memory and 15 TByte disk space.
The nodes are connected by a (non-blocking) "production network" and a "service network", both built with 100/1000 Mbit/s Ethernet technology. The service network is used to switch on and off the single nodes, as it cannot be done for all in one moment. Typical applications of this cluster are modelling with finite elements, fluid dynamics, problems in thermodynamics and elasticity. One of the biggest obstacles was the fact that there is no "ready to run" solution for the various aspects of management of such a huge system. Other critical aspects are parallelization and communication support for the applications.
Huebner commented that they did not use SCI or Myrinet, as these technologies are rather expensive. He liked to buy more nodes.
As all the application software is self-made, the users can directly optimize their codes. The system is divided into 2 to 32 nodes for development and testing of parallel algorithms and 64 to 324 for production. The full system for one job can only be started by "hand".
He mentioned that the cost of network to nodes comes to 2 : 1. The total cost of the system of 1.25 Euro has to be compared to the cost of the systems that rank in the same range, e.g. a Fujitsu VPP700 with 48 processors at ECMWF.
More information about the communication:
http://www.tu-chemnitz.de/~huebner/cliccom
Uwe Harms
[News on Advanced IT][Calendar][Analysis][IT in Medicine]
|