Intel Computing Cluster

The last year has seen a further shift of computing capacity available within the LCE (i. e., not including the LCE's share of the Clux cluster) from the original Alpha processor based architecture towards off-the-shelf Intel based computers. By now, 50% of the total installed capity, as measured by the LINPACK benchmark [1], is installed using Intel based architecture. Half of this is on a computing cluster that was installed during the first half of 2002.

This cluster, internally named acis, consists of 22 nodes (with a total of 27 cpu's) and 150 GByte of available harddisc capacity. The nodes can communicate (only) via inexpensive 100 MBit/s Ethernet network cards, resulting in a relatively large communication penalty to be paid. This makes the other clusters available at the LCE more suitable for heavily parallel applications [2].

Still, within the LCE there is a significant demand for serial and modestly parallel applications. The Intel cluster, in contrast to the other clusters available at the LCE, shares the same user base, directory structure etc. as the regular workstation computers, thereby introducing no obstacles to using it. Over the past year, the computing capacity offered by the acis cluster was utilised for approximately 95%, making it the most used computer system at the LCE.

[1] J. Dongarra, J. Bunch, C. Moler, and G. W. Stewart,
LINPACK User's Guide, SIAM (Philadelphia), 1979.

[2] A. Lukkarinen and A. Selonen, in E. Lampinen (ed.),
Annual report of the Laboratory of Computational Engineering and Research Centre for Computational Science and Engineering, Helsinki, 2001.