|
A Linux compute cluster in its simplest form is built of a number of compute nodes, i.e. standard off-the-shelf hardware, plus a head-node, all connected by a network, the cluster interconnect. As the name says, the operating system is Linux. Additional software is required to tie the individual computers together in a way that makes it possible to start and run jobs in parallel across a number of compute nodes.
Linux clusters at Risø DTU
Pingo
Our first Linux cluster, Pingo, was installed in 2003 and phased out in December 2008. It had 12 nodes, each with two processors, one login server (head node) and a file server for data storage. Pingo was installed as a Rocks cluster, an open-source Linux cluster distribution by NPACI.
The "old" workhorse is Mary, with 240 single-processor nodes, one login server (head node) and two file serves. Mary was installed in the autumn 2004. The cluster installation was done by means of the program package OSCAR (Open Source Cluster Application Resources).
In preparation for the installation of a new cluster at Risø, a test cluster named Valde has been set up. This cluster, also an OSCAR cluster, has 24 compute nodes, each with two dual-core processors, and therefore effectively four-processor nodes.
The newest Risø Linux cluster is Thyra, with 128 compute nodes, one login server and three file servers. It was installed in 2008. Like Mary and Valde, Thyra is an OSCAR cluster.
The compute nodes are blade servers, and each server has two dual core CPUs, or from a user's point of view four CPUs. This gives a total compute power of 512 CPUs.
Cluster interconnect is both InfiniBand and Gigabit Ethernet.
See also: DCSC, Danish Center for Scientific Computing.
DCSC Linux clusters:
DCSC/DTU: Niflheim + Yggdrasil — DCSC/KU: Steno — DCSC/SDU: Horseshoe
|