In January of 2002, the high-performance computing cluster was
built and put into operation in the Institute for Condensed Matter
Physics of NAS of Ukraine. Since that time the cluster was already upgraded for several
times, and due to this it remains as one of the powerful clusters in
Ukraine.
After the last upgrade in december of 2006, the cluster of
ICMP was extended by 17 new nodes of dual-core 64-bits processors (Xeon
5130) and RAM of 4 Gb. Due to this the test performance of the cluster
increased by 310 GFlops. At the same time the theoretical peak performance of the whole cluster is estimated as 680 GFlops.
In this moment the cluster of ICMP consits of two parts. One
part of the cluster is built of 13 nodes based on Athlon MP 2200+ and 6
nodes based on Opteron 246 interconnected by by Gigabit Ethernet.
Another part of the cluster is built of 17 nodes of Xeon 5130
interconnected by Infiniband 4x SDR. All nodes of the ICMP Cluster are dual, i.e. each of them contains two
corresponding CPUs. Therefore, the ICMP cluster is composed of 72 CPUs totally.
Except the PGI compillers, all computers within the ICMP cluster are running the freely available and open
source operating system Linux based on
Red Hat 9 òà
CentOS 4.0-4.4
distributions. Also the job management system
OpenPBS is used to
distribute the cluster resources and to control job queues.
Parallel capabilities
are achieved through the use of MPICH libraries – a portable implementation of
MPI. In addition to the PGI compilers, the usual GNU compilers are avaible.