Cluster Computing
Cluster Computing
N. D'Imperio
Linux clusters have become important providers of computing cycles in
the scientific world. (http://www.ccd.bnl.gov/bcf/cluster/). The CSC
cluster, named Galaxy, consists of 554 Intel P3, P4 and Mac G5
processors running at speeds up to 3.2 GHz. Arranged in dual processor
configurations, installed memory is 2 Gbyte per node for the highest
performing nodes. A RAID storage array with a 10 terabyte capacity is
attached to this machine. Communication is gigabit Ethernet through a
Force10 E600 Terascale switch with 240 ports. Parallel computations are
performed using the message passing interface, MPI (www-unix.mcs.anl.gov/mpi)
and through threads using Open MP.
Our goal is to continue to explore the frontier of the commodity
component world, as processor, motherboard bus, and memory speeds
increase, as fiber cable decreases in price, and as switch speeds
increase in performance and decrease in price. We will determine optimal
configurations and architectures, within the overall framework of our
design. Results from this research will be published and posted on the
network, to benefit others seeking a similar price/performance
advantage.
The Galaxy computer provides computing cycles to CSC affiliates and
projects. Convenient and assured access to local parallel computing is
an essential requirement for the development of parallel simulation
codes, and it is thus a key component of the CSC strategy to make
state-of-the-art computing technology available to benefit BNL
scientific programs.
Last Modified: April 23, 2009
Please forward all questions about this site to:
Denise Monteleone
|
