Subhasis Mahapatra,
Physical Research Laboratory,
Navrangpura, India
emails: subhasis76@hotmail.com and subhasis@prl.ernet.in
In distributed computing environments we need a careful selection of computers, networks, protocols and algorithms used by an application for high performance execution. This selection and configuration require access to accurate, up-to-date on the structure and state of available resources. Here I suggest a model based on "BEOWULF Cluster" as a low budget solution for HPC as used in Institute of Physics, Bhubaneswar, India.
We have used XEON dual processor machine as computing element. Each machine has two Ethernet card belonging to two different networks. For usual login connection and so, it uses the cost effective "Fast Ethernet network" and of inter-processor communication, it uses a "Gigabit network".
The cluster is running on RedHat Linux. I suggest to use MPICH or ProMPI on the cluster for parallel execution.