Cluster for Parallel Computing


The heart of the computing system is the cluster Zeus. It is a multicomputer architecture in Beowulf configuration that can be used to perform parallel computations. It consists of a server node and 11 clients connected together using Ethernet connections. The operating system is Red Hat Enterprise 5 and uses the Message Parsing Interface (MPI) libraries to execute parallel codes. The complete control of the entire cluster is delegated at the server node that provides all kind of resource to client nodes. It’s also the role of the system console and connecting point (gateway) with the surrounding world. The jobs management is entrusted to TORQUE resource manager that implements all necessary functions for manipulation and job execution in a batch system, including the utilities for finding information on the system state and for same jobs execution. In order for the resource manager has full access to system resources that administers, it is quite clear that its program should be distributed somehow in the system. TORQUE is constantly in execution as daemon on all computing nodes and central manager node of the system.