Cluster Overview

NJIT is in the process of standardizing on groups ("clusters")  of commodity computers as the vehicle for providing high performance  computing services for researchers. These computers (nodes) can act  independently, or in parallel, to handle large-scale, usually  floating-point intensive, computationally demanding tasks.

Clusters consist of a master (head) node, and slave nodes. The master node  provides various management and software distribution functions; the slave  nodes perform the computations.

Clusters are especially effective when computational tasks can be divided  into sub-tasks that can be performed independently and simultaneously  (parallelization). Parallel computing requires fast communication between  processes running on the slave nodes; this communication is enabled by  a Message Passing Interface (MPI).

The user interacts with the cluster via scheduling and resource  management software. This software can perform these same functions  for groups of clusters (or other computational elements).

Scheduling and allocation of resources are determined by policies that  are dependent on various factors, including group and individual ownership  of nodes in a cluster.

There are currently three clusters in operation at NJIT : cappl.njit.edu,  hydra.njit.edu, and kong.njit.edu. All are managed by IST University Computing  Systems.

Cluster Software

All of the standard Unix/Linux compilers, programs and utilities are  available on all clusters. In addition, various special-purpose software  is available.

Requests for additional software for use on clusters should be sent to  ucssys@njit.edu. Free software can usually be installed within 14 days of the  request. Requests for commercial software should identify the funding source  for that software.

Access to Clusters

Access to two of the clusters, cappl.njit.edu and hydra.njit.edu,  is determined by the groups that have funded those machines  (cappl : Electrical and Computer Engineering; hydra : Department of  Mathematical Sciences).

Researchers wishing to access either cappl or hydra should contact  the following :

cappl : Dr. Jie Hu, jie.hu@njit.edu

hydra : Dr. Michael Siegel, michael.s.siegel@njit.edu

Researchers wishing to access kong.njit.edu should send mail to  that effect to ucssys@njit.edu in order to initiate the access process.

Cluster Software