This paper discusses ongoing research at Oak Ridge National Laboratory (ORNL) to make computing clusters easier to use. Cluster administration, setup, and use is an active research area with many different components. Two systems for cluster control and administration, that have been experimented with at ORNL, are Managing Multiple Clusters (M3C) and Cluster Control GUI (C2G). M3C uses a Java Servlet in conjunction with a Java application server to handle communications between a remote user and the head node. C2G takes an alternate approach and uses the sshd to handle these messages. Another important issue to consider is the mechanism that is used to handle communications between compute nodes.

A new system that is under construction at ORNL is designed to allow a user to easily keep track of software that is loaded on a node. This system has two components, a node manager daemon and a package services back-end that is basically a database. Multiple software configurations for a compute node can be stored and loaded on a node with this system.

...

Download PDF.