Cluster Administration Tool

Rolf Spuler

Semester Thesis Summer 1998
Supervisors: F. Rauch, Prof. T. Stricker
Institute for Computer Systems, ETH Zürich

Objectives

There are usually many users on our Cluster of PCs (CoPs). If one would like to know which machines are currently in use, a combination of some shell-commands like finger, rsh/ssh, rusers need to be started. Sometimes machines are dying or down, so more commands like ping come into play. As there are different operating systems on the cluster, it gets even more difficult to find out what's exactly going on.

The goal of this thesis was to ease this task with a tool, so that one always has an overview over the cluster and can administrate the machines more easily.

Results

The final Cluster Administration Tool supports Linux and Windows NT and consists mainly of two parts: The server daemons gather information on the machine on which they are running, such as: Reachability, operating system currently running, users loged in and their idle-times, load. To do so they use the scripting capabilities of Tcl/Tk on Linux as well as a few very small C-programs which handle requests to the OS which could not easily be done in the scripting language. The daemons can easily be ported to other UNIX operating systems. On Windows NT the daemons are completely written in C as this fits better into the system.

The client is a frontend running on Linux. It connects to the server daemons with TCP and communicates with them using a simple text-based protocol. The returned results are then grafically displayed.

Additional functionalities include the possibility to reboot a remote machine to a specified operating system, to copy files to a number of selected machines, to watch remote log-files and to simply log into a machine by clicking on its name.

[ CS-Department | Up ]
ETH Zürich: Department of Computer Science
Comments to Felix Rauch <rauch@inf.ethz.ch>
26. July 1999