Implementation of A Condor Pool At The University Of Huddersfield That Conforms To A Green IT Policy

The University of Huddersfield has a large number of computers laboratories on campus that are used to capacity during timetabled session but out of these sessions the machine can be unused for long periods of time. Whilst these machines have power management software installed to reduce the universities electricity bill. The idle machines could be used to perform complex calculations and simulations to benefit the research community through cycles stealing techniques and High Throughput Computing (HTC) middlewares.

In order to provide a suitable HTC service an investigation was undertaken into what middlewares are available and how they compared against each other. This study also looked at what is green IT and how the chosen HTC middleware has been adapted to conform. The investigation also involved looking into publication from other universities to see how it is used.

A survey was conducted into how useful a Condor HTC grid could fit in with other universities High Performance Computing (HPC) clusters and how beneficial Condor is. The survey also looked at how Condor is funded and how it is administered. Overall the results show that Condor is an extremely useful and low cost HTC solution.

Condor has been deployed within Canalside East within the University of Huddersfield as a test bed with plans to expand the Condor pool across campus. To help Condor fit within the green IT policy the compute nodes were configured to allow the machines to go into a low power state when required.

To be able to prevent the possibility of having a large job queue with very few nodes online a number of scripts were created that would collect the information required to remotely wake machines up using Wake on LAN (WoL). The scripts will wake machine when a number of jobs are idle and there are machine available that are offline.

In order to make Condor able to run programs that have been developed for Windows and Linux a dual Condor client system has be implemented. This has been achieved by using the standard Windows client and a virtualized Linux client with Condor on called Pools of Virtual Boxes (PoVB). These clients run as a Windows service that can be remotely switched on and off when required remotely within the same script that can wake machines when required.