Using Condor

Our main source of spare compute capacity comprises the PC cluster system running Microsoft Windows® 7. Condor jobs can be compiled (or cross-compiled) for this platform or make use of existing installed software (e.g. MATLAB® or R). In addition certain "well behaved" executables can be down-loaded and run as a scripted job. Jobs must only communicate through files or the network, any attempt by programs to access the keyboard, video or mouse will cause jobs to fail. Since cluster PCs are re-booted at 5am every day your jobs must be arranged to complete before that time.

A Condor job (or cluster of jobs) is submitted using a set of commands listed in a file, the jobs are then queued up and when matched with an available computer they will be sent to execute there. Jobs may run at any time and in any order, and even jobs that have started to run may be preempted (e.g. by an interactive user or system update), killed off and re-entered in the queue to run again later. Condor will preferentially send jobs to the most power effective compute resource, favouring machines that are already powered up, and will wake up computers to process jobs only if necessary. Note that usage of the Condor system is monitored.

Condor jobs are submitted and managed on condor1.ncl.ac.uk. This is a Unix system and requires Unix access to be added to your computer user account: contact the IT Service Desk to request this.

Note that to access condor1 from off-campus you will need to make a secure shell connection via the UNIX time-sharing service.