[torqueusers] Historical workload traces
michaeljack at trentu.ca
Thu Jan 5 16:59:08 MST 2006
My name is Michael Jack. I am an NSERC graduate student at Trent University (Ontario, Canada) in the Modeling and Simulation program. I am currently working in the area of parallel job scheduling under the supervision of Dr. Brian G. Patrick.
Part of our research involves the collection and analysis of workload data on high performance, multiprocessor systems. This data is used to model as realistically as possible, important job factors such as size, memory requirements, minimum and maximum number of processors (workstations) required for execution, number of threads, malleability, degree of parallelism, time-sensitivity etc.
Most parallel systems maintain accounting logs for administrative use. These logs contain valuable information about the job activity on all machines, and in particular, about the attributes of each job. The format of the logs is typically an ASCII file with one line per job (although some systems maintain a much more detailed log). Analyzing such logs can lead to important insights into the workload.
I have gotten some workload traces from www.clusterresources.com but they are pretty outdated and deprecated. It is understandable that most sites consider the traces to be somewhat confidential. However, I have been recommended by support department at www.clusterresources.com to contact you and see if you happen to have any historical workload traces of the jobs run on your system and if it is at all possible for us to obtain them for research purposes only. We could then use the data as a reference in designing workload models that would be used to drive the evaluation of our models of parallel job schedulers.
Your help is greatly appreciated.
More information about the torqueusers