Computer cluster workload analysis

Performance of computer clusters is greatly affected by a nature of the submitted workload. Early characterization of different workload types allows for scheduler fine tuning as well as predictions on the system load. Statistical analysis and visual representation of the workload data provide valuable insight to the overall system utilization and may reveal potential bottlenecks and points for improvement. In this paper we describe important cluster job features and introduce a tool for statistical analysis and manipulation of the workload data that is a part of a cluster simulation and runtime prediction system.