Predicting the workload of a computer system

A technique based on clustering is described for characterizing the current workload of a computer system and for predicting future workload. The technique was applied to predicting the load for an installation operating two machines processing about 20,000 job steps per month.