Workload characterization for trend analysis

The methodology of analysis proposed in this paper aims at predicting the workload of a computer. This methodology consists of applying an algorithm of clustering to the workload, its jobs being identified by a pair (X,P), where X is the resource-vector of the job and P stands for the priority given to the job by the user.The hereby obtained clusters are then associated to the ai activities developed in the system and determine the influence of each ai to the overall workload. By repeating this operation at different times, either the periodicity or the monotonic changes that may occur in each activity are determined. This makes it possible to predict the evolution of the overall workload and consequently to evaluate changes to be carried out in the system.The above methodology is applied to a specific case and is illustrated in its various phases. The results obtained have validated the method. The study is still going on, with continuous periodical observations in order to update the data.