Understanding how your cluster behaves is fine, but predicting its behavior is even better. When jobs are submitted on the cluster, some of them cannot terminate on time and results are not obtained. For profitability and production reasons, it is necessary to ensure that all submitted computations will end-up correctly.
That is why we have developed Predict-IT. By analyzing the job submission historical data of your cluster, Predict-IT identifies which jobs are likely to end-up in failure and also provides predictions on the time you need to specify for the jobs to end correctly (runtime).
Configured specifically for your cluster, Predict-IT improves through time: it adapts to your HPC environment by learning from your cluster and new jobs, each time becoming more and more accurate in its predictions. The more data you can feed it, the more efficient it will be.