Modeling for computing center planning

Existing empirically founded techniques for timing individual jobs are not suitable for long-range planning. The main reasons are they are tied to an existing system structure; necessary detail on individual jobs is not available on a long-range basis; and the types of jobs are subject to considerable uncertainty. To improve this situation, we suggest a planning technique for many system alternatives. In addition, this technique should require only relatively general information about the work load at the planning horizon. The basis for the model is a probability-based macromodel of multiprogramming by Gaver.