Latent Demand Considerations in Mainframe Analytical Modeling

Analytical modeling is a capacity planning function often used to accurately project future computing resource requirements. However, many capacity planners ignore activity "outside of the box" network changes, human behavior, and changing requirements which can introduce latent demand to an upgraded system and adversely affect the results of their best planning efforts. What is latent demand? Latent demand in a mainframe system can be best explained as work that would have arrived on the system had resources been available to initiate it. Latent demand can take the form of batch work waiting to process on the input queue or additional online transactions that would have been processed but were prevented from creation due to poor response time. The effects of latent demand vary from one environment to another. In some instances only a minor revision to a capacity plan is necessary after changing the system configuration to account for latent demand of resources. This is truly a latent demand situation but when it falls within an acceptable margin of approximation in resource consumption predictions it is often ignored or even unrecognized. If major delays in processing were alleviated by the upgrade, new unanticipated work can immediately consume any newly acquired capacity. Should the capacity planner fail to account for large quantities of work that was previously enqueued either in the system, or by human factors but now arrive for execution, the plan and associated forecasts will be invalidated, and were probably too optimistic in predicting performance results. Some may look at latent demand as causing a vicious circle – capacity is acquired, it’s immediately consumed by new applications couldn’t run, and the process of planning for and acquiring new computing resources begins again. How do you identify it? If the CPU is running at full capacity, and any work is queued and waiting to be processed, it's obvious that there is latent demand. If the environment is expanded to handle more work, then it stands to reason that you will have to factor in work that is currently waitingmore throughput will be processed, but there may not be an improvement in response time. Furthermore, when demand is slowed by network anomalies or general slow response time in online systems, you must also attempt to estimate, by hand, the volume of arrivals that will be created when response time problems are alleviated. Batch scenarios, online processing, and