Estimating CPU Latent Demand

Mark Cohen Austrowiek – Fabio Massimo Ottavini
May 2002

When all tuning actions have been done and the bottleneck is the fastest device (CPU) the problem seems to be fairly simple: you have to buy more or bigger processors.How many and how big are relatively easy questions to answer. In fact there are a lot of tools and techniques you can use to estimate the processing power you need.

But in some particular situations things become a little more complicated.Performance analysts know that, if the CPU utilisation is at 100% for long periods of time, there is a resource demand in the system, hidden in the CPU queue, that needs to be estimated. This demand is normally named “CPU latent demand”.

In these situations, using only the “normal” tools and techniques, the risk to plan a CPU constrained system can be very high.

This paper describes a technique, based on CPU utilization and queue metrics, you can use to estimate the CPU latent demand for OS/390 and Z-OS operative systems.

This technique will also be applied to the Unix and Windows platforms.