As part of the z/OS support of Workload License Charges, you can set a defined capacity limit, also called a soft cap, for the work running in a logical partition. This defined capacity limit is measured in millions of service units (MSUs) per hour. It allows for short-term spikes in the CPU usage, while managing to an overall, long-term four-hour rolling average utilization. It applies to all work running in the partition, regardless of the number of individual workloads the partition may contain.
See PR/SM Planning Guide for more information about how to set a defined capacity limit.
WLM enforces the defined capacity limit by tracking the partition's CPU usage and continually averaging it over the past four hours. Spikes higher than the defined capacity limit are possible, as long as they are offset by low points that keep the four-hour average at or below the limit. When this four-hour average goes over the defined capacity limit, then WLM caps the partition. At that point, it can use no more than the defined capacity limit, until the average drops below the limit.
At IPL, WLM defaults to a four-hour time interval that contains
no partition CPU usage. 
This means the four hour
rolling average starts with zero.
Consider the example shown in Figure 1:
This partition has a 50 MSU defined capacity limit, as shown by the grey dashed line. The solid black line is the actual MSU consumption of the partition, and the dashed black line is the four-hour rolling average utilization. At IPL (09:00), the partition’s CPU usage starts with more than 100 MSUs. The four-hour average, however, is below the defined capacity limit at that point, due to the fact that WLM is using the default four-hour interval time containing no partition CPU usage. No action is taken. Just before 11:00, the four-hour average hits the defined capacity limit. Now WLM caps the partition’s CPU usage at 50 MSUs.
WLM allows the CPU usage to remain at the defined capacity limit. Therefore, the four-hour average can continue to go up. In this example, this occurs because the low usage numbers of the 4-hour default time interval at IPL are falling off the back end of the four-hour horizon, and being replaced by the new usage numbers starting at IPL (09:00). This is a consequence of managing to the four-hour average.
In summary, WLM caps a partition's CPU usage only when the four-hour average reaches the defined capacity limit. Before that point, you may see CPU usage spikes over the limit. You may see the four-hour average go over the limit after capping, until the passage of time gradually brings the average back down to the limit.