Suitable workloads

Workloads best suited for sharing logical partitions are those that have a widely fluctuating ETR or would not fit well into the capacity of a dedicated logical partition. If a workload can use only a small portion of the capacity of a dedicated logical partition, redefine the logical partition to use shared processors to free the available capacity for use by other logical partitions.

A workload with a widely fluctuating ETR would experience peaks and valleys in its processor-utilization curve. Such fluctuations can occur over extremely short periods of time (minutes or seconds). This type of workload could take advantage of the time- and event-driven dispatching available. With event-driven dispatching, a sharing logical partition receives the resources required as needed and leaves the capacity free for other logical partitions when not needed.

When combining workloads on a processor complex by means of logical partitions, examine their average and peak requirements carefully. If the workloads fluctuate over very short intervals, the total capacity of the system must meet the sum of the average requirements for each workload. If processor utilization fluctuates over longer periods, and the peak utilization periods for these workloads occurs simultaneously, then the total capacity of the logical partitions must meet the sum of the peak requirements for each workload.

Sharing logical partitions that use event-driven dispatching are better able to maintain high transaction rates with fluctuating demand while being responsive. However, the Internal Throughput Rate (ITR) for a sharing logical partition is lower than the ITR for a dedicated logical partition.

The capability to limit CPU usage for any or all logical partitions with shared processors is provided by the PR/SM capping function. The capping function enhances PR/SM workload balancing controls for environments with a requirement that the CPU resources for a logical partition be limited. Capped logical partitions are recommended for use when CPU resources must be limited for business reasons (in accordance with a contract), or when the impact that one logical partition can have on other logical partitions needs to be limited.