8.5 Sharing the technologies

8.5 Sharing the technologies

WebSphere clustering works well in scaling for performance in conjunction with the advanced power virtualization features of shared CPU partitions, partition load management and work load management, as described.
All of these workload management mechanisms have two things in common:

The workload and resource management mechanisms only take effect when there is a shortage of system resources and there is competition for these resources. This means that they will only shift resources when the system resources are insufficient to meet the requirements of the workload and there are two or more active workloads. When the system is normally loaded or when there is only one resource consumer, then the resource managers do not have a significant effect even if they intervene.

All workload and resource management mechanisms require an explicit policy describing the (relative) priorities of the managed components, which means identifying those components that will be penalized and those that will be given an advantage whenever there is a resource shortage.

WLM manages resource usage conflicts within a partition. If the partition is known to have sufficient resources for the workload, or there is only one application running in the partition, there will be no competition for resources and WLM will have no role. WLM is most often used when consolidating several different applications on a single AIX 5L server or partition.
Shared-processor partitions and the POWER Hypervisor can move CPU resources almost instantaneously from one partition to another. Shared processor partitions are appropriate when there is marked, short-term fluctuation in the workload when consolidation in to a single AIX 5L partition is inappropriate.
Shared processor partitions can be used when a partition requires a fractional part of a POWER5 processor (for example, two partitions, each with an entitlement of 1.5 running on three POWER5 processors in a shared pool).
Partition Load Manager has a relatively long latency and cannot manage workload peaks that are of short duration. PLM manages the medium-term and long-term trends; it can handle the necessary migration of resources as operations move from the daily transactions to the overnight batch and back again.
PLM can be used to optimally configure servers with stable workloads. By setting the initial partition configuration with minimal resources, leaving unassigned resources in the free pool (memory and processor), PLM will move just enough resources (assuming that they are available) in to each partition to satisfy the workload. This alleviates the need to perform any precise estimation regarding the distribution of the server hardware between partitions.
This chapter details suggestions for AIX WLM with WebSphere, in a specific scenario (where there is a desire to share LPARs among a number of WebSphere applications). For more information about combining the forms of workload management, refer to Introduction to pSeries Provisioning, SG24-6389, and PowerVM Virtualization on IBM System p: Introduction and Configuration Fourth Edition, SG24-7940.