A few key concepts should be understood before diving into the example WLM scenarios. These terms are used throughout the remainder of this chapter and throughout the WLM product documentation.
Service-Level Objectives: measurable goals used to ensure that workloads receive the necessary resources. SLOs specify which partition the workload is running in, a goal that defines how WLM should determine when to reallocate resources, a priority that indicates the relative priority of the workload, and optional conditions. These conditions further customize the SLOs by, indicating what circumstances should be in effect in order for the SLO to apply. Conditions can be the time of day, day of week, day of month or any event that can be detected on the system, such as a Serviceguard failover. Each workload must have at least one SLO.
Priority: the relative importance of an SLO. Each SLO has an associated priority that is used to allocate resources. The workload with the highest priority will be allocated resources until it is able to meet its SLO. Multiple workloads may have the same priority, in which case they are treated equally when WLM allocates excess resources. The use of priorities may cause lower-priority workloads to receive very little, if any, resources. Therefore, configuration of the SLO priority should be thoroughly considered, as the consequences on a low-priority workload can be dramatic when a high-priority workload is busy.
Entitlement: the minimum allocation of CPU, memory, or disk bandwidth resources a workload receives. Depending on the resource type, the entitlement may be specified in shares, in percentages, or in absolute units.
Workload: a set of processes that is monitored and managed as a group. The definition of a workload may vary depending on the technology being used to allocate resources. If Secure Resource Partitions are being used to isolate multiple workloads within a single OS instance, then it is necessary to define which processes belong to each Secure Resource Partition. This is done by identifying the executables that belong to a particular application and/or a list of users. If the workload is all the processes on an OS instance using vPars, nPars, Temporary Instant Capacity, or Pay per use to reallocate resources, then there is no need to define the details of the processes or users for the workload.
Goals: a measure of workload performance or resource utilization. Goals can be defined in a number of ways. The easiest to configure are CPU utilization goals. These require no special data about the application and can be used for any workload. Alternatively, real-time metrics such as response time, queue length, number of users connected, or number of processes can be used in a goal. For example, if response-time data is already being collected for the application, a goal can be configured that specifies the response time should stay below five seconds. If the response time goes above this value, WLM will allocate additional CPU resources until the response time drops back down below the goal.
