Workload Manager
| Industry | Computer software |
|---|---|
| Products | Workload management for IBM mainframe and AIX operating systems |
| Parent | IBM |
In IBM mainframe environments, Workload Manager (WLM) serves as an essential feature within MVS/ESA and its successors up to z/OS. It governs the allocation of system resources for workloads on z/OS based on objectives set by administrators. IBM also provides versions of Workload Manager tailored for other operating systems, including AIX operating system.
Workload Manager
Mainframe systems execute numerous applications concurrently, necessitating consistent processing times and reliable database access. On z/OS, WLM fulfills these requirements by regulating resource availability in accordance with externally defined administrative policies.
Administrators categorize workloads into service classes using criteria such as transaction identifiers, user credentials, or program names. For each class, they establish goals alongside importance ratings, defining expected performance levels which may be specified as response time targets, relative processing speed (known as velocity), or marked discretionary if no explicit goal exists. Response time refers to the interval from when a request enters the system until the application signals completion.
When applications do not provide response time information, WLM relies on execution velocity—a measure calculated as the proportion of time actively using resources versus waiting in delay states—to assess workload performance independently:
| Execution Velocity Defined |
|---|
| <math>\text{Execution Velocity}=100\cdot\frac{\text{Total Using Samples}}{\text{Total Using Samples}+\text{Total Delay Samples}}</math> |
To determine resource prioritization during periods of high demand, administrators assign importance levels to service classes. These classes and their associated goals compose service policies stored in service definitions accessible by all z/OS instances within an Parallel Sysplex cluster, thereby maintaining consistent performance objectives across the environment.
WLM functions as a closed-loop control mechanism that continuously gathers data on workloads and resources, compares actual performance against targets, and modifies resource allocation when goals are not achieved. This evaluation occurs at regular intervals, computing a performance index defined as follows:
| Performance Index Formulas |
|---|
| <math>\text{Response Time: }PI=\frac{\text{Actual Achieved Response Time}}{\text{Response Time Goal}}</math>
|
This performance index reveals whether targets are being met or missed. Utilizing this metric alongside importance values, WLM predicts necessary adjustments and implements changes aimed at optimizing workload efficiency. Decisions are based on data collected over intervals ranging from 20 seconds up to 20 minutes to ensure statistical significance, with only one adjustment applied per interval to maintain controlled system behavior.
Resource management under WLM encompasses processor cycles, I/O devices, memory allocation, and control over process initiation and termination. Processor access is managed through dispatch priorities assigned per service class; these priorities are dynamically modified according to goal attainment, current system load, and demand levels. Comparable dynamic controls regulate other resource types. This goal-driven workload management approach contrasts with entitlement-based models often found in larger UNIX systems that rely on fixed resource allocations.
A distinguishing feature of z/OS WLM is its deep integration with key applications, middleware layers, and subsystems. It offers interfaces allowing subsystems to notify WLM upon work unit start and completion events while providing classification details for precise workload handling. Furthermore, WLM facilitates load distribution across Parallel Sysplex clusters and supplies instrumentation enabling database and resource managers to report contention issues. This capability allows WLM to prioritize lock holders effectively, enhancing overall throughput.
Over successive iterations, z/OS WLM has become the primary component governing performance control within the operating system. In Parallel Sysplex configurations, WLM components coordinate to deliver a cohesive performance perspective throughout the cluster. On System z platforms featuring multiple virtual partitions, WLM collaborates with the LPAR Hypervisor to influence partition weighting and allocate CPU capacity among logical partitions accordingly.
Literature
- Paola Bari and colleagues: Guide for System Programmers on Workload Management. IBM Redbook, SG24-6472
External links
See also
- Unit Control Block, explaining WLM's approach to handling dynamic Parallel Access Volumes (PAVs)