Working Set

Key Characteristics

• • Dynamic Nature: The working set is not static; it changes dynamically as a program executes, referencing different code paths and data areas over time.
  • Real Storage Residency: Pages within the working set are physically present in real storage, allowing the CPU to access them directly without incurring the overhead of a page fault and subsequent I/O operation to retrieve them from auxiliary storage.
  • Impact on Performance: A well-managed working set, where frequently accessed pages remain in real storage, is crucial for optimal application and system performance, reducing CPU wait time due to paging.
  • Locality of Reference: Programs exhibiting good locality of reference (accessing a small set of pages repeatedly over a short period) tend to have smaller, more stable working sets, leading to fewer page faults.
  • SRM Management: The z/OS System Resource Manager (SRM) actively monitors and manages the working sets of active address spaces, attempting to keep essential pages in real storage based on system policies and resource availability.
  • Virtual Storage Subset: It is a subset of an address space's total virtual storage that is currently mapped to and resident in real storage.

Use Cases

• • Performance Tuning: Analyzing an application's working set behavior (e.g., through RMF or SMF data) helps identify programs with poor locality or excessive memory demands, guiding optimization efforts to reduce paging.
  • System Capacity Planning: Understanding the aggregate working set requirements of critical workloads is vital for determining the optimal amount of real storage needed on a z/OS LPAR to maintain acceptable performance.
  • Troubleshooting Paging Issues: When a system experiences high page-in or page-out rates, examining the working sets of active address spaces can pinpoint the culprits consuming excessive real storage or causing page thrashing.
  • Application Design: Developers consider working set principles when designing applications, aiming to structure data and code to improve locality and minimize the active memory footprint, especially for frequently executed routines.

Related Concepts