Stream: Woodcote
Time: 16:30 - 17:15
Drawing from a 'real observation @KBC', we will examine how seemingly small configuration choices can significantly influence CPU performance index (CPI), cache utilization, and overall system throughput. We begin by analyzing the size of the system trace table and its interaction with processor cache behavior, highlighting how trace configuration can introduce hidden CPI overhead. Closely related, we investigate the effect of CPENABLE(SYSTEM) on CPI and processor distribution of I/O interrupts, demonstrating how Workload Manager (WLM) dynamically shifts interrupt handling across processors. The presentation then moves into library concatenation behavior, challenging common assumptions about STEPLIB and USERLIB search order, and discussing performance implications of STEPLIB allocations with excessive dataset concatenations. Building on this, we explore module preloading techniques, including how to leverage SDSF’s Module Fetch display and activation approaches to reduce fetch latency and improve runtime efficiency. From a measurement perspective, we emphasize the importance of high-frequency performance data, including the role of SMF 113 interval data at one-minute granularity and CPI tracking at five-second intervals via SMF 98, enabling fine-grained performance diagnostics and tuning insights.
There is currently no attachment for Lessons from the Field: z/OS Performance Tuning & Optimizations & Understanding
+35 years experience on Mainframe, majority of time spend as a Db2 systems programmer, last 10 years I'm busy as a Mainframe Capacity Performance Specialist