What's-Different Analysis and its Application to Performance Management in VM SP/HPO

Abstract

Performance analysts spend a great deal of time answering questions that involve comparing measurement data (e.g., Why is performance now worse than last hour?). For performance analysts responsible for 5 to 10 mainframes, answering such questions is time consuming. With the advent of departmental computing, the situation becomes worse since there are more questions and fewer analysts per system. This paper describes a technique called what's-different analysis that answers comparison questions. Qualitative what's-different analysis only requires knowing directional effects (e.g., CPU waits increase with CPU utilization). If it is possible to compute new values of variables (e.g., from algebraic relationships), quantitative what's-different analysis can be used to quantify the contribution of individual variables to performance problem (e.g., 55% of the increase in system expansion factor is due to waits for storage). An analysis may be purely qualitative, purely quantitative, or a mixture of both. We illustrate what's-different analysis by applying it to measurements taken from an IBM 3081K running IBM's operating system VM SP/HPO (Virtual Machine System Product with High Performance Option) with a workload from the petroleum industry.