CapSys: A tool for macroscopic capacity planning of IBM mainframe systems

摘要

Capacity planning for large-scale systems like mainframe installations is typically based on rather coarse historical workload data. Hence, it is advisable to employ techniques and tools that use the same macroscopic (high-level) view of the overall system. We follow the classic approach known from performance modelling to compose a model essentially out of two parts, the system model (or machine) and the workload model. The system model is an aggregated mainframe system model that is focussed primarily on processing resources that are given by processors of different types (i.e. general all purpose processors and processors dedicated to Java, Linux, Database and others) and their processing capacity given by scaling tables. The workload model follows the structuring given by logical partitions and is built from historical utilization data that are permanently collected during normal operation. Based on this framework it is possible to define and improve the allocation of logical partitions to processors or processor pools. Employing the CapSys tool the designer can interactively vary the configuration of processor pools as well as the shape and intensity of the workloads to be allocated. In this report we describe background, architecture and functionality of the CapSys tool.