Tuning a Virtual Storage System

methodology for performance-tuning a virtual storage system is discussed. This methodology encompasses Performance measurement , workload characterization, performance evaluation, and planned experimentation. Use of the methodology is illustrated by describing results of a case study involving the IBM Research Division's T S S l 3 6 0 system. The performance objectives of different computer installations are diverse. Therefore an operating system should provide a flexible and easily modifiable mechanism for implementing system resource management poli~ies.l,~J Specifying a policy for a particular computer installation is an iterative process, in which there are five logically distinct steps: Understanding the resource management components of the Recording workload and system activity Characterizing the workload and identifying the system vari-Experimenting with various resource management policies Determing the policy that best satisfies the installation's per-system ables that most significantly affect performance formance objectives. The intent of this paper is to illustrate these steps by means of a case study of a virtual storage system, the T S S D ~ O (Time Sharing System/360) installation at IBM'S T. J. Watson Research Center in Yorktown Heights, New York. Emphasis is on the methodology of the five steps rather than on the TSS installation itself. We also stress our use of data analysis techniques to evaluate the effects of system changes on performance. A benefit of our methodology, with regard to modeling system performance, is a simplification of the system description. Our objective is to improve system responsiveness to a typical user transaction, which we define later in this paper.