Execution and processing time in computer performance measurements
Borivoje Djokić · Proceedings of the 17th conference on ACM Annual Computer Science Conference · 1989
Tasks cooperate with each other to perform processing and compete with each other for limited resources. Relationships between technical components are represented by differential equations and, could be solved analitically. In large multiuser systems stochastic processes are introduced: concurent requests for resources, transactions. The level of aggregation in choosing descriptions of workload is a crutial choice. Measures have been done on IBM 370/3031, OS MVS, over a four years period. Ratio elapsed time/processing time for the applications in background had a slight postive slope after an initial decrease. A steeper slope characterizes the next period although the curve leveled at 6 units of execution per one processing time unit. Fast increase characterised next two years, and the curve reached critical level of 47 time units of execution for one unit of actual processing. This ratio tourned in a paramemeter that shows given workload and throughput resulting from straight CPU activities, and other resources participating in the form of processor waiting for other services. The curve is divided in two sections as a result of unbalanced system's structure after increasing number of users, number of tasks and channel operations. Average duration of a terminal session yields a significant increas, with an influence of the users training level in the first part of a curve, and overall systems inefficiency in the second part. A ratio total terminal session time/total session processing time shows a growth of of approximately 40% in the entire period. More experienced users wait longer for the terminal to react, therby loosing the efficiency of interactive work. Unbalanced growth is a result of the attempts to keep the interactive component at a relatively high level, thus sacrificing background performance. Overall workload and processing time are highly positively corelated. Different execution/processing time ratios are used for measuring global parameters, as a relatively machine independent, including more system components (not accurate to the level of single process) for functional approach to the computer system modeling.