Modelling interactive workloads of time-sharing computer systems

Abstract

One of the major obstacles to performance evaluation studies is the definition of a representative or accurate workload. User scripts, a detailed description of user activities to specify an interactive workload, can be used as input for remote terminal emulators as well as a representation of an interactive workload the users impose on a system. User scripts are often based on the widely accepted user behaviour model 'LIST-MODS-RUN'. A method of analysis is described which suggests that this user behaviour is not realistic. More accurate models are given to describe user behaviour at system command language level. It is also studied how different users groups adapt to different loads. The results are then used to define a workload model. An automatic workload genrator is described to generate workload models in form of user scripts ready for execution 'on ERTE which is a system designed to exert interactive workloads on a multiaccess system. Experimental runs with ERTE demonstrate the accuracy of the model and provide the basis for more realistic performance measures on EMAS and the development of ERTE.