Modelling distributed computing workloads to support the study of scheduling decisions

Abstract

Process scheduling is one of the most important issues in distributed computing. However, this problem still requires further formalisation to understand the consequences of scheduler decisions. To overcome this drawback, this paper defines the behaviour of computer workloads in terms of a dynamical system model, in which next workload states depend on previous ones. The model considers all variables which influence a computer workload at a time instant t, i.e., received, migrated and processed workloads, as well as the degree of dependence among application processes. It has been validated by a set of experiments which consider: 1) a real-world application, running on a GNU/Linux system; 2) a simulated model, in which all variables are modelled according to probability density functions; 3) an emulated scenario, which provides an environment similar to a real-world distributed system. The experiments allowed the conclusion that the proposed model is consistent with the real-world environment and, therefore, both simulator and emulator present the same tendencies of the real-world scenario.