Effective scheduling of local interactive processes and parallel processes in a non-dedicated cluster environment

摘要

This paper presents a performance evaluation of the interactions between local sequential processes running on behalf of interactive applications and parallel processes running as part of an overall parallel application on a non-dedicated cluster environment. To control the interactions between the two types of processes we propose to constrain the scheduling of local interactive processes (IPs) by a measure of the maximum response time (MRT) expected by the workstation (WS) user. The measure is assumed obtained through empirical studies. We propose a mathematical model of the scheduling problem based on the usage of the MRT measure. In addition, we propose a scheduling scheme that within the MRT cycle computes the time quanta needed to satisfy the requirements of both local IPs and the parallel task process present in the system. A colored Petri net (CPN) is used to model the scheduling scheme. Simulations of the CPN model and numerical results have shown the effectiveness of the proposed scheduling scheme in allowing the parallel task to ensure a minimum speedup even in heavy-loaded situations and to maximize the speedup adaptively depending on load conditions. In addition the simulation results revealed the sensitivity of the interactions to factors such as interactive job demand, parallel job demand, and arrival rate of interactive jobs among other. Finally, simulation and analytical results have been found to agree nicely thus confirming the correctness of the proposed analytical model.