Probabilistic and Fuzzy Process Classifiers for Operating Systems Scheduler

Abstract

The schedulers residing in kernel of Operating Systems employ patterns of resource affinities of concurrent processes in order to make scheduling decisions. The scheduling decisions affect overall resource utilization in a system. Moreover, the resource affinity patterns of a process may not be pos sible to profile statically in all cases. This paper proposes a novel probabilistic estimation model and a classifier algorithm to queuing processes based on respective resource affinities. The proposed model follows probabilistic estimation using execution traces, which can be either online or statically profiled. The algorithm tracks the resource affinities of processes based on periodic estimation and classifies the processes accordingly for scheduling. The effects of variations of estimation periods are investigated and fuzzy refinements are introduced. Experimental results indicate that the classifier algorithm successfully determines resource affinities of a set of processes online. However, the algorithm can determine inherent affinity pattern of a process in the presence of uniform distribution having enhanced IO frequency.