Probabilistic Estimation of Resource Affinities of Processes in Computing Systems

The determination of various resource affinities of processes is an important parameter for making effective scheduling decisions by the schedulers in any operating systems kernel. The resource affinities of a set of processes may be dynamic in nature based on application logic and execution environments. This paper proposes a novel probabilistic estimation model and corresponding classifier algorithm to segregate processes in different queues based on respective resource affinities. The classifier algorithm is online in nature and tracks the dynamic variations of resource affinity patterns of processes. The algorithm classifies processes according to resource affinities for scheduling purposes. The effects of dilated estimation periods are investigated. Experimental results indicate that the estimation model and algorithm successfully classifies a set of processes based on execution traces.