Estimating Working Set Size by Guest OS Performance Counters Means

Cloud infrastructures imply virtual machines dynamic hosting. The distribution of resources is performed in dependence on the volume and a pattern of used resources. An estimation of current load and prediction of future load are mandatory for effective resources management. The statistical processing of internal counters of a guest operating system gives good prospects. The disadvantage of the method is the complexity of data collection and its processing. The large number of parameters and variance in their behavior in different operating systems and configurations introduce extra complexity for the virtualization case. The present research covers the estimation of a virtual machine working set based on guest OS internal counters. The correction of the estimated value is done in accordance on the feedback of donor guest OS. Keywords-virtual machine; cloud computing; memory management; working set.