Software Rejuvenation in Cloud Systems Using Neural Networks

Abstract

Virtual Machine Monitor (VMM) is very important for the cloud and data center environment. VMM runs continuously for a long time and hence encounters the problem of software aging. VMM experiences failure because of software aging. In order to prevent the VMM failure caused by software aging, a proactive fault management approach called software rejuvenation is used. There are various software rejuvenation approaches existing in literature that can be broadly categorized into two categories namely model based approaches and measurement based approaches. Time to failure is predicted in measurement based approaches by monitoring the resource usage statistics. There can be any non-linear relationship between resource usage statistics and the time to failure. Such a nonlinear function can be approximated using Artificial Neural Networks (ANN). The change in the value of attributes of resources is given as input to ANN and new value of time to failure is generated as output. Experiments shows that if there is some pattern in the arrival and departure of the VMs, then the prediction is more accurate.