Modeling and analysis of memory virtualization in cloud computing

Abstract

Virtualization is a promising approach to consolidating multiple online services onto a smaller number of computing resources. The adoption of virtualization and mappings between physical and virtual resources can further exacerbate challenges. Stochastic reward net (SRN) is an extension of stochastic Petri nets which provides compact modeling facilities for system analysis. In this paper, we address SRN hierarchical modeling of memory virtualization in cloud computing. Memory overcommit techniques such as ballooning, page swapping, and sharing, are applied and modeled for virtualization as a lower-level model. We propose three memory management policies, combination of ballooning and swapping intended for busy virtual memory (VMEM), to increase virtual memory utilization. These policies are handled by using guard functions of transition in SRN. The hypervisor with these policies can relinquish the memory pages of not so busy VMEM and let the busy VMEM get the pages. Memory availability with failure related behavior is modeled and analyzed as a upper-level model. We combine above two level models together and get performability measures. The upper-level model is the structure state model representing the state of the system with regard to failures and repairs of memory pages (banks). There are $$i$$ i active memory pages in progress. Each state of $$i$$ i is assigned a reward rate equal to the memory utilization from the lower-level performance model. The lower-level model captures the performance of the system, especially several memory utilization measures, within a given structure state. Measures of interest are considered as follows: memory utilization for allocated and actually used memory, memory availability with memory failure/repair behavior. These measures are expressed in terms of the expected values of reward rate functions for SRNs.