Performance Modeling to Divide Performance Interference of Virtualization and Virtual Machine Combination

Abstract

When we evaluate performance of a virtualized system, we should consider two aspects, performance interference and the number of virtual machine (VM) combinations. Performance interference is caused by sharing physical resources among VMs and virtual machine monitor (VMM) scheduling. We should precisely incorporate these effects into performance evaluation. We also need to evaluate many VM combinations to find the optimal consolidation of servers, therefore, it is important we apply an efficient evaluation method to reduce evaluation time. We propose a layered performance model to address these aspects. We regard a virtualized system as a combination of two layers: one consisting of VMs and the other consisting of VMM and physical resources. We construct a performance model for each layer. We apply the white-box approach to the VM layer model and the black-box approach to the VMM/physical resource layer model. The white-box model is flexible for representing many VM combinations and the black-box model incorporates performance interference. By allocating each aspect to each model, our proposed model evaluates performance precisely and efficiently. We discuss the effectiveness of our proposed model with a case study of storage I/O contention.