VMcol: A Collector of Garbage for Virtual Machine Image Files

Abstract

Virtual disk for a virtual machine (VM) is a virtual image file on a physical node. Inside a VM, the guest VM operates the virtual disk as the general OS, while outside the VM, the virtual image file grows larger and larger with the data operation of the VM, because of the semantic gap between the guest VM and the virtual machine monitor (VMM), the delete operation in guest VM cannot be received to the VMM. This leads to data space vanishing on physical node, even there are a large volume of data space released inside the VM. To target this, we present the design, implementation and evaluation of the VMcol, a system that collects the garbage space lively which has been deleted in the guest VM. When a file is deleted in the guest VM, VMcol will reclaim the deleted data for the corresponding virtual image files without interrupting the service of the VM and requiring additional physical space. The performance evaluation shows that VMcol improves the storage utilization with little performance penalty of VMs in terms of CPU utilization and I/O bandwidth.