A Garbage Collector For Virtual Machine Images

In virtualization environment, virtual machine image on a physical node saves operating environment of VM (virtual machine) and all data of users. But semantic isolation, introduced by virtualization, makes VM image preserve all the useless data which have been deleted in guest VM permanently. This not only wastes the storage resource but also harm interests of users. Presently, image management tools, which collect the garbage of VM images, need to interrupt guest VM service and scan the virtual disks and records. Therefore, there is a key question about how to improve storage resource, by reclaiming garbage of VM images with ensuring user transparent and VM's available.A collector of garbage for VM images, named VMcol, lively collects garbage for VM images without scanning VM images, while ensuring user transparent and VM's available. VMcol implements user-deletion monitor, information transform, fast context deletion and so on. Therefore, when file was deleted in VM VMcol can reclaim garbage automatically for the corresponding VM image. The system comprises of user-operation monitor and garbage processor. When user deletes files in a guest VM, the user-deletion monitor uses capture deletion and analyze information of deleting file in real time to get the valid information of deleted files in the guest VM. Then the user-operation monitor sends the information to the garbage processor with the information transmission model. According to the information, the garbage processor uses information transforming to fix area of worthless data in the image without scanning the image, and then deletes worthless data in the image with ensuring running of the guest VM.Based on Xen 3.3.1 virtualization platform and Qcow image, we use C and Python to implement the collector of garbage for VM images, and then test the system. As expected, the function test shows that the system collects, to improve the disk space utilization, garbage of VM images automatically without stopping guest VM service. Besides, the performance test shows that the system hardly affects the delete operation performance in the guest VM and never introduces overheads of the CPU and disk read in server. Furthermore, the virtual disk I/O performance is not affected but improved a little, the write bandwidth of virtual disk increases by 8% and read bandwidth of virtual disk increases by 2.7%.