BuddyFS: A File-System to Improve Block Level Sharing of Disk Images in Virtualization Environments

Abstract

Virtualization based computing is rapidly expanding as an alternative to conventional physical machine based computing because virtualization environments o er an over- all nancial bene t to the end users where they do not need to maintain physical resources. Increased availability of high speed internet and corporate network connec- tions has also in uenced this high demand . As virtualization environments become more wide spread, accessing, storing and managing the virtual resources become an equally tedious job to managing physical resources. The cost bene ts expected to achieve via virtualization will reduced if new optimizations are not integrated to ex- isting techniques. Most virtualization environments demand users to download complete disk images to create virtual machines. A disk image has the virtual resemblance of a physical machine. Thus, there is a tendency to consider disk images as separate, isolated devices when storing, accessing and managing them, similar to physical machines. However unlike physical machines, disk images in one host, are not actually physi- cally dispersed. We note that the system data in disk images hardly varies from user to user, and there is a possibility to have common data blocks amongst the remote disk images and the local virtualization environment. Thus, we propose a new le-system BuddyFS, to identi ed these common data blocks and reused them to create disk images in the local machine instead of tranfering the whole package from a remote machine. BuddyFS identify blocks as kernal pages, library pages and uses this semantic information to enable new levels of sharing. We prove that reusing the common blocks in virtualization environments can improve transfer times and storage space. It can be used to reduce the overhead of managing virtual devices in now rapidly expanding virtualization environments such as cloud computing.