A buddy-file-system to improve block level sharing of disk images in virtualization environments

We evaluate how existing problems in virtualization environments can be rectified by exploiting common data blocks. BuddyFS is a new file-system which identifies common blocks and recreates the disk images with fewer disk block exchanges between the hosting environment and the remote machine. We have also discussed how semantic information can be used to enable new levels of sharing. The idea is to avoid transferring data blocks from remote servers if equivalent data exist in the hosting environment. It is the system data such as the kernel pages, library pages that are most likely to reside in the hosting machines. In order to achieve this main goal, we needed a method to identify data that are equivalent in disk images and to evaluate whether the commonalities amongst disk images can effectively reduce the amount of data that has to be transferred to create the virtual machine. Then using this comparison, a controlling module needs to map the disk image blocks and produce a structured view for the virtualization environment. Even though sharing is the main focus, unique data alterations by different users must be saved. These are the main design requirements and considerations. The fundamental approach was to build a file-system to reuse client-side cached blocks and transmit less number of disk image blocks. The results suggest that introducing this concept in the hosting environment is a plausible optimization when transferring and storing data in a virtualization environment. When creating virtual machines it is possible to identify reusable common blocks in the hosting environment that can be used to recreate the disk images. BuddyFS can be used to reduce the overhead of managing virtual devices in now rapidly expanding virtualization environments such as cloud computing.