Dmdedup: Device-mapper Deduplication TargetVasily Tarasov (firstname.lastname@example.org)
We present Dmdedup, a versatile and practical primary-storage deduplication platform suitable for both regular users and researchers. Dmdedup operates at the block layer, so it is usable with existing file systems and applications. Since most deduplication research focuses on metadata management, we designed and implemented a flexible backend API that lets developers easily build and evaluate various metadata management policies. We implemented and evaluated three backends: an in-RAM table, an on-disk table, and an on-disk COW B-tree. We have evaluated Dmdedup under a variety of workloads and report the evaluation results here. Although it was initially designed for research flexibility, Dmdedup is fully functional and can be used in production. Under many real-world workloads, Dmdedup's throughput exceeds that of a raw block device by 1.5 to 6 times.