UPDS: Reliable Storage for Personal Data in Online Services

We all have a large amount of data stored in Gmail, Facebook, Twitter, Google Docs and many other online services. With the increasingly importance of these online data, we need an effective way to protect it. This paper proposes a personal data storage system, UPDS, to improve personal data's availability and durability. Two common approaches to improve availability and durability are complete replication and parity scheme such as erasure coding. Although the erasure coding scheme can reduce storage cost, the self-contained feature of online service data source increases its bandwidth cost and decreases its durability. Leveraging the temporal locality of personal data's access model, we propose a lazy-committed log-structure storage architecture and content-based block striping technology to solve the problem. This paper explores the distributed storage system on self-contained data source for the first time and analyzes it system model. The system analysis results show that UPDS can improve availability and durability of personal data at low cost.