Database Scalability, Elasticity, and Autonomy in the Cloud

This extended abstract has more pages than many full papers. The authors provides some textbook-like explanation on generals of cloud computing in the half of the paper (to Section 2.1) and their ongoing projects with some backgrounds in the other half (from Section 2.2)

There could be two different approaches to achieve scalability and atomicity of database at the same time. One way is to improve current key-value storages (inherently scalable) to support some level of atomicity. The other way is to make conventional RDBMSs (with transactional guarantees) to split a large database unit into relatively independent shards for better scalability. The authors name each approach Data Fusion and Data Fission, respectively. The paper introduces G-Store and ElasTraS as prototypes of each approach.

In addition to scalability, the concept of elasticity more focuses on the change of system size (expanding and shrinking). Such events requires data migration to be lightweight enough, to minimize the impact of live services. The authors introduces their prototype, Zephyr, which enables seamless query processing even in the middle of process of migration.