Concurrency control algorithm for shared disk cloud DBMS
Islam Eisa, Rashed Salem · 2017
Using shared disk architecture for relational cloud DBMSs enhances their performance and throughput and increases the scalability. In such architectures, transactions are not distributed between database instances and data are not migrated, whereas any database instance can read and access any database object. Lock technology control for concurrent transactions ensures their consistency especially in shared disk architecture, using traditional granularity database locks for cloud database, can cause numerous problems. This paper proposes an optimistic concurrency control algorithm that uses soft locks and minimizes the number of accessed database instances for validating a transaction. It creates a lock manager for all database objects and distributes it over database instances until it does not have to validate the transaction, neither with single database instance if it is owned by only a database instance nor with all database instances if it is replicated on all database instances. The proposed algorithm is evaluated against other cloud concurrency control algorithms and the results confirms its effectiveness.