Concurrency control based on distributed cycle detection

This paper presents a novel optimistic concurrency control mechanism for distributed database systems. The optimistic concurrency control mechanism is based on a distributed algorithm for cycle detection which determines whether a transaction violates serializability or not. There are three advantages of this concurrency control mechanism. (1) It certifies execution of transactions in a distributed fashion and in parallel. (2) Local concurrency control is performed at each site independently. Global concurrency control verifies consistency of local concurrency control only when a transaction intends to commit. The separation of local and global concurrency controls enables us to integrate different local concurrency control mechanisms such as two-phase locking and timestamping into a single concurrency control mechanism. This is suitable for concurrency control in heterogeneous database systems. (3) No assumption is made on how data are distributed. It works in partially replicated database systems as well as in partitioned or fully replicated database systems.