Title : Bounded Locking for Optimistic Concurrency Control

Optimistic methods of concurrency control are gaining popularity. This is especially true with the resurgence of mobile and distributed databases, which inherently rely on optimistic techniques to improve availability and performance of the database system. A key problem with optimistic techniques is that they do not perform well in highly conflict prone environments. Pessimistic techniques, especially locking, perform much better under these circumstances. In this paper we explore a hybrid technique that provides locking for high conflict data items and optimistic access for the rest. While hybrid techniques have been proposed in earlier literature, our technique is unique in that it is self tuning and does not require the transaction manager, the transaction or the user to incorporate any additional knowledge or to specify which data items or transactions are optimistic. Rather, the system uses an LRU data structure called lock buffer to maintain an optimal level of locks in the system. This data structure enhances the performance of the basic optimistic model by automatically providing locking for highly conflict prone data. A unique feature of our algorithm is that locks may be evicted from the lock buffer in an LRU fashion, if the number of data items for which locks are requested exceeds the size of the lock buffer. All transactions affected by such an eviction of locks automatically “become” optimistic with respect to the evicted data items.