Deadlock Detection in Distributed Database

I. Priyadarshini, Dr.S.S. Sane


In a distributed database, data resides in various sites and several transactions originate at all the sites randomly needing data item from two or more sites. These transactions execute concurrently and thus may lead to a deadlock in which transactions may enter into an infinite waiting state. Therefore, the deadlock handling is an important activity in transaction processing. Thus the database system uses an algorithm for detecting deadlock. The task of deadlock detection is complex in case of a distributed database systems as compared to the centralized system because of presence of multiple sites and communication links and a fear of failure of a site or a link and message delays etc.The deadlock detection algorithm usually maintains a wait-for-graph and periodically checks that it is free of cycles to ensure that the system is not in a deadlock. The algorithm needs to take care of false deadlock cycles under the circumstances of site/link failures and delays in message delivery. Various approaches exist for detecting deadlock in a distributed database system. These algorithms are either centralised or distributed. However many of the these algorithms does not support data replication. Data replication improves availability at the cost of additional storage space. However, as the cost of storage is becoming cheaper and cheaper, the modern distributed systems prefer algorithms that are capable of detecting deadlocks in the presence of data replication. The paper deals with the implementation details of one such deadlock detection algorithm in an replicated database environment.

Full Text:



  • There are currently no refbacks.