Deadlock Detection in Multidatabase Systems: a Performance Analysis

Deadlock detection is an interesting problem in MultiDataBase Systems (MDBS), since if all local transaction managements are blocking and force direct conflicts between global transactions, the problem of ensuring the global serializability in the MDBS is reduced to detecting and resolving global deadlocks.Unfortunately the autonomy of the local systems precludes the visibility of the state of local transactions and the contention on items, and therefore the classical approaches proposed for homogeneous distributed database systems, and based on necessary and sufficient conditions, cannot be extended to the MDBS case. A few specific methods have been proposed in the literature that exploit weaker necessary conditions to detect potential global deadlocks, that not necessarily correspond to real ones. In this paper we present a comparative performance study of several global deadlock detection methods. The results of the analysis have suggested a new Hybrid Deadlock Detection method, that we present in the paper and that is very well suited for a distributed implementation and has a performance that , according to our experiments, compares favorably with all the other methods in a variety of workload conditions