Concurrency control in multidatabase systems: A performance study

Abstract

A multidatabase system (MDBS) is a federation of multiple autonomous local database systems (LDBSs). Over the years, it has been recognized that the most challenging issue of concurrency control in an MDBS has been a concern about how to resolve indirect conflicts and thereby to ensure global serializability (GSR), without sacrificing the local autonomy. Even though the global serializability issue is well understood and many concurrency control algorithms have been proposed, it is noticeable that the performance trade-offs of those algorithms are not yet understood. To address issues about the performance of concurrency control algorithms in an MDBS, we examine two MDBS concurrency control algorithms through simulation: optimistic ticket method (OTM) and global ticket method (GTM) that we have proposed. In OTM, the GSR is assured by way of accessing a local object (ticket), by which any indirect conflicts that may cause non-serializable execution are resolved. The global transaction manager (Global_TM) in GTM maintains a global variable (ticket), which is similar to timestamp, and the commit order of a GT is predefined by the order of assigned global ticket value. We examine the performance trade-offs of the two concurrency control methods (CCMs) with respect to ensuring GSR in particular. To ensure GSR, the (GTs) are expected to frequently restart in OTM and block in GTM, in this study, we also investigate the implications of time-out interval on the performance as we assume that the time-out approach is used as a global deadlock resolution method for both algorithms. Our experimental results showed that GTM outperforms OTM in cases that short time-out values are given. However, in case that the time-out value relatively becomes long, our results demonstrated that OTM outperforms GTM. (C) 1998 Elsevier Science B.V. All rights reserved.