A New Distributed Optimistic Concurrency Control Method and a Comparison of its Performance with Two-Phase Locking

Alexander Thomasian IBM T. J. Watson Research Center P.O. Box 704 Yorktown Heights, NY 10598, USA The performance of high-volume transaction processing systems is determined by the degree of contention for hardware resources as well as data. Data contention is especially a problem in the case of data partitioned (e.g., distributed) systems with global transactions accessing and updating objects from multiple systems. While the conventional two-phase locking (2PL) method of centralized systems can be adapted to data partitioned systems. it may restrict system throughput to levels inconsistent with the available processing capacity. This is clue to a significant increase in lock holding times and associated transaction waiting time1' for locks in distributed data partitioned systems, as compared to centralized systems. Optimistic concurrency control (OCC) is similarly extensible to data partitioned systems, but has the disadvantage of repeated transaction restarts, which is a weak point of currently proposed methods. We present a new distributed OCC method followed by locking, such that locking is an integral part of distributed validation and two-phase commit. This OCC method assures that a transaction failing its validation will not be re-executed more than once. in general. Furthermore deadlocks, which are difficult to handle in a distributed environment, are avoided by serializing lock requests. We outline implementation details and compare the performance of the new scheme with distributed 2PL through a detailed simulation, which incorporates queueing effects at the devices of the computer systems, buffer management, concurrency control, and commit processing. It is shown that in the case of higher data contention levels, the hybrid OCC method allows a much higher maximum transaction throughput than distributed 2PL. We also report the performance of the new method with respect to variable size transactions. It is shown that by restricting the number of r·estarts to one, the performance achieved for variable size transactions is comparable to fixed size transactions with