Improved optimistic concurrency control and its use in distributed database systems
The original approach of optimistic concurrency control has some serious weaknesses with respect to the validation technique, the treatment of long transactions and the problem of starvation. In the first part of this paper some design alternatives of the validation-phase are considered which not only avoid these disadvantages but additionally improve the performance of the original approach. Very promising is a new validation technique which especially supports readtransactions, therefore decreasing the number of backups substantially. In the second part of this paper an adaptation of the new optimistic concurrency control schemes to a distributed environment is presented. Although a lot of algorithms based on two-phase locking have been proposed, concurrency control for distributed databases is still the subject of intensive research because most locking approaches have the common disadvantage that deadlocks may occur. As deadlock detection is a difficult and expensive task in a distributed database system, nonblocking algorithms like the time-stamp approach or the optimistic approach are promising alternatives. We present an optimistic concurrency control algorithm for a distributed database system, which is based on a global coordinator. This algorithm benefits from the improved validation techniques presented in the first part. Its advantage is high flexibility and good performance due to high parallelism and small message overhead. Of course, the algorithm also is deadlock-free. lt will be shown that the main disadvantage, the existence of a global coordinator, does not have that much weight.
Nutzung und Vervielfältigung:
Alle Rechte vorbehalten