skip to main content
Optimistic concurrency control algorithms for distributed database systems
Publisher:
  • State University of New York at Stony Brook
  • Stony Brook, NY
  • United States
Order Number:AAI8812694
Pages:
140
Bibliometrics
Skip Abstract Section
Abstract

We present optimistic concurrency control algorithms for non-replicated and replicated database systems in this thesis. The optimistic concurrency control algorithm as proposed by Kung and Robinson is criticized for its problem of excessive transaction rollback. We have generalized the validation conditions of the original algorithm which results in reducing the amount of rollback in the system by allowing additional interleavings of transactions. We also show that the generalized algorithm permits more concurrency than the optimistic concurrency control algorithm. Another problem of the optimistic concurrency control algorithm is that the transactions may see inconsistent views of the database. This problem can be eliminated by using a multiversion implementation of the database. We have combined the generalized validation condition with a multiversion database to yield a generalized multiversion optimistic concurrency control algorithm. This algorithm guarantees consistent views to the transactions and reduces transaction rollback significantly. Also, read-only transactions are not subjected to validation in this algorithm.Next, we examine the synchronization overhead associated with quorum based algorithms for replicated data. We identified the reasons for delays in these algorithms--synchronization delays due to pessimistic concurrency control and latency incurred to accumulate required quorum for each action of a transaction. We have developed a family of optimistic algorithms for replicated data by making different optimistic assumptions about the underlying system. A partially optimistic algorithm is developed, for an environment in which the probability of conflict between transactions is low and the probability that the local replica is current is not high, by combining optimistic concurrency control with a pessimistic blocking quorum technique. A second partially optimistic algorithm is developed, for an environment in which both the probability of conflict between transactions and the probability that the local replica is current are high, by using pessimistic concurrency control with an optimistic non-blocking quorum accumulation mechanism. Finally, we present a fully optimistic algorithm that uses optimistic concurrency control with an optimistic non-blocking quorum accumulation technique. Depending on the characteristics of the underlying system any one of these algorithms can be chosen which may reduce the two kinds of delays present in the algorithms for replicated data.

Contributors
  • University of California, Santa Barbara

Recommendations