Safety by Uninterpreted Locks

The locking of a static set of database transactions is discussed in the setting of uninterpreted locks. Uninterpreted locks generalize entity locks to the case where any variable may be used for locking, not only those associated with the entities accessed by the transactions. A new condition for safe lockings is presented. There exist safe lockings which could not be accepted by an safe locking in the entity lock model. Moreover, it is shown that the condition is necessary for safety whenever only entity locks are used. It is also shown that there are safe sets of locked transactions not satisfying the condition. This implies that safe locking by uninterpreted locks is too powerful to be completely characterized by our condition; however, our condition is more powerful than safe locking in the entity lock model. The given condition for safety is expressed in geometric terms and it can be used for proving the safety of specific lock policies. In particular, safe pre-analysis lock policies can be designed which strictly dominate the 2-phase lock policy with respect to the sets of accepted schedules.

Authors

• G. Lausen
• E. Soisalonsoininen

Keywords

No keywords are indexed for this paper.