Information

First Name : Woochun

Last Name : Jun

Title : assistant professor

Company : Dept. of Computer Education, Seoul National Univ. of Education

Address : 1650 Seocho Dong, Seocho Gu

City : Seoul, Korea

State :

Mail Code : Telephone : 82-2-3475-2504

FAX : 82-2-3475-2263

E-mail : wocjun@ns.seoul-e.ac.kr

Cyberpaper : Transaction Processing in Object-oriented Databases

Object-oriented databases (OODBs) have been adopted for non-standard applications such as computer-aided design and artificial intelligence. These applications require advanced modeling power in order to handle complex data and relationships among data. In OODBs, a database is collection of classes and instances where classes and instances are called objects. Especially, a class object consists of a group of instance objects and class definition objects. The class definition consists of a set of attributes and methods that read or modify attributes of an instance or a set of instances. Users can access objects by invoking methods. A typical transaction consists of a set of method invocations on objects. One of the important characteristics in database system is manipulation of shared data. That is, database systems, including OODBs, allow shared data to be accessed by multiple users at the same time. Concurrency control involves synchronization of accesses to database so that the consistency of database is maintained. In order to provide good performance, it is very important that concurrency control techniques incur low overhead and increase concurrency among transactions so that more transaction can run in parallel. Among concurrency control schemes, locking- based schemes are widely used due to its simplicity. In this work, I am concerned with relationship between locking granularity and locking-based concurrency control performance. That is, in general, with bigger locking granularity, lower locking overhead incurs but lower concurrency can be achieved. On the other hand, with smaller locking granularity, higher locking overhead incurs but higher concurrency can be met. For the comparison of performance depending on locking granularities, I construct an analytical model for concurrency control in OODBs using 007 OODB benchmark. Based on this model, various locking granularities are adopted to compare their effects on performance. Performance studies show that concurrency control scheme with smaller locking granularities has better performance than schemes with bigger locking granularities. That is, achieving high concurrency is more important than reducing locking overhead in OODB concurrency control performance.