Browse Theses

The concurrency-control algorithms allow concurrent transactions to process database, and also introduce the interference between transactions. This feature complicates the performance evaluation of concurrency-control algorithms. Two concurrency-control algorithms, the simplified locking algorithm and the optimistic timestamps algorithm, are studied here in details, by which I introduce a general modeling procedure. The simplified locking algorithm allows the least concurrency. In comparison with the optimistic timestamps algorithm, the simplified locking algorithm works well in heavy traffic. According to the relationships between database items, the transactions are classified into a number of classes. The average number of ACTIVE transactions in a class and the thruput of a class of transactions represent respectively the concurrency and the interference between transactions. The results of analyses of the algorithms also provide the other quantities such as the communication cost per accepted transaction, the data item utilization.

With the availability of inexpensive microcomputers, we are seeing many changes in database management system. It is likely that the database system of the future uses one mainframe to manage the stored data, together with microcomputers to handle the human to machine interface in interactive transaction-based system. The database items are sent back and forth between the mainframe and the microcomputer. Most of the life time of transaction is due to the random transmission delays which are very large in comparison with the time, caused by the resource contention in the mainframe and the transaction processing in the microcomputer. Therefore, the paper is concentrated on the study of data contention in database management system.

A database is composed of a number of items where an item is the smallest lockable entity in the database. We may imagine that the items in the database are each a server and the transactions are customers. Performance modelling of database management systems is achieved by specifying a strategy which ensures the integrity of the data, by classifying the transactions according to the relationships between database items, and by describing the transaction behavior. Queueing model and probability model are most useful for analyses.