An Evaluation of Precompilation and Interpretation in Distributed Database Management Systems

Abstract

In this paper we present a simple model for the performance evaluation of Distributed Database Management Systems (DDBMS). We first define a transaction-processing model. Then we provide a set of expressions to evaluate the impact of various factors on the performance. The performance is evaluated in terms of transaction response time. Finally we compare the pre-compilation and the interpretation approaches using these expressions. briefly some of the studies that have focused on distributed database systems. Ries 17 compared four concurrency control algorithms, two based on centralised control and two based on distributed control. The centralised methods were variations of centralised two-phase locking; the differ- ence in the two algorithms was essentially concerning transaction scheduling. The distributed control methods differed in the way they solved the deadlock. In one, a deadlock detection algorithm was invoked periodically, in the other the 'wound-wait' model was adopted, which prevents deadlock situations. The simulation model used involved about twenty input parameters that described the database, the transactions, the sites and the network. Performance measures included I/O utilisation and average response time. The simulation results indicated that choice of the best algorithm in terms of the overall database system performance is application-dependent. However, the results indicated that when most trans- actions can be handled locally the distributed control leads to better performance, while if most transactions are non-local centralised control performs better. Lin and Nolte 14 have first evaluated the two-phase locking in a centralised DBMS. In the simulation model, the application environment is characterised by the trans- action size, and the system environment is characterised by the number of transactions running concurrently and the total number of lockable units. Performance