Performance models of distributed database systems
Victor Ok Li · DSpace@MIT (Massachusetts Institute of Technology) · 1980
A distributed database (DDB) consists of redundant copies of data files geographically distributed on a computer network. This thesis develops a performance model of a DDB. This model can be used to compare the performance (i.e. response time, utilization, etc.) of different concurrency control algorithms. We started by developing a network of queues model of a communication subnetwork. We have originally attempted to employ Jackson's Model but have concluded that Jackson's Model is inadequate for our purposes. The Independent Queues Model that we developed in this thesis make somewhat stronger assumptions than Jackson's Model, but has more flexibility and approximates better a real communication subnetwork. We found that in a general DDB, concurrency control algorithms could not be modelled accurately without taking into consideration the particular query processing strategy used. We have therefore developed two new query processing strategies: the MST and the MDT Algorithms. These two algorithms are easy to analyze and to implement. We next modelled the competition among different transactions in the DDB for the services of the database management system. Probabilistic arguments were used to determine the probability of conflicts between different database transactions and the delay due to conflicts. (Author)