Transaction scheduling and concurrency control for time critical applications
Student thesis: Master's Thesis
Related Research Unit(s)
In a time critical application, the transactions are associated with time constraints on their completion times. It is important to the usefulness of the system to meet the time constraints of the transactions. In this study, we have chosen two typical time critical applications, programmed stock trading systems and mobile computing systems for command and control, as examples. After studying their system characteristics, we have identified two factors that could seriously affect the predictability in transaction execution in these systems. They are triggering of transactions and interactions between different types of transactions in the system. Triggering of transactions is important to the system to respond to the changes in system environment. It is common to see that a time critical application, i.e., programmed stock trading system, may have different types of transactions with different characteristics and performance requirements. The transaction scheduling methods and concurrency control protocols proposed from previous research work in real-time database systems research may not be suitable for scheduling of mixed transactions. In this thesis, we first study how triggering of transactions affects the system performance in real-time database systems. Committing a triggered transaction requires meeting the deadline constraints of its triggering transaction and also the data deadlines of its accessed temporal data objects. Generations of triggered transactions also increase the system workload and the probability of data conflicts. Different approaches for assigning deadlines and priorities to the triggered transactions are proposed with the objectives to satisfy the timing requirements of triggered transactions especially for the more critical ones. The performance of the new approaches and their relationships with different coupling modes has been studied using simulation. In the second part of the thesis, a general real-time database system model is defined in which various types of real-time transactions, i.e., hard, soft and non-real-time, are co-existed in the system. Due to the different performance requirements of hard and soft real-time transactions, these proposed concurrency control protocols are not suitable to mixed real-time database systems (MRTDBS). Strategies for resolving data conflicts between different types of transactions in a MRTDBS are studied and formulated so that the performance requirement of each individual transaction type can be achieved and, at the same time, the overall system performance can be improved. Finally, we investigate the location update problem in mobile real-time database systems. Many new mobile computing applications require providing support to location-dependent continuous queries (LDCQs) on moving objects. The result of a location dependent query depends on the current locations of the moving objects on which the query has been issued as well as the location of the mobile client, which issues the query. When the query is specified as continuous, the requesting client can get continuously changing result. In order to provide correct and timely results to requesting clients, the locations of moving objects in the system as well as the locations of the requesting clients have to be closely monitored. Although various update generation methods have been proposed in the literature, those methods mainly aim to optimize utilization of the limited wireless bandwidth. The correctless and timeliness of query results reported to the requesting clients have been greatly ignored. An adaptive monitoring method (AMM) for managing the locations of moving objects is proposed. The aim of AMM is to maintain the correctness of the results of query evaluation without significantly increasing the wireless bandwidth requirements.
- Real-time data processing, Transaction systems (Computer systems)