Performance Analysis of Data Sharing Environments: (ACM Distinguished Dissertation)

Large transaction-processing applications, such as banking, flight reservations, telecommunications, and networking, demand increasingly high throughput and fast response times. Coupling multiple systems for database transaction processing can provide these systems with higher capacity and availability at a lower cost. Data sharing is one promising architecture for such an environment. This monograph develops a comprehensive model for analyzing the design and performance of a complex data-sharing environment that consists of multiple, loosely coupled transaction-processing nodes with a common database at the disk level. By capturing the salient features of this data sharing architecture - database buffers at each node with their buffer management policies, concurrency control, buffer coherency, nonuniform database access, and CPU queuing delay - the model can be used to answer a number of design questions about scalability, buffer utilization, skewed access, policy selection, and optimal system configuration. Included are two new submodels, for private and shared buffers.

Table of Contents:
Part 1 Introduction: design issues in a data sharing environment; problem statement; earlier work; research contributions; overview of the monograph. Part 2 Data sharing environment: system architecture; buffer management policy - private buffer management policy, shared buffer management policy; transaction execution - two-phase locking protocol, optimistic protocol with read timestamp; simulation model. Part 3 Integrated system model (uniform access): concurrency control model; system resource model; buffer model - simple buffer model, validation of the simple buffer model, buffer model for transaction; validation and experiments. Part 4 Data sharing model under skewed access: buffer model - buffer model without rerun transactions, buffer model with rerun transactions, buffer flushing; concurrency control model; system resource model; validation and results - buffer hit probability, transaction response time, buffer flushing; effect of skewness. Part 5 Data sharing model with shared buffer: modeling framework for SBMPs - analysis of the policies ignoring dependency, analysis capturing dependency; validation and results - selection of the SBMP; optimal system configuration. Part 6 Summary and conclusions: summary of results; subsequent work - comparison of transaction processing architectures, use of an analytical model as a configuration tool, performance study of buffer coherency policies, analysis of buffer coherency policies through lock retention; future extensions.