Content area
This thesis describes an infrastructure for scaling distributed transaction processing applications from local area networks to wide area networks. The infrastructure includes efficient and scalable communication software for transaction processing, a facility for experimenting with distributed database software on an internet, and a surveillance model for controlling performance during the transition.
This thesis proves that a scalable communication subsystem does not need to sacrifice performance for any particular communication environment. For communication on an internet, on a local network, or within a computer, the thesis identifies a set of techniques that are useful to build a specialized version of the communication subsystem to achieve maximum performance in each of these environments. All of these versions can be integrated by an adaptable approach. These techniques can be easily incorporated into existing operating systems with few changes.
The thesis develops a tool to simplify experimentation with distributed database software on an internet. It uses local computers to emulate sites that are far away and to force communication to travel a real path on an internet, thus allowing the experimentation without having to move database sites to the remote internet hosts. By using techniques like source routing and echoing, this method does not require administrative control over the remote experimental sites, providing an inexpensive and accurate replacement for real experiments.
The thesis also describes an experimental analysis of the performance of a distributed transaction processing system in a wide area network setting. It studies the communication behavior of an internet, analyzes its impact on transaction processing, and evaluates various transaction processing algorithms.
Finally, this thesis suggests a surveillance model to control performance in a wide area network environment. Special processes running in different sites monitor the system and communication conditions and report them to higher level applications. The transaction processing system that implements adaptable algorithms will respond quickly to avoid performance degradation caused by the network fluctuation.