Autonomous Transaction Processing Using Data Dependency in Mobile Environments

Mobile clients retrieve and update databases at servers. They use transactions in order to ensure the consistency of shared data in the presence of concurrent accesses. Transaction processing at mobile clients faces new challenges to accommodate the limitations of mobile environments, such as frequent disconnections and low bandwidth. Caching of frequently accessed data in a mobile computer can be an effective approach to continue transactions in the presence of disconnections or other reasons for losing messages. It can help to reduce contentions on the narrow bandwidths of wireless channels. Concurrency control schemes using caching ensure consistency among data items from the server and from the client caches. We present a scheme that can increase the autonomy of mobile clients for validating transactions, using caching and pull-based data delivery. In the scheme, mobile clients can decide to commit read-only transactions locally, without interaction with the server and can detect transaction aborts earlier. The clients receive from the server dependency information, from which they build partial serialization graphs. Dependency information is based on the notion of i-order dependency introduced in the paper. We study the performance of the proposed protocol by means of simulation experiments. This research was supported by CERIAS and NSF grants CCR-9901712 and CCR-0001788. Currently with Samsung, Seoul, South Korea.