14th International Conference on Database and Expert Systems Applications (DEXA 2003)
Neighborhood-Consistent Transaction Management for Pervasive Computing Environments
September 2, 2003
This paper examines the problem of transaction management in pervasive computing environments and presents a new approach to address it. We represent each entity as a mobile or static semi-autonomous device. The purpose of each device is to satisfy user queries based on its local data repository and interactions with other devices currently in its vicinity. Pervasive environments, unlike traditional mobile computing paradigms, do not differentiate between clients and servers that are located in a fixed, wired infrastructure. Consequently, we model all devices as peers. These environments also relax other assumptions made by the mobile computing paradigm, such as the possibility of reconnection with a given device, support from wired infrastructure, or the presence of a global schema. These fundamental characteristics of pervasive computing environments limit the use of techniques developed for transactions in a mobile computing environment. We define an alternative optimistic transaction model whose main emphasis is to provide a high rate of successful transaction terminations and to maintain a neighborhood-based consistency. The model accomplishes this via the help of active witnesses and by employing an epidemic voting protocol. The advantage of our model is that it enables two or more peers to engage in a reliable and consistent transaction while in a pervasive environment without assuming that they can talk to each other via infrastructure such as base stations. The advantage of using active witnesses and an epidemic voting protocol is that transaction termination does not depend on any single point of failure. Additionally, the use of an epidemic voting protocol does not require all involved entities to be simultaneously connected at any time and, therefore, further overcomes the dynamic nature of the environments. We present the implementation of the model and results from simulations.
Downloads: 4868 downloads
Google Scholar Citations: 11 citations