Accuracy-Based Cache Consistency Management for Numerical Object Replication

Object replication and cache consistency have been one of major design issues in the recent Internet applications. In this paper, we forcus on accuracy-aware applications (i.e., sensor applications), which use numerical data with allowing some errors. As long as the cache consistency between the originpsilas and replicated objects is guaranteed to some error levels, the system do not need to refresh its cache, resulting in saving network workloads. We propose a numerical cache consistency model and a cache control method for such applications. The results of our experiments on weather temperature data have shown (1) network traffic was dynamically reduced depending on application requested accuracy, and (2) the rate of exceedance to the threshold of allowable error was controllable, which was not in the traditional time-based cache validity control.