A Model-Driven Approach to Distributed Information Management

In this paper, we present a model-driven distributed information management system called MDIM that can resolve multi-attribute range queries about large-scale networked systems. Different from previous work, MDIM can adaptively configure its information collection and query resolution operations based on a dynamically maintained information model. The information model captures both statistical information query patterns (e.g., frequently queried attributes, frequently queried value ranges) and information attribute properties (e.g., node attribute distributions). Thus, MDIM can better scale to large numbers of system nodes and attributes than static schemes by intelligently minimizing monitoring overhead. We have implemented a prototype of MDIM and evaluated its performance using both extensive simulations and micro-benchmark experiments. Our experimental results show that MDIM always produce much smaller system overhead than static monitoring schemes. More importantly, when system conditions or query patterns change, MDIM can adaptively reconfigure itself in response to the changes.