Adaptive parameter collection in dynamic distributed environments

Cost-effectively collecting distributed state information is a challenging problem. There is perhaps no single perfect solution since different distributed application environments pose different requirements from the information collection process. Knowledge of the environment, in terms of traffic conditions, load models etc., plays a key role in determining tradeoffs between accuracy (needed to ensure quality-of-service requirements) and cost-effectiveness. We develop an adaptive information collection algorithm that utilizes network traffic knowledge characterized using a time series model. The algorithm utilizes an information collection architecture consisting of a directory service integrated into the middleware layer with monitoring modules distributed across the network. The cost-effectiveness of the proposed information collection algorithm proposed is verified in simulations over diverse network traffic patterns, i.e. Internet WAN (TCP), MPEG (multimedia) and Web access traffic traces. Our results show that the proposed adaptive information collection algorithm compensates for inaccuracies in network traffic predictions in a cost-effective manner.