Bright, Gal, and Raschid deal with the policies used for timely propagation of up-to-date information to thousands of clients over a wide area network. It seems as if pull-based data delivery is becoming the preferred solution for rapid and widespread deployment of wide area applications.
The authors present adaptive pull-based policies that explicitly aim to reduce the overhead of contacting remote servers, compared to existing pull-based policies. They model updates to data sources using update histories, and present two novel policies to estimate when updates occur based on individual history and aggregate history. They further develop two adaptive policies to handle objects that initially may have insufficient history or objects that experience changes in update patterns. Their extensive experimental evaluation uses three data traces from diverse applications to show that history-based policies can reduce contact between clients and servers by up to 60 percent compared to existing pull-based policies, while providing a comparable level of data freshness. They provide experiments to demonstrate that their adaptive policies can select the best policy to match the behavior of an object and perform better than any individual policy.
According to the authors, this work serves as a first step toward scalable pull-based data delivery that minimizes communication between clients and servers, while meeting client freshness requirements. The authors suggest that future work could involve a more thorough analysis of burst detection and the identification of more adaptive policies for managing the uncertainty involved in the use of stochastic update models when estimating update events.