|M.Sc Thesis||Department of Computer Science|
|Supervisors:||Assoc. Prof. Gal Avigdor|
|Prof. Pinter Ron|
Wide Area Applications (WAA) involve hundreds of servers and tens of thousands of clients. In such systems, the need to monitor the availability of resources residing at servers is of great importance for the sake of supplying high quality of service to the clients. To this end, we propose an algorithm for background active collection of resource availability patterns, assuming that resources have a limited lifespan.
First, we present a model which describes a WAA environment and show that without periodic probing of the resources in the WAA, the quality of service delivered to the clients is poor. Then we develop an algorithm that periodically probes resources to study their availability status, and which adapts its probing frequency online according to resource popularity and dynamic availability status changes. The estimates of resource availability take into account both basic resource characteristics and the history of previous probings and therefore is adaptive to changes.
We show by simulation that this algorithm outperforms common alternatives. Specifically, we show that it does better than uniform probing which was found to give good results in the different but close area of web crawling.