Demand Forecasting in P2P and GRID Systems

摘要

In this paper we propose to analyze and forecast requests for data objects in P2P or GRID systems by using Box and Jenkins time series modelling. Thereby we presume a hybrid system providing a centralized instance named as Active Rendezvous Server (ARS). P2P participants are offering their disk capacity to the community but are not aware what content is stored on their disks (similar to Freenet). We furthermore assume that popularity requests regarding individual content objects may change rapidly. The ARS performs popularity forecasts, which base on past request observations. Thus, the ARS is able to create replicas before the request boost occurs. As an example, we apply a distributed Video on Demand P2P application.