Data popularity and shortest-job-first scheduling of network transfers

This paper presents a strategy for scheduling transfers of data items from a single source to a single destination. We are motivated by the scenarios in which an author with limited connectivity relies on a remote publishing server for sharing data items with readers. Our strategy is based on the shortest job first (SJF) algorithm with the computational cost expressed in terms of the transfer time and the number of readers waiting for an item. We consider cases with static and dynamic popularity of items. We prove that our strategy is optimal in the static case. For the dynamic case, we use simulation results to demonstrate that our strategy results in lower average waiting time when compared to an SJF algorithm that bases its decision only on the expected transfer time or only on the popularity of a data item