Abstract

This paper presents a subscription-based overlay network that supports efficient file parallel downloading. First, the system lets users register to a central server and allows the server to incrementally build a topology graph containing the network connections among the subscribers. With this topology graph in place, we address the challenges of minimizing network traffic and choosing the set of nodes from which a chosen file is downloaded through several concurrent connections. Active subscribers and other users are organized in an arbitrary peer-to-peer network. When a subscriber wants downloading a chosen file, it submits a search query to the overlay network and obtains a list of nodes having this file. Then, this list is submitted to the server, which ranks the nodes considering both their network distance to the client and and their workloads. Second, based on these top nodes, a bandwidth-aware parallel downloading technique is executed. If a top node becomes offline, either a new replacement is included or the downloading is resumed with the remaining active nodes. Finally, in addition to being topology-aware, the proposed system also features leveraging idling nodes for data downloading. More specifically, the subscribers who are online but not participating in downloading/uploading are recruited to reduce network traffic and to further disseminate files