Topology-aware server selection method for dynamic parallel downloading

For distributed file transfers, parallel downloading is one of effective schemes to shorten a transmission time and/or to get more file transfer rate. In parallel downloading, single file is transferred by using multiple connections to different file servers. However, the effectiveness of parallel downloading is strongly dependent on the physical network topology, i.e., if some connections share the same bottleneck link, the throughput is not increased and the additional connection simply wastes the server resources. In this paper, we propose a server selection method for parallel downloading by taking the physical network topology into consideration. Our approach tries to minimize the total number of shared links between a client and downloading servers. Through simulation results, we show that our method can keep high throughput regardless the network topology and decrease the number of selected servers compared with random server selection.