Dumpling: An Efficient Method Based on Clustering for Network Positioning

The prediction for network distance, as a function of network coordinate system, has already been proved by existing works. This paper, presents Dumpling, a decentralized network coordinate system for Internet distance prediction. It combines the mechanism of physical mass spring system and the clustering method in the internet which based on inter-host latencies. We have implemented the Dumpling algorithm and evaluated it through simulations. By performing experiment on the datasets collected from Planetlab, we show that Dumpling can predict network distances quickly and accurately, it also can protect the virtual network structure in the coordinate space and reduce the number of system oscillation.