A Hierarchical Internet Path Available Bandwidth Inference Mechanism

Abstract

Distributed applications such as cloud computing and CDN need the Internet state information to optimize their performance. Although the present network coordinate systems are effective in latency inference, they cannot infer performance metric such as available bandwidth. Therefore, this paper attempts to use a novel mechanism to concentrate on how to infer the Internet path available bandwidth. Based on theoretical and Internet measurement data analysis result, we discover that Internet path available bandwidth has ultra metric space feature, based on which we propose an available bandwidth inference mechanism called PathGuru. In PathGuru, each node uses the partial measurement data to and from landmarks to obtain the outgoing and the incoming bandwidth vectors, and then the available bandwidth between any two nodes can be inferred according to the two vectors as well as the ultra metric space constraint. Experimental results on the Internet measurement dataset indicate that PathGuru has reasonable inference accuracy and can also deal with the available bandwidth symmetry violation problem.