Through our PhD student Roberto Roverso we are collaborating with Peerialism. Hive Streaming, as a result of this collaboration, is a solution for large scale distribution of streaming video; live and on-demand. The product enables customers to offload up to 98% of the traffic from their streaming servers and core parts of the network. In addition to this system, a number of research papers and demos were published, i.e., On HTTP Live Streaming in Large Enterprises (SIGCOMM 2013), Through the Wormhole: Low Cost, Fresh Peer Sampling for the Internet (P2P 2013), HTTP-Live Streaming Goes Peer-To-Peer (Networking 2012), DTL: Dynamic Transport Library for Peer-To-Peer Applications (ICDCN 2012), and Peer2View: a Peer-To-Peer HTTP-Live Streaming platform (P2P 2012).

Through our PhD student Raul Jimenez, we are collaborating with Spotify. The result of this work was a paper submitted to IEEE INFOCOM 2014 (currently under review). The paper explores the challenges of integrating mobile devices on P2P networks. In particular, energy consumption is empirically studied by enabling P2P download on Spotify's Android app. When P2P is enabled, the experiments show a stark contrast between 3G and Wifi, with a dramatic increase in energy consumption on 3G and a much smaller one on Wifi. We implement a backwards-compatible protocol modification that reduces energy consumption to levels close to the standard non-P2P Spotify app. This collaboration not only produced research results but also uncovered energy inefficiencies on the app which are currently being addressed at Spotify.

recorded future   gavagai
Through our PhD student Fatemeh Rahimian we are collaborating with Recorded Future and Gavagai. The result of this work is a paper that is going to be submitted to IPDPS 2014. This paper proposes a solution to large-scale Word Sense Disambiguation (WSD), which is currently a very challenging task for both companies. With the exponential growth of data over the current web, most of the existing algorithms for WSD render infeasible. Our idea is to transforms the WSD problem to a well-known graph problem, i.e., community detection in graphs. This problem is NP-Hard and also extremely data-intensive. However, we propose a highly parallel heuristic-based algorithm that can solve the problem efficiently. Our solution easily scales up to deal with millions of documents and billions of words, by making use of the recent developments in graph processing frameworks. We show that the accuracy of our result outperforms that of the state of the art, while the convergence time is relatively low.