Were You There? Bridging the Gap to Unveil Users' Online Sessions in Networked, Distributed Systems

The comprehension of users' behavior is paramount for evaluating improvements to networked, distributed systems. To this end, several strategies have been proposed to obtain traces, based on the capture of usage information, which can then serve for evaluation purposes. One main strategy consists of taking snapshots of online users, using instrumented clients. In spite of its popularity, related proposals have fallen short in ensuring accuracy of obtained data. For a variety of reasons, users may fail to appear in some snapshots, although online. In this paper, we propose a methodology to correct ill-collected snapshots and build more accurate traces from them. In summary, we estimate the probability that a given snapshot is missing some users. The snapshot is corrected if the probability exceeds a given threshold. We use ground-truth data to assess the effectiveness of our methodology. The impact of our proposal is evidenced by means of an analysis of traces from a BitTorrent swarm.