Thomas Locher

eQuus: A Provably Robust and Locality-Aware Peer-to-Peer System

Peer-to-peer systems (p2p) are highly dynamic in nature. They may consist of millions of peers joining only for a limited period of time, resulting in hundreds of join and leave events per second. In this paper we introduce eQuus, a novel distributed hash table (DHT) suitable for highly dynamic environments. eQuus guarantees that lookups are always fast - in terms of both the delay and the total number of routing hops -, although peers may join and leave the network at any time and concurrently

Push-to-pull peer-to-peer live streaming

In contrast to peer-to-peer file sharing, live streaming based on peer-to-peer technology is still awaiting its breakthrough. This may be due to the additional challenges live streaming faces, e.g., the need to meet real-time playback deadlines, or the increased demands on robustness under churn. This paper presents and evaluates novel neighbor selection and data distribution schemes for peer-to-peer live streaming. Concretely, in order to distribute data efficiently and with minimal delay, our algorithms combine low-latency push operations along a structured overlay with the flexibility of pull operations. The protocols ensure that all peers are able to obtain the required data blocks of a live stream in time, and that due to the loop-free dissemination paths, the overhead is low.

Sensor networks continue to puzzle: selected open problems

While several important problems in the field of sensor networks have already been tackled, there is still a wide range of challenging, open problems that merit further attention. We present five theoretical problems that we believe to be essential to understanding sensor networks. The goal of this work is both to summarize the current state of research and, by calling attention to these fundamental problems, to spark interest in the networking community to attend to these and related problems in sensor networks.

Rescuing Tit-for-Tat with Source Coding

BitTorrent has become one of the most popular Internet applications, given the number of users and the fraction of the Internet traffic it consumes. Its wide adoption has exposed some potential problems, like selfish peer behavior. Related research efforts so far have focused on modeling the dynamics of swarms, as well as devising incentive mechanisms that improve fairness without sacrificing efficiency. To the best of our knowledge, this is the first paper to evaluate the impact of attacks that exploit BitTorrent vulnerabilities with the sole intention of harming a swarm. The paper sheds light on BitTorrent behavior by presenting state diagrams, describes two attacks, and then evaluates their negative impact in realistic swarm settings. To evaluate the impact of attacks, a discrete-event simulator was developed and validated against an experimental evaluation performed in a controlled environment. Our findings show the seriousness of the problem and should be the basis for the development of new mechanisms to increase BitTorrent security.Tit-for-tat is widely believed to be the most effective strategy to enforce collaboration among selfish users. However, it has been shown that its usefulness for decentralized and dynamic environments such as peer-to-peer networks is marginal, as peers can rapidly end up in a deadlock situation. Many proposed solutions to this problem are either less resilient to freeloading behavior or induce a computational overhead that cannot be sustained by regular peers. In contrast, we retain tit-for-tat, but enhance the system with a novel form of source coding and an effective scheme to prevent peers from freeloading from seeding peers. We show that our system performs well without the risk of peer starvation and without sacrificing fairness. The proposed solution has a reasonably low overhead, and may hence be suitable for fully distributed content distribution applications in real networks.

Tight bounds for distributed selection

We revisit the problem of distributed <i>k</i>-selection where, given a general connected graph of diameter <i>D</i> consisting of <i>n</i> nodes in which each node holds a numeric element, the goal is to determine the <i>k^th</i> smallest of these elements. In our model, there is no imposed relation between the magnitude of the stored elements and the number of nodes in the graph. We propose a randomized algorithm whose time complexity is <i>O</i>(<i>D</i>log_<i>D</i> <i>n</i>) with high probability. Additionally, a deterministic algorithm with a worst-case time complexity of <i>O</i>(<i>D</i>log2<over><i>D</i> <i>n</i>) is presented which considerably improves the best known bound for deterministic algorithms. Moreover, we prove a lower bound of Ω(<i>D</i> log_<i>D</i><i>n</i>) for any randomized or deterministic algorithm, implying that the randomized algorithm is asymptotically optimal.

Oblivious gradient clock synchronization

We study the gradient clock synchronization (GCS) problem, in which the worst-case clock skew between neighboring nodes has to be minimized. In particular, we consider oblivious clock synchronization algorithms which base their decision on how to adapt the clock solely on the most accurate timing information received from each neighbor. For several intuitive clock synchronization algorithms, which attempt to minimize the skew at all times, we show that the clock skew between neighboring nodes can be significantly larger than the proven lower bound of

Gradient clock synchronization in dynamic networks

\Omega(\frac{\log D}{\log\log D})

Optimal gradient clock synchronization in dynamic networks

, where D denotes the diameter of the network. All of these natural algorithms belong to the class of oblivious clock synchronization algorithms. Additionally, we present an oblivious algorithm with a worst-case skew of

Clock Synchronization with Bounded Global and Local Skew

O(d + \sqrt{D})

Poisoning the Kad network

between any two nodes at distance d.