Andri Lareida

From SimpleWiki
Jump to navigationJump to search
Summary
Student: Andri Lareida
Title: Ph.D. Student
e-mail: lareida@ifi.uzh.ch
Affiliation: University of Zurich, CSG group
Supervisor: Prof. Dr. Burkhard Stiller
Start: 1 July 2012
End: 31 December 2015
Funding: UZH / FLAMINGO

Biography

Andri Lareida is a junior researcher and Ph.D. candidate at the Communications Systems Group since July 2012 supervised by Prof. Dr. Burkhard Stiller. He received his Master’s Degree (M. Sc.) in Informatics in 2012 from the University of Zurich where he also completed his Bachelor studies. During his Master Thesis he was implementing and evaluating B-Tracker on a BitTorrent client. Andri Lareida currently works the research projects: Management of the Future Internet (FLAMINGO) and Socially-aware Management of New Overlay Application Traffic combined with Energy Efficiency in the Internet (SmartenIT). He has experience in software engineering in academia as well as in industry from the AMAAIS project and his former position at IBM Switzerland. Andri Lareida’s research interest mainly lies with decentralized (P2P) systems and their optimization, focusing on exploiting locality in overlay networks and using information present in networks to predict user behavior.

PhD project description

Although, video streaming is causing the main part of internet traffic, peer-to-peer (P2P) file sharing applications still account for a large portion of the total internet traffic. The most prominent among these applications is BitTorrent. Internet Service Providers (ISPs) face the challenge of network congestion during peak traffic hours, which is made more difficult by this P2P traffic. Furthermore, the overlay networks used by file sharing applications, such as BitTorrent, cause unnecessary inter-domain traffic because they are typically not aware of the underlying physical network structure. While several solutions targeting the reduction of inter-domain traffic have been proposed by the research community, the reduction of P2P traffic peaks during rush hour has received little to no attention. Solutions to improve locality and reduce inter-domain traffic of P2P systems rely on services offered by third parties. However recent research shows that potential locality is present in BitTorrent but not exploited. The reason might be that the service required for the solutions are not available because they do not scale or deployment cost is too high. Therefore, the hereby proposed project, named LoCoRe (Locality Aware Cooperative Recommender), will pursue a different approach. The major difference to the existing locality approaches is the absence of any central service, this is made possible by the integration B-Tracker, which makes peers responsible for tracking their shared files. This approach scales with very large swarms since every member of the swarm shares a portion of the tracking overhead. To exploit the hidden locality, the peers will do network measurements, such as trace route, to determine which are the best peers to connect to. Furthermore, when receiving a tracking request, peers will select the best neighbors to reply to. To reduce traffic peaks a recommender system will be designed that identifies relevant content and prefetches it, at least partly, to a disk attached to a home router (UNaDa) when Internet traffic is low (e.g. at night). The content will be discovered by asking neighbors for their content and then ranking them according to similarity to the own content. Furthermore, when prefetching content the availability of that content inside one domain will be considered to not create another copy of content if it is already highly available. This cooperative caching approach works best if the system is deployed on home routers with attached disks, since they are typically always on and more energy efficient than PCs. For the evaluation of LoCoRe detailed real life traces of many BitTorrent swarms over a longer period of time (at least 1 month) will be necessary. Unfortunately, this kind of data does not exist yet and will have to be collected. A proof of concept prototype has already been tested and the results gave insights on how to design such a measurement system for a larger scale. Traces of this type are invaluable for evaluation research since they reflect user behavior (when user x downloaded which file) and can be reused in other projects. The advantages of LoCoRe are, besides the main goals of reduced traffic peaks and inter-domain traffic, the easy deployment which does not require new server deployments. Three deployment strategies, which can be pursued in parallel, are possible. First, end users with high technical skill can install the LoCoRe software on their open home router system, e.g. Open-WRT. Second, the ISP integrates LoCoRe in its customized router firmware. Third, the vendor of the router hardware integrates it in the standard firmware, like they do with BitTorrent clients. As long as the same protocol is used, the three strategies are compatible. LoCoRe will produce beneficial effects even if it is not installed in every Internet access connection. The expected results of the LoCoRe project are a set of mechanisms, algorithms, and deployment scenarios which significantly decrease inter-domain traffic and noticeably reduce traffic peaks. Based on the former research on locality in P2P systems the potential is high. Furthermore, the real life traces will be an early result of the project which can be reused by the research community.

Recent publications

  1. Lareida, A., Bocek, T., Waldburger, M., & Stiller, B. (2013, May). RB-tracker: A fully distributed, replicating, network-, and topology-aware P2P CDN. In Integrated Network Management (IM 2013), 2013 IFIP/IEEE International Symposium on (pp. 1199-1202). IEEE.
  2. Lareida, A., Bocek, T., Golaszewski, S., Luthold, C., & Weber, M. (2013, September). Box2Box-A P2P-based file-sharing and synchronization application. In Peer-to-Peer Computing (P2P), 2013 IEEE Thirteenth International Conference on (pp. 1-2). IEEE.

You can find a complete list of publications here.

External links