Publications on Peer-to-Peer

• Peer-Assisted Content Distribution in Akamai NetSession. Mingchen Zhao, Paarijaat Aditya, Ang Chen, Yin Lin, Andreas Haeberlen, Peter Druschel, Bruce Maggs, Bill Wishon, and Miroslav Ponec. 13th ACM SIGCOMM Conference on Internet Measurement (IMC '13), Barcelona, Spain, Oct 2013.
  
  × @inproceedings{zhao-2013-netsession,
    author = {Mingchen Zhao and Paarijaat Aditya and Ang Chen and Yin Lin and Andreas Haeberlen and Peter Druschel and Bruce Maggs and Bill Wishon and Miroslav Ponec},
    title = {Peer-Assisted Content Distribution in {Akamai} {NetSession}},
    booktitle = {13th ACM SIGCOMM Conference on Internet Measurement (IMC '13)},
    location = {Barcelona, Spain},
    month = oct,
    year = 2013
  }
  ×

  Content distribution systems have traditionally adopted one of two architectures: infrastructure-based content delivery networks (CDNs), in which clients download content from dedicated, centrally managed servers, and peer-to-peer CDNs, in which clients download content from each other. The advantages and disadvantages of each architecture have been studied in great detail. Recently, hybrid, or "peer-assisted", CDNs have emerged, which combine elements from both architectures. The properties of such systems, however, are not as well understood.

  In this paper, we discuss the potential risks and benefits of peer-assisted CDNs, and we study one specific instance, Akamai's NetSession system, to examine the impact of these risks and benefits in practice. NetSession is a mature system that has been operating commercially since 2010 and currently has more than 25 million users in 239 countries and territories. Our results show that NetSession can deliver several of the key benefits of both infrastructure-based and peer-to-peer CDNs – for instance, it can offload 70-80% of the traffic to the peers without a corresponding loss of performance or reliability – and that the risks can be managed well. This suggests that hybrid designs may be an attractive option for future CDNs.

  PDF BibTeX Abstract
• Reliable Client Accounting for Hybrid Content-Distribution Networks. Paarijaat Aditya, Mingchen Zhao, Yin Lin, Andreas Haeberlen, Peter Druschel, Bruce Maggs, and Bill Wishon. 9th USENIX Symposium on Networked Systems Design and Implementation (NSDI '12), San Jose, CA, Aug 2012.
  
  × @inproceedings{aditya-2012-rca,
    author = {Paarijaat Aditya and Mingchen Zhao and Yin Lin and Andreas Haeberlen and Peter Druschel and Bruce Maggs and Bill Wishon},
    title = {Reliable Client Accounting for Hybrid Content-Distribution Networks},
    booktitle = {9th USENIX Symposium on Networked Systems Design and Implementation (NSDI '12)},
    location = {San Jose, CA},
    month = aug,
    year = 2012
  }
  ×

  Content distribution networks (CDNs) have started to adopt hybrid designs, which employ both dedicated edge servers and resources contributed by clients. Hybrid designs combine many of the advantages of infrastructure-based and peer-to-peer systems, but they also present new challenges. This paper identifies reliable client accounting as one such challenge. Operators of hybrid CDNs are accountable to their customers (i.e., content providers) for the CDN’s performance. Therefore, they need to offer reliable quality of service and a detailed account of content served. Service quality and accurate accounting, however, depend in part on interactions among untrusted clients. Using the Akamai NetSession client network in a case study, we demonstrate that a small number of malicious clients used in a clever attack could cause significant accounting inaccuracies.

  We present a method for providing reliable accounting of client interactions in hybrid CDNs. The proposed method leverages the unique characteristics of hybrid systems to limit the loss of accounting accuracy and service quality caused by faulty or compromised clients. We also describe RCA, a system that applies this method to a commercial hybrid content-distribution network. Using trace-driven simulations, we show that RCA can detect and mitigate a variety of attacks, at the expense of a moderate increase in logging overhead.

  PDF BibTeX Abstract
• Efficient Replica Maintenance for Distributed Storage Systems. Byung-Gon Chun, Frank Dabek, Andreas Haeberlen, Emil Sit, Hakim Weatherspoon, M. Frans Kaashoek, John Kubiatowicz, and Robert Morris. 3rd Symposium on Networked Systems Design and Implementation (NSDI '06), San Jose, CA, May 2006.
  
  × @inproceedings{chun-2006-maintenance,
    author = {Byung-Gon Chun and Frank Dabek and Andreas Haeberlen and Emil Sit and Hakim Weatherspoon and M. Frans Kaashoek and John Kubiatowicz and Robert Morris},
    title = {Efficient Replica Maintenance for Distributed Storage Systems},
    booktitle = {3rd Symposium on Networked Systems Design and Implementation (NSDI '06)},
    location = {San Jose, CA},
    month = may,
    year = 2006
  }
  ×

  This paper considers replication strategies for storage systems that aggregate the disks of many nodes spread over the Internet. Maintaining replication in such systems can be prohibitively expensive, since every transient network or host failure could potentially lead to copying a server's worth of data over the Internet to maintain replication levels.

  The following insights in designing an efficient replication algorithm emerge from the paper's analysis. First, durability can be provided separately from availability; the former is less expensive to ensure and a more useful goal for many wide-area applications. Second, the focus of a durability algorithm must be to create new copies of data objects faster than permanent disk failures destroy the objects; careful choice of policies for what nodes should hold what data can decrease repair time. Third, increasing the number of replicas of each data object does not help a system tolerate a higher disk failure probability, but does help tolerate bursts of failures. Finally, ensuring that the system makes use of replicas that recover after temporary failure is critical to efficiency.

  Based on these insights, the paper proposes the Carbonite replication algorithm for keeping data durable at a low cost. A simulation of Carbonite storing 1 TB of data over a 365 day trace of PlanetLab activity shows that Carbonite is able to keep all data durable and uses 44% more network traffic than a hypothetical system that only responds to permanent failures. In comparison, Total Recall and DHash require almost a factor of two more network traffic than this hypothetical system.

  PDF BibTeX Abstract
• Experiences in Building and Operating ePOST, a Reliable Peer-to-Peer Application. Alan Mislove, Ansley Post, Andreas Haeberlen, and Peter Druschel. 1st European Conference on Computer Systems (EuroSys '06), Leuven, Belgium, Apr 2006.
  
  × @inproceedings{mislove-2006-epost,
    author = {Alan Mislove and Ansley Post and Andreas Haeberlen and Peter Druschel},
    title = {Experiences in Building and Operating {ePOST}, a Reliable Peer-to-Peer Application},
    booktitle = {1st European Conference on Computer Systems (EuroSys '06)},
    location = {Leuven, Belgium},
    month = apr,
    year = 2006
  }
  ×

  Peer-to-peer (p2p) technology can potentially be used to build highly reliable applications without a single point of failure. However, most of the existing applications, such as file sharing or web caching, have only moderate reliability demands. Without a challenging proving ground, it remains unclear whether the full potential of p2p systems can be realized.

  To provide such a proving ground, we have designed, deployed and operated a p2p-based email system. We chose email because users depend on it for their daily work and therefore place high demands on the availability and reliability of the service, as well as the durability, integrity, authenticity and privacy of their email. Our system, ePOST, has been actively used by a small group of participants for over two years.

  In this paper, we report the problems and pitfalls we encountered in this process. We were able to address some of them by applying known principles of system design, while others turned out to be novel and fundamental, requiring us to devise new solutions. Our findings can be used to guide the design of future reliable p2p systems and provide interesting new directions for future research.

  PDF BibTeX Abstract
• Proactive replication for data durability. Emil Sit, Andreas Haeberlen, Frank Dabek, Byung-Gon Chun, Hakim Weatherspoon, Robert Morris, and M. Frans Kaashoek. 5th International Workshop on Peer-to-Peer Systems (IPTPS '06), Santa Barbara, CA, Feb 2006.
  
  × @inproceedings{sit-2006-tempo,
    author = {Emil Sit and Andreas Haeberlen and Frank Dabek and Byung-Gon Chun and Hakim Weatherspoon and Robert Morris and M. Frans Kaashoek},
    title = {Proactive replication for data durability},
    booktitle = {5th International Workshop on Peer-to-Peer Systems (IPTPS '06)},
    location = {Santa Barbara, CA},
    month = feb,
    year = 2006
  }
  ×

  Many wide-area storage systems replicate data for durability. A common way of maintaining the replicas is to detect node failures and respond by creating additional copies of objects that were stored on failed nodes and hence suffered a loss of redundancy. Reactive techniques can minimize total bytes sent since they only create replicas as needed; however, they can create spikes in network use after a failure. These spikes may overwhelm application traffic and can make it difficult to provision bandwidth.

  This paper explores a proactive approach that creates additional copies not in response to failures, but periodically at a fixed low rate. We introduce Tempo, a distributed hash table that allows each user to specify a maximum maintenance bandwidth and uses it to perform proactive replication. Results from a simulation study suggest that Tempo can deliver high durability despite only using several kilobytes per second of bandwidth, comparable to state-of-the-art reactive systems.

  PDF BibTeX Abstract
• Fallacies in evaluating decentralized systems. Andreas Haeberlen, Alan Mislove, Ansley Post, and Peter Druschel. 5th International Workshop on Peer-to-Peer Systems (IPTPS '06), Santa Barbara, CA, Feb 2006.
  
  × @inproceedings{haeberlen-2006-fallacies,
    author = {Andreas Haeberlen and Alan Mislove and Ansley Post and Peter Druschel},
    title = {Fallacies in evaluating decentralized systems},
    booktitle = {5th International Workshop on Peer-to-Peer Systems (IPTPS '06)},
    location = {Santa Barbara, CA},
    month = feb,
    year = 2006
  }
  ×

  Research on decentralized systems such as peer-to-peer overlays and ad hoc networks has been hampered by the fact that few systems of this type are in production use, and the space of possible applications is still poorly understood. As a consequence, new ideas have mostly been evaluated using common synthetic workloads, traces from a few existing systems, testbeds like PlanetLab, and simulators like ns-2. Some of these methods have, in fact, become the "gold standard" for evaluating new systems, and are often a prerequisite for getting papers accepted at top conferences in the field.

  In this paper, we examine the current practice of evaluating decentralized systems under these specific sets of conditions and point out pitfalls associated with this practice. In particular, we argue that (i) despite authors' best intentions, results from such evaluations often end up being inappropriately generalized; (ii) there is an incentive not to deviate from the accepted standard of evaluation, even if that is technically appropriate; (iii) research may gravitate towards systems that are feasible and perform well when evaluated in the accepted environments; and, (iv) in the worst-case, research may become ossified as a result. We close with a call to action for the community to develop tools, data, and best practices that allow systems to be evaluated across a space of workloads and environments.

  PDF BibTeX Abstract
• ePOST. Alan Mislove, Andreas Haeberlen, Ansley Post, and Peter Druschel. Chapter in Peer-to-Peer Systems and Applications, LNCS 3485, Springer Verlag, Ralf Steinmetz and Klaus Wehrle, Eds., Aug 2005.
  
  × @incollectionmislove-2005-epostchapter,
    author = {Alan Mislove and Andreas Haeberlen and Ansley Post and Peter Druschel},
    title = {{ePOST}},
    booktitle = {Peer-to-Peer Systems and Applications, LNCS 3485},
    publisher = {Springer Verlag},
    editor = {Ralf Steinmetz and Klaus Wehrle},
    month = aug,
    year = 2005
  }
  BibTeX Book Software
• Consistent Key Mapping in Structured Overlays. Andreas Haeberlen, Jeff Hoye, Alan Mislove, and Peter Druschel. Technical report TR05-456, Department of Computer Science, Rice University, Aug 2005.
  
  × @techreport{haeberlen-2005-consistent,
    author = {Andreas Haeberlen and Jeff Hoye and Alan Mislove and Peter Druschel},
    title = {Consistent Key Mapping in Structured Overlays},
    institution = {Department of Computer Science, Rice University},
    number = {TR05-456},
    month = aug,
    year = 2005
  }
  ×

  Most structured peer-to-peer overlays rely on consistent hashing to determine the node that is responsible for a given key. For consistent hashing to work properly, it is necessary that the nodes have a consistent view of their neighborhood in the identifier space. However, if routing anomalies occur in the underlying network, this view can become inconsistent, causing unstable overlay behavior and, worse, allowing more than one node to assume responsibility for ranges of keys.

  We present a set of techniques for preventing inconsistencies under routing anomalies, and we propose to adopt strategies from mobile ad-hoc networking for maintaining connectivity in the presence of path failures. We evaluate our design in the context of Pastry and present results from a deployment in the PlanetLab testbed.

  PDF BibTeX Abstract
• Glacier: Highly durable, decentralized storage despite massive correlated failures. Andreas Haeberlen, Alan Mislove, and Peter Druschel. nd Symposium on Networked Systems Design and Implementation (NSDI '05), Boston, MA, May 2005.
  
  × @inproceedings{haeberlen-2005-glacier,
    author = {Andreas Haeberlen and Alan Mislove and Peter Druschel},
    title = {Glacier: {H}ighly durable, decentralized storage despite massive correlated failures},
    booktitle = {nd Symposium on Networked Systems Design and Implementation (NSDI '05)},
    location = {Boston, MA},
    month = may,
    year = 2005
  }
  ×

  Decentralized storage systems aggregate the available disk space of participating computers to provide a large storage facility. These systems rely on data redundancy to ensure durable storage despite of node failures. However, existing systems either assume independent node failures, or they rely on introspection to carefully place redundant data on nodes with low expected failure correlation. Unfortunately, node failures are not independent in practice and constructing an accurate failure model is difficult in large-scale systems. At the same time, malicious worms that propagate through the Internet pose a real threat of large-scale correlated failures. Such rare but potentially catastrophic failures must be considered when attempting to provide highly durable storage.

  In this paper, we describe Glacier, a distributed storage system that relies on massive redundancy to mask the effect of large-scale correlated failures. Glacier is designed to aggressively minimize the cost of this redundancy in space and time: Erasure coding and garbage collection reduces the storage cost; aggregation of small objects and a loosely coupled maintenance protocol for redundant fragments minimizes the messaging cost. In one configuration, for instance, our system can provide six-nines durable storage despite correlated failures of up to 60% of the storage nodes, at the cost of an eleven-fold storage overhead and an average messaging overhead of only 4 messages per node and minute during normal operation. Glacier is used as the storage layer for an experimental serverless email system.

  PDF BibTeX Abstract Slides Software

Show all publications