Overall Objectives
Scientific Foundations
Application Domains
New Results
Contracts and Grants with Industry
Other Grants and Activities

Section: New Results

Resource management in large-scale dynamic systems

Heterogeneous gossip

Participants : Davide Frey, Anne-Marie Kermarrec.

This work has been done in collaboration with Vivien Quéma (CNRS Grenoble), Maxime Monod and Rachid Guerraoui (EPFL). Gossip-based information dissemination protocols are considered easy to deploy, scalable and resilient to network dynamics. Load-balancing is inherent in these protocols as the dissemination work is evenly spread among all nodes. Yet, large-scale distributed systems are usually heterogeneous with respect to network capabilities such as bandwidth. In practice, a blind load-balancing strategy might significantly hamper the performance of the gossip dissemination.

These observations have led to the development of HEAP, HEterogeneity-Aware gossip Protocol , where nodes dynamically adapt their contribution to the gossip dissemination according to their bandwidth capabilities. Using a continuous, itself gossip-based, approximation of relative bandwidth capabilities, HEAP dynamically leverages the most capable nodes by increasing their fanout, while decreasing by the same proportion those of less capable nodes. HEAP preserves the simplicity and proactive (churn adaptation) nature of gossip, while significantly improving its effectiveness. We extensively evaluated HEAP in the context of a video streaming application on a 270 PlanetLab node testbed. Our results show that HEAP significantly improves the perceived quality of the streaming over standard gossip protocols, especially when the stream rate is close to the average available bandwidth. This work has been published at DSN 2009 [43] and Middleware [42] .


Participants : Kévin Huguenin, Anne-Marie Kermarrec.

This work has been done in collaboration with Maxime Monod and Rachid Guerraoui (EPFL). LiFTinG is the first protocol to detect freeriders, including colluding ones, in gossip-based content dissemination systems with asymmetric data exchanges. LiFTinG relies on nodes tracking abnormal behaviors by cross-checking the history of their previous interactions, and exploits the fact that nodes pick neighbors at random to prevent colluding nodes from covering each other up. We present extensive analytical evaluations of LiFTinG, backed up by simulations and PlanetLab experiments. In a 300-node system, where a stream of 674 kbps is broadcast, LiFTinG incurs a maximum overhead of only 8%.With 10% of freeriders decreasing their contribution by 30%, LiFTinG detects 86% of the freeriders after only 30 seconds and wrongfully expels only a few honest nodes.


Participants : Kévin Huguenin, Anne-Marie Kermarrec, Nicolas Le Scouarnec.

This work has been done in collaboration with Mary-Luc Champel (Thomson Labs). Network coding has been successfully applied in large-scale content dissemination systems. While network codes provide optimal throughput, its current forms suffer from a high decoding complexity. This is an issue when applied to systems composed of nodes with low processing capabilities.

We propose a novel network coding approach based on LT codes [103] , initially introduced in the context of erasure coding. Our coding scheme, called LTNC, fully benefits from the low complexity of belief propagation decoding. Yet, such decoding schemes are extremely sensitive to statistical properties of the code. Maintaining such properties in a fully decentralized way with only a subset of encoded data is challenging. This is precisely what the recoding algorithms of LTNC achieve.

We evaluate LTNC against random linear network codes in an epidemic content-dissemination application. Results show that LTNC slightly increases communication overhead (20%) and convergence time (30%) but greatly reduces the decoding complexity (99%) when compared to random linear network codes. In addition, LTNC consistently outperforms dissemination protocols without codes, thus preserving the benefit of coding.


Participants : Anne-Marie Kermarrec, Nicolas Le Scouarnec.

Phosphite [38] is a mechanism to preserve out of order download in peer to peer video-on-demand applications, in the presence of selfish peers. In such applications, peers have a natural trend to download blocks in order to start watching videos as soon as possible. Without specific mechanism to enforce a fair amount of out of order download, the last blocks of the video tend to be lost due to peers leaving soon after having downloaded the last blocks thus forcing peers to rely on the central server for re-introducing those lost blocks. This issue can be solved if peers dedicate a portion of their bandwidth for out of order downloads. Yet, this heavily relies on the goodwill of peers to collaborate. Phosphite is a simple yet efficient approach ensuring that all peers dedicate a part of their bandwidth to out of order download. Phosphite relies on a computational challenge where peers are provided with a combination of the requested blocks and other blocks. This forces peers to download out of order blocks to be able to decode the requested blocks. We evaluate Phosphite and show that it successfully prevents the system from losing blocks, even in the presence of selfish peers, thus offering an appealing alternative to state of the art approaches. With Phosphite, the last blocks remain available (with a probability higher than 0.98), while such result cannot be guaranteed (with a probability lower than 0.5) without enforcement mechanism. Phosphite ensures that a peer to peer download is almost always possible, even in the presence of selfish peers.

This work has been done in collaboration with Mary-Luc Champel. A preliminary paper has been accepted at the French conference Algotel 2009 [37] and the final version of the paper has been accepted at the international conference P2P 2009 [38] .

Distributed channel switching in P2P-IPTV

Participants : Erwan Le Merrer, Anne-Marie Kermarrec.

It is now common for IPTV systems attracting millions of users to be based on a peer-to-peer architecture. In such systems, each channel is typically associated with one P2P overlay network connecting the users. Yet, the joining process resulting in a peer to be integrated in channel overlay usually requires a significant amount of time. As a consequence, switching from one channel to another is far to be as fast as in IPTV solutions provided by telco operators. In this work, we tackle the issue of efficient channel switching in P2P IPTV system. We formulate the switching problem, and propose a simple distributed algorithm, as an illustration of the concept, which aims at leveraging the presence of peers in the network to fasten the switch process. This work has been published at Euro-Par 2009 [94] .

Combining gossip and fountain codes for fast content dissemination

Participants : Anne-Marie Kermarrec, Nicolas Le Scouarnec.

Gossip-based dissemination reaches all nodes with high probability. However, some peers may not receive all the data. In order to enhance the performance of gossip-based dissemination protocol, fountain codes can be used to recover the missing data. We propose a protocol that combine a gossip-based dissemination protocol and fountain codes. The gossip protocol builds a structured split-graph overlay to split the peers between encoders and forwarders. Forwarders become encoders as soon as they have received the whole content so they can start encoding. In order to benefit even further from encoders, we deliberately bias the dissemination process through the gossip protocol so that forwarders become earlier potential encoders. We implemented this protocol on the PlanetLab testbed. By simulation in PeerSim, we assess that this protocol outperforms a simple protocol that does not introduce a bias or that does not update the overlay to match the peers'role.

Peer-to-peer collaborative back-up

Participants : Anne-Marie Kermarrec, Fabrice Le Fessant.

The storage capacity of computers has increased a lot in the past years: in the meantime, final users have started using this storage for important personal data, with the democratization of digital cameras, and professional data with the rise of telecomputing. Backuping all this data has become a new challenge for peer-to-peer systems, since these users are connected most of the time, often with large unused storage capacity on their disks, and unfortunately seldom take the time to properly save these important data.

Anne-Marie Kermarrec and Fabrice Le Fessant are currently designing a platform for a collaborative backup system, and this problem tackles a large set of problems: making the backup resilient to the large number of failures characterizing peer-to-peer networks, choosing where to backup the data, designing the protocols to place and retrieve the data from the network, while ensuring secrecy/privacy of the data. The prototype, currently developed by Fabrice Le Fessant within the Peerple open-source project, uses both a structured overlay, to localize stored data during restoration, and an unstructured overlay, to query for storage availability among neighbors. Contrary to most peer-to-peer backup systems, files are not stored separately on the overlay network, but gathered in volumes, encrypted using strong cryptography for privacy, and replicated using Reed-Solomon coding, to ensure availability even in the presence of high failure rates at a minimal extra storage cost. This work is done in collaboration with Laurent Viennot from the Gang project-team, INRIA Paris - Rocquencourt.

First results have been published in DAMAP'09[30] .

Measuring availability in peer-to-peer systems

Participants : Anne-Marie Kermarrec, Fabrice Le Fessant.

Tracking peer availability in a peer-to-peer network is of utmost importance for many collaborative applications. For instance, such information is invaluable for identifying the most stable peers or group of peers with similar uptime characteristics. However, as many applications tend to reward the most stable peers, there is a clear incentive for peers to try to appear more available than their real availability. We developed a scalable and lightweight protocol that enables nodes to measure the peer availability in the presence of such selfish peers. In our protocol, which is called Pacemaker each peer is in charge of maintaining proofs of its own availability over time by collecting pulses disseminated by a trusted entity using asymmetric cryptographic signatures. Essentially, using these pulses, peers gain the ability to challenge other peers and verify that their real uptime matches the advertised one. Simulation results show that our protocol provides accurate availability measures even in the presence of selfish peers. Furthermore, our results are verified by experiments in Planetlab, which also illustrates the deployability of Pacemaker in real networks.

Finding good partners in availability-aware p2p networks

Participants : Fabrice Le Fessant, Erwan Le Merrer.

We studied the problem of finding peers matching a given availability pattern in a peer-to-peer (P2P) system. Motivated by practical examples, we specified two formal problems of availability matching that arise in real applications: disconnection matching, where peers look for partners expected to disconnect at the same time, and presence matching, where peers look for partners expected to be online simultaneously in the future. As a scalable and inexpensive solution, we proposed to use epidemic protocols for topology management; we provided corresponding metrics for both matching problems. We evaluated this solution by simulating two P2P applications, task scheduling and file storage, over a new trace of the eDonkey network, the largest available with availability information. We first proved the existence of regularity patterns in the sessions of 14M peers over 27 days. We also showed that, using only 7 days of history, a simple predictor could select predictable peers and successfully predicted their online periods for the next week. Finally, simulations showed that our simple solution provided good partners fast enough to match the needs of both applications, and that consequently, these applications performed as efficiently at a much lower cost. We believe that this work will be useful for many P2P applications for which it has been shown that choosing good partners, based on their availability, drastically improves their performance and stability. This work has been done in collaboration with Stevens Leblond from INRIA Sophia-Antipolis.

Our results have been published in SSS'09 [61] , CFSE'09 [70] , RR-6795 [88] .


Logo Inria