Description
Date depot: 1 janvier 1900
Titre: Adaptive overlay technology for media distribution in the OSN context
Encadrant :
Artur HECKER (LTCI (EDMH))
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Non defini
Resumé:
This thesis aims at studies of efficient media distribution architectures in the context of Online Social Networks (OSN).
Recently, implementing OSNs as a distributed system has been proposed as a promising alternative to better preserve user privacy [1][2]. Indeed, just like security, privacy is majorly a question of control; yet, in the centralized environment the authority is necessarily with the OSN platform provider, who alas has non-negligible commercial incentives to exploit user data for other purposes. Specifically, the P2P technology has been used to tackle these issues: here, the OSN is implemented on top of a P2P system [3][4]. Several initiatives are ongoing, the most well-known being, from the academic perspective, Safebook [1] and PeerSON [3], and, from the open-source community, Diaspora [5].
Yet, beyond privacy issues, several additional questions deserve a deeper study. Today, the OSNs are used in different manners, and the usage is not limited to the mere message exchange. Owing to their communitarian nature, the OSNs increasingly become a way for content dissemination – and often not only for dissemination of links to the content stored elsewhere (e.g. to the videos on Youtube), but for the direct dissemination of the user-produced content like photos, videos, sounds, etc. Therefore, such content distribution and streaming capabilities need to be integrated in any modern OSN. That means that a given P2P system realization has to be prepared to efficiently deliver services very different in nature. In other words, the OSN communities and OSN developers should be offered possibilities to bend the underlying P2P technology to their respective needs. We explain that in more details.
Indeed, while the environment, the required services, the available knowledge from the OSN, the participation in the P2P network (as of churn) will dynamically change in operation, the implementation of a participating peer stays the same. Therefore, the same running implementation participates in an actually constantly changing P2P system required to fulfil different services’ demands. To do that, the peer will need to be capable of adapting to the new situation or needs, triggered to do so through some external stimuli (e.g. through a specific reconfiguration command from a content provider in an operated environment, or through a dynamic adaptation in an autonomous setup).
A visible part of such adaptation will happen at the service interface: the respective QoS could be enhanced/degraded (e.g. video could be delivered in a lower resolution because a mobile user has moved into a worse connectivity zone), data security could be changed (e.g. encryption could be activated – in spite of battery considerations – because the new channel is not trustworthy), etc. Such adaptation can be, for most, considered locally by the peer, (pre-)specified by some policies.
Yet, for us, there is a more intriguing question of the best general topology of the P2P network and of the relationship between the latter and:
• the structure of the social graph;
• media transport and delivery;
• the systemic properties like resilience, robustness and security;
• user privacy.
The study of these relationships is the actual goal of this thesis. What is the best strategy for the P2P/OSN merger? What is the best topology for media delivery over a P2P OSN? Should the P2P topology directly follow the OSN contacts (darknets or F2F-like architecture, claimed to be better for privacy [6][7]), or should it rather be used as an independent communication middleware for the OSN (therefore increasing the degrees of freedom in communications, as e.g. in Skype)? Can different service requirements be fulfilled by the same structure? (e.g. should it be different for file transfer and delay-sensitive applications? or maybe even different for video and voice?)
Besides, the overall system resilience – another study goal of this thesis – will also depend on the structural properties of the used P2P graph. Centralized P2P networks (i.e. P2P networks exhibiting high centrality) have very different properties than fully distributed networks, both in terms of routing (average path length is related to the network diameter, local routing table size is related to the node degree) and in terms of error robustness and security (e.g. disturbing nodes with high betweenness has a particularly high impact on the whole structure [8]).
Finally, there seems to be a need for adaptation to the constraints of the underlying network. P2P topology adaptation to the underlying network in general and ISP/P2P collaborations in particular have received a lot of interest in the past [9][10][11]. More recent proposals include Oracle-like architectures, where the ISP essentially aids in the construction of network locality aware P2P structures [12]. While this seems like a very promising way to optimize data transport over the P2P overlay (to shorten delays in s
Doctorant.e: Famulari Antonino