AFT: Adaptive and fault tolerant peer-to-peer overlay—A user-centric solution for data sharing

Highlights

• AFT: User-centric network for data-sharing.

• Fault Tolerance and Adaptive Peer-to-Peer overlay.

• Comparison of different Peer-to-Peer overlays.

• Performance evaluation of network construction.

Abstract

The widespread of interconnectable computers gives systems the chance to operate more efficiently, by better utilizing the cooperation between individual components. User-centric solutions address the devices themselves and, since there is no network infrastructure and a device powerful enough to assume the role of a coordinator, adopting a peer-to-peer model tends to be the best solution. In this paper we propose AFT, an overlay that adapts to a changing number of nodes, is resilient to faults and is the foundation for an efficient implementation of a reputation based trust system. The AFT overlay is designed to be a solution for systems that need to share transient information, performing a synchronization between various components, like in mobile ad-hoc networks, M2M networks, urban networks, and wireless sensor networks. The operations supported by the overlay, like joining, leaving, unicast transmission, broadcast sharing and maintenance can be accomplished in a duration belonging to $\mathcal{O}\left(\sqrt{N}\right)$, where $N$ is the number of nodes which are part of the structure. We proved these properties and we evaluate the time performance related to overlay creation and node joining.

Introduction

With the rapid adoption of ubiquitous computing, smart devices and computers tend to appear everywhere, under various forms ranging from powerful servers, to smart-phones, and to lightweight sensors. The fact that they are so many and scattered all over the place means that we can increase their usefulness and efficiency if the systems manage to communicate with each other. The challenge becomes even greater as we take into consideration the fact that computers may misbehave, acting in a way that destabilizes the entire system. Such a situation can occur as a result of a malfunction or it can be what the device was actually created for.

While the previous motivation for the use of a distributed system comes naturally, as the components are weakly linked and can enter or exit the network at any given time, there are also other ways to get to a solution based on this type of system. The fact that single computer architectures and even centralized systems scale up to a limit generates demand for alternatives. The mentioned limit may not be an important factor at the beginning, when the number of users is low, but as time passes and more participants come into play it becomes an impediment to growth  [1].

A centralized system has both performance issues and is also more susceptible to failures. Basically, we face a trade-off between simplicity and fault tolerance. In a purely distributed system the role of a given note can be fulfilled by any other peer. This change may generate other modification in the structure of the network (for example, in the overlay-specific naming of the nodes, or for interactive broadcasting  [2]), but after running a specialized procedure the absence of the initial peer does not affect the system any more.

For all applications based on content distribution networks, the decentralization over different P2P overlays improves the efficiency. P2P models and technologies have been successfully used for on-line communities, where users post contents (video, images, text) in order to share them. In these communities, similarity search is supported by semantic overlay networks, where data management, data integration and documents retrieval systems play an important role. An open issue in very large-scale P2P systems refers to how to chose a representative answer for a user query, where there may be a huge number of answers most of them being uninteresting or redundant  [3], [4].

Because applications that require large processing power and scalability vary very much, a system that sits on top of the traditional networking technologies, but below the application layer is needed. Reducing the problem to its bare essentials, what we need is a manner of organizing devices such that they can cooperate with low penalties, that adapts to changes and that enables fast data sharing  [5]. The first and third previously mentioned properties should be analysed relative to centralized models or, by using the transitivity of the better than relation, to other distributed systems that have proven to be successful at managing given issues. Other challenges for Peer-to-Peer overlays are related to interoperability for mobile nodes  [6], optimized of throughput  [7], and performance modelling and evaluation  [8], [9].

The above gives rise to the challenge we address in this study, namely: we seek to organize the computers in a decentralized self-organizing overlay network that can handle failing devices. Such a structure should also adapt to a varying number of peers and assign an equal importance to both unicast and broadcast data traffic. While this will not directly deal with bad intentioned peers, it will provide support to adequate solutions. This is AFT, and Adaptive and Fault-Tolerant peer-to-peer overlay, with nodes virtually linked in a torus, which is designed to be a user-centric solution for data dissemination and sharing. As an example, we can augment the network with a reputation-based trust layer that will filter out disturbing devices. One of the requirement for such a trust system is that the layers below it have to provide a quick manner in which nodes can share some data with all other peers and learn from them at the same time  [10].

To clarify why the previously mentioned property is needed, we analysed the Secured Trust model  [11]. As in many other systems that rely on the ability to compute the reputation of its peers, from time to time this property is recomputed for each node. The way in which this is done is the same as the manner in which we would do it in real life, when talking about the reputation of a person: we would take into account the feedback of others after they interacted with the subject.

At certain moments in time, nodes can synchronize in the matter of how they perceived interactions with the peers they came in contact with. This leads us precisely to the requirement we stated previously: nodes should be able to send and receive data to, respective from, other peers without wasting much time.

The need to handle devices that may, at some point, fail is a key feature of any system that wants to face real-life scenarios. Hardware tends to misbehave after being used a certain period of time and such an unfortunate event should not take down the entire network, either by causing healthy nodes to misbehave or simply by degrading performance to the point of un-usability.

The same is true for adaptability: the overlay should modify its structure depending on the number of nodes that compose it, in order to operate efficiently in the given situation. This also means that performances should degrade as little as possible when increasing the number of peers.

To summarize, the main contributions of this paper are as follows.

• AFT, a user-centric network for data-sharing, structured as a torus, having joining, leaving, unicast transmission, broadcast sharing and maintenance operations with $\mathcal{O}\left(\sqrt{N}\right)$ cost.

• A fault tolerance and adaptive Peer-to-Peer overlay with a specific consolidation procedure, a resilience scheme and a mechanism for contacting a ring from the overlay.

• Comparison of proposed overlay with two other Peer-to-Peer overlays: Chord and HoneyComb.

• Detailed performance evaluation of network construction and theoretical evaluation of operations cost for the proposed overlay.

This rest of the paper is structured in 7 sections as follows. Section  2 presents the related work and existing solutions. Section  3 gives a general description of the proposed design, leaving the next two sections, Section  4 and Section  5, to present the details regarding the important characteristics of the overlay: adaptability and fault tolerance. Aside from formally proving the properties of our solution, in Section  6 we present experimental measurements that reassert these features. Section  7 reveals a comparison between AFT and other Peer-to-Peer overlays. The conclusions and future development ideas are drawn in the final part of the paper, in Section  8.