Artificial immune system for the Internet

We investigate the usability of the Artificial Immune Systems (AIS) approach for solving selected problems in computer networks. Artificial immune systems are created by using the concepts and algorithms inspired by the theory of how the Human Immune System (HIS) works. We consider two applications: detection of routing misbehavior in mobile ad hoc networks, and email spam filtering. In mobile ad hoc networks the multi-hop connectivity is provided by the collaboration of independent nodes. The nodes follow a common protocol in order to build their routing tables and forward the packets of other nodes. As there is no central control, some nodes may defect to follow the common protocol, which would have a negative impact on the overall connectivity in the network. We build an AIS for the detection of routing misbehavior by directly mapping the standard concepts and algorithms used for explaining how the HIS works. The implementation and evaluation in a simulator shows that the AIS mimics well most of the effects observed in the HIS, e.g. the faster secondary reaction to the already encountered misbehavior. However, its effectiveness and practical usability are very constrained, because some particularities of the problem cannot be accounted for by the approach, and because of the computational constrains (reported also in AIS literature) of the used negative selection algorithm. For the spam filtering problem, we apply the AIS concepts and algorithms much more selectively and in a less standard way, and we obtain much better results. We build the AIS for antispam on top of a standard technique for digest-based collaborative email spam filtering. We notice un advantageous and underemphasized technological difference between AISs and the HIS, and we exploit this difference to incorporate the negative selection in an innovative and computationally efficient way. We also improve the representation of the email digests used by the standard collaborative spam filtering scheme. We show that this new representation and the negative selection, when used together, improve significantly the filtering performance of the standard scheme on top of which we build our AIS. Our complete AIS for antispam integrates various innate and adaptive AIS mechanisms, including the mentioned specific use of the negative selection and the use of innate signalling mechanisms (PAMP and danger signals). In this way the AIS takes into account users' profiles, implicit or explicit feedback from the users, and the bulkiness of spam. We show by simulations that the overall AIS is very good both in detecting spam and in avoiding misdetection of good emails. Interestingly, both the innate and adaptive mechanisms prove to be crucial for achieving the good overall performance. We develop and test (within a simulator) our AIS for collaborative spam filtering in the case of email communications. The solution however seems to be well applicable to other types of Internet communications: Internet telephony, chat/sms, forum, news, blog, or web. In all these cases, the aim is to allow the wanted communications (content) and prevent those unwanted from reaching the end users and occupying their time and communication resources. The filtering problems, faced or likely to be faced in the near future by these applications, have or are likely to have the settings similar to those that we have in the email case: need for openness to unknown senders (creators of content, initiators of the communication), bulkiness in receiving spam (many recipients are usually affected by the same spam content), tolerance of the system to a small damage (to small amounts of unfiltered spam), possibility to implicitly or explicitly and in a cheap way obtain a feedback from the recipients about the damage (about spam that they receive), need for strong tolerance to wanted (non-spam) content. Our experiments with the email spam filtering show that our AIS, i.e. the way how we build it, is well fitted to such problem settings.

Signaling and Reciprocity

Wojciech Galuba

Complex networks exist for a number of purposes. The neural, metabolic and food networks ensure our survival, while the social, economic, transportation and communication networks allow us to prosper. Independently of the purposes and particularities of the physical embodiment of the networks, one of their fundamental functions is the delivery of information from one part of the network to another. Gossip and diseases diffuse in the social networks, electrochemical signals propagate in the neural networks and data packets travel in the Internet. Engineering networks for robust information flows is a challenging task. First, the mechanism through which the network forms and changes its topology needs to be defined. Second, within a given topology, the information must be routed to the appropriate recipients. Third, both the network formation and the routing mechanisms need to be robust against a wide spectrum of failures and adversaries. Fourth, the network formation, routing and failure recovery must operate under the resource constraints, either intrinsic or extrinsic to the network. Finally, the autonomously operating parts of the network must be incentivized to contribute their resources to facilitate the information flows. This thesis tackles the above challenges within the context of several types of networks: 1) peer-to-peer overlays – computers interconnected over the Internet to form an overlay in which participants provide various services to one another, 2) mobile ad-hoc networks – mobile nodes distributed in physical space communicating wirelessly with the goal of delivering data from one part of the network to another, 3) file-sharing networks – networks whose participants interconnect over the Internet to exchange files, 4) social networks – humans disseminating and consuming information through the network of social relationships. The thesis makes several contributions. Firstly, we propose a general algorithm, which given a set of nodes embedded in an arbitrary metric space, interconnects them into a network that efficiently routes information. We apply the algorithm to the peer-to-peer overlays and experimentally demonstrate its high performance, scalability as well as resilience to continuous peer arrivals and departures. We then shift our focus to the problem of the reliability of routing in the peer-to-peer overlays. Each overlay peer has limited resources and when they are exhausted this ultimately leads to delayed or lost overlay messages. All the solutions addressing this problem rely on message redundancy, which significantly increases the resource costs of fault-tolerance. We propose a bandwidth-efficient single-path Forward Feedback Protocol (FFP) for overlay message routing in which successfully delivered messages are followed by a feedback signal to reinforce the routing paths. Internet testbed evaluation shows that FFP uses 2-5 times less network bandwidth than the existing protocols relying on message redundancy, while achieving comparable fault-tolerance levels under a variety of failure scenarios. While the Forward Feedback Protocol is robust to message loss and delays, it is vulnerable to malicious message injection. We address this and other security problems by proposing Castor, a variant of FFP for mobile ad-hoc networks (MANETs). In Castor, we use the same general mechanism as in FFP; each time a message is routed, the routing path is either enforced or weakened by the feedback signal depending on whether the routing succeeded or not. However, unlike FFP, Castor employs cryptographic mechanisms for ensuring the integrity and authenticity of the messages. We compare Castor to four other MANET routing protocols. Despite Castor's simplicity, it achieves up to 40% higher packet delivery rates than the other protocols and recovers at least twice as fast as the other protocols in a wide range of attacks and failure scenarios. Both of our protocols, FFP and Castor, rely on simple signaling to improve the routing robustness in peer-to-peer and mobile ad-hoc networks. Given the success of the signaling mechanism in shaping the information flows in these two types of networks, we examine if signaling plays a similar crucial role in the on-line social networks. We characterize the propagation of URLs in the social network of Twitter. The data analysis uncovers several statistical regularities in the user activity, the social graph, the structure of the URL cascades as well as the communication and signaling dynamics. Based on these results, we propose a propagation model that accurately predicts which users are likely to mention which URLs. We outline a number of applications where the social network information flow modelling would be crucial: content ranking and filtering, viral marketing and spam detection. Finally, we consider the problem of freeriding in peer-to-peer file-sharing applications, when users can download data from others, but never reciprocate by uploading. To address the problem, we propose a variant of the BitTorrent system in which two peers are only allowed to connect if their owners know one another in the real world. When the users know which other users their BitTorrent client connects to, they are more likely to cooperate. The social network becomes the content distribution network and the freeriding problem is solved by leveraging the social norms and reciprocity to stabilize cooperation rather than relying on technological means. Our extensive simulation shows that the social network topology is an efficient and scalable content distribution medium, while at the same time provides robustness to freeriding.

EPFL2011

Routing and search on large scale networks

Dominique Florian Tschopp

In this thesis, we address two seemingly unrelated problems, namely routing in large wireless ad hoc networks and comparison based search in image databases. However, the underlying problem is in essence similar and we can use the same strategy to attack those two problems. In both cases, the intrinsic complexity of the problem is in some sense low, and we can exploit this fact to design efficient algorithms. A wireless ad hoc network is a communication network consisting of wireless devices such as for instance laptops or cell phones. The network does not have any fixed infrastructure, and hence nodes which cannot communicate directly over the wireless medium must use intermediate nodes as relays. This immediately raises the question of how to select the relay nodes. Ideally, one would like to find a path from the source to the destination which is as short as possible. The length of the found path, also called route, typically depends on how much signaling traffic is generated in order to establish the route. This is the fundamental trade-off that we will investigate in this thesis. As mentioned above, we try and exploit the fact that the communication network is intrinsically low-dimensional, or in other words has low complexity. We show that this is indeed the case for a large class of models and that we can design efficient algorithms for routing that use this property. Low dimensionality implies that we can well embed the network in a low-dimensional space, or build simple hierarchical decompositions of the network. We use both those techniques to design routing algorithms. Comparison based search in image databases is a new problem that can be defined as follows. Given a large database of images, can a human user retrieve an image which he has in mind, or at least an image similar to that image, without going sequentially through all images? More precisely, we ask whether we can search a database of images only by making comparisons between images. As a case in point, we ask whether we can find a query image q only by asking questions of the type "does image q look more like image A or image B"? The analogous to signaling traffic for wireless networks would here be the questions we can ask human users in a learning phase anterior to the search. In other words, we would like to ask as few questions as possible to pre-process and prepare the database, while guaranteeing a certain quality of the results obtained in the search phase. As the underlying image space is not necessarily metric, this raises new questions on how to search spaces for which only rank information can be obtained. The rank of A with respect to B is k, if A is B's kth nearest neighbor. In this setup, low-dimensionality is analogous to the homogeneity of the image space. As we will see, the homogeneity can be captured by properties of the rank relationships. In turn, homogeneous spaces can be well decomposed hierarchically using comparisons. Further, it allows us to design good hash functions. To design efficient algorithms for these two problems, we can apply the same techniques mutatis mutandis. In both cases, we relied on the intuition that the problem has a low intrinsic complexity, and that we can exploit this fact. Our results come in the form of simulation results and asymptotic bounds.

EPFL2010

A cross-layer design of wireless ad-hoc networks

We consider a cross-layer design of wireless ad-hoc networks. Traditional networking approaches optimize separately each of the three layers: physical layer, medium access and routing. This may lead to largely suboptimal network designs. In this work, we propose a jointly optimal design of the three layers, and we show a significant performance improvement over the conventional approach. In the first part of this thesis, our goal is to select appropriate performance metrics for the joint optimization problem. To that respect, we analyze several existing rate-maximization performance metrics for wireless ad-hoc networks: maximizing the sum of rates, max-min fairness and proportional fairness. We first show with several examples that it is not clear if and how max-min fairness can be defined on each one of the examples. We give a formal proof that the max-min fair rate allocation exists on a large class of sets of feasible rates, among which are the feasible-rate sets of all known ad-hoc networking examples. We also give a centralized algorithm to compute the max-min fair allocation whenever it exists. Next, we compare the three metrics for ad-hoc scenarios in terms of efficiency and fairness. We prove that, similar to wired networking,maximizing the sum of rates leads to gross unfairness and starvation of all but the flows with the best channel conditions. We also prove that, contrary to wired networking, max-min fairness yields all flows having the same rate, thus causing large inefficiencies. These findings offer theoretical explanations to the inefficiency and unfairness phenomena previously observed in the contexts of 802.11 and UWB networks. Finally, we show that proportional fairness achieves a good trade-off between efficiency and fairness and is a good candidate for a rate-based performance metric in wireless ad-hoc settings. Having shown that the proportional fairness is an appropriate optimization objective for our problem, in the second part of the thesis we consider a joint optimization of rates, transmission powers, medium access (scheduling) and routing, where the goal of the optimization is to achieve proportional fairness. We first analyze networks built on physical layers that have a rate which is a linear function of SNR at the receiver (such as UWB or low-gain CDMA systems). We find that the optimal solution is characterized by the following principles: (1)Whenever a node transmits, it has to transmit with the maximum power; otherwise it has to remain silent (0 - PMAX power control). (2) Whenever data is being sent over a link, it is optimal to have an exclusion region around the destination, in which all nodes remain silent during transmission, whereas nodes outside of this region can transmit in parallel, regardless of the interference they produce at the destination. (3) When a source transmits, it adapts its transmission rate according to the level of interference at the destination due to sources transmitting in parallel. (4) The optimal size of this exclusion region depends only on the transmission power of the source of the link, and not on the length of the link nor on positions of nodes in its vicinity. As for the routing, we restrict ourselves to a subset of routes where on each successive hop we decrease the distance toward the destination. We also show that (5) relaying along a minimum energy and loss route is better than using longer hops or sending directly, which is not obvious since we optimize rate and not power consumption. Finally (6), the design of the optimal MAC protocol is independent of the choice of the routing protocol. We present a theoretical proof of optimality of 0 - PMAX power control, and the remaining findings we show numerically on a large number of random network topologies. Next, we consider narrow-band networks, where rate function is a strictly concave function of SNR. There, previous findings do not always hold. We show that in some cases, the size of the exclusion region and the optimal routing depend on transmission powers, and that the optimal MAC design depends on the choice of routing. Nevertheless, as we show with the example of 802.11 networks, a significant improvement over the existing 802.11 MAC can be achieved even with simpler, suboptimal strategies. Although this result is shown by simulations on a simplified model, it still gives further directions on how to improve the performance of RTS/CTS based protocols.

EPFL2005