Section: New Results

Resource and Traffic Management

Traffic offloading; infrastructure deployment; opportunistic routing; traffic modeling; intermittently connected networks.

On the Interaction between Content Caching and Routing

Participants : Kolar Purushothama Naveen, Laurent Massoulié, Emmanuel Baccelli, Aline Carneiro Viana, Don Towsley.

Nowadays Internet users are mobile over 60% of their time online, and mobile data traffic is expected to increase by more than 60% annually to reach $15.9$ exabytes per month by 2018. This evolution will likely incur durably congested wireless access at the edge despite progress in radio technologies. To alleviate congestion at the Internet edge, one promising approach is to target denser deployments of wireless access points. As a result, mobile users are potentially within radio reach of several access points (AP) from which content may be directly downloaded. In this context, distinct AP's can have very different bandwidth and memory capacities. Such differences raise the following question: When requests can be sent to several such access points, how to optimize performance through both load balancing and content replication?

In this work, we introduce formal optimization models to address this question, where bandwidth availability is represented via a cost function, and content availability is represented either by a cost function or a sharp constraint. For both formulations we propose dynamic caching and request assignment algorithms. Crucially our request assignment scheme is based on a server price signal jointly reflecting content and bandwidth availability. Using mean field approximation and Lyapunov functions techniques, we prove that our algorithms are optimal and stable in a limiting fluid regime with large arrival rates and content chunking. Through simulations we exhibit the efficacy of our request assignment strategy in comparison to the common practices of assigning requests purely based on either bandwidth or content availability. Finally, using the popular LRU (Least Recently Used) strategy instead for cache replacements, we again demonstrate the superior performance of our request assignment strategies. This work was published in the ACM SIGCOMM'15 workshop on All Things Cellular.

From Routine to Network Deployment for Data Offloading in Metropolitan Areas

Participants : Eduardo Mucceli, Aline Carneiro Viana.

Smartphone sales are booming, nearly half billion were sold in 2011; more smartphones, more mobile data traffic, and Currently, 3G cellular networks in metropolitan areas are struggling to attend the recent boost up of mobile data consumption. Carefully deploying WiFi hotspots allow to maximize WiFi offloading and can both be cheaper than upgrade the cellular network structure and concede substantial improvement in the network capacity. In this context, in this work, we first propose a new way to map into a graph the people behavior (i.e., mobility context) in an urban scenario. Our proposed behavior-to-graph solution is simple, take into consideration the restrictions imposed by transportation modes to traffic demand, the space-time interaction between people and urban locations, and finally, is powerful to be used as input to any popular area identification problem (key points for an efficient network planning). Secondly, we propose a metric to identify locations more capable of providing coverage for people and consequently, more suitable for receiving hotspots. Deploying a small percentage of hotspots ranked by the herein proposed metric provides high percentages of coverage time for people moving around in the city. Using a real-life metropolitan trace, we show our routine-based strategy guarantees higher offload ratio than current approaches in the literature while using a realistic traffic model. This work, including new characterization results of the used trace and new analysis of space-traffic correlation, is under submission in a trasaction.

Mobile Data Traffic Modeling: Revealing Hidden Facets

Participants : Eduardo Mucceli, Aline Carneiro Viana, Kolar Purushothama Naveen, Carlos Sarraute.

Smartphone devices provide today the best means of gathering users information about content consumption behavior on a large scale. In this context, the literature is rich in work studying and modeling users mobility, but little is publicly known about users content consumption patterns. The understanding of users' mobile data traffic demands is of fundamental importance when looking for solutions to manage the recent boost up of mobile data usage and to improve the quality of communication service provided. Hence, the definition of a usage pattern can allow telecommunication operators to better foreseen future demanded traffic and consequently, to better (1) deploy data offloading hotspots or (2) timely plan network resources allocation and then, set subscription plans.

Using a large-scale dataset collected from a major 3G network in a big metropolitan area, in this work, we present the first detailed measurement-driven modeling of mobile data traffic usage of smartphone subscribers. Our main outcome is a synthetic measurement-based mobile data traffic generator, capable of imitating traffic-related activity patterns of different categories of subscribers and time periods of a routinary normal day in their lives. For this, we first characterize individual subscribers routinary behaviour, followed by the detailed investigation of subscribers' usage pattern (i.e., "when" and "how much" traffic is generated). Broadly, our observations bring important insights into network resource usage. We then classify the subscribers into six distinct profiles according to their usage pattern and model these profiles according to two different journey periods: peak and non-peak hours. We show that the synthetic trace generated by our data traffic model consistently imitates different subscriber profiles in two journey periods, when compared to the original dataset. We discuss relevant issues in traffic demands and describe implications in network planning and privacy. This work, including a new characterization results of the used trace, including analysis correlating age and gender to traffic demands, as well as new profiling results, is under submission in a transaction.

Data Delivery in Opportunistic and Intermittently Connected Networks

Participants : Ana Cristina Vendramin, Anelise Munaretto, Myriam Delgado, Aline Carneiro Viana, Mauro Fonseca.

The pervasiveness of computing devices and the emergence of new applications and cloud services are factors emphasizing the increasing need for adaptive networking solutions. In most cases, this adaptation requires the design of interdisciplinary approaches as those inspired by nature, social structures, games, and control systems. The approach presented in this work brings together solutions from different, yet complementary domains, i.e., networking, artificial intelligence, and complex networks, and is aimed at addressing the problem of efficient data delivery in intermittently connected networks.

As mobile devices become increasingly powerful in terms of communication capabilities, the appearance of opportunistic and intermittently connected networks referred to as Delay Tolerant Networks (DTNs) is becoming a reality. In such networks, contacts occur opportunistically in corporate environments such as conferences sites, urban areas, or university campuses. Understanding node mobility is of fundamental importance in DTNs when designing new communication protocols that consider opportunistic encounters among nodes. This work proposes the Cultural Greedy Ant (CGrAnt) protocol to solve the problem of data delivery in opportunistic and intermittently connected networks. CGrAnt is a hybrid Swarm Intelligence-based forwarding protocol designed to address the dynamic and complex environment of DTNs. CGrAnt is based on: (1) Cultural Algorithms (CA) and Ant Colony Optimization (ACO) and (2) operational metrics that characterize the opportunistic social connectivity between wireless users. The most promising message forwarders are selected via a greedy transition rule based on local and global information captured from the DTN environment. Using simulations, we first analyze the influence of the ACO operators and CA knowledge on the CGrAnt performance. We then compare the performance of CGrAnt with the PROPHET and Epidemic protocols (two well known related protocols in the literature) under varying networking parameters. The results show that CGrAnt achieves the highest delivery ratio (gains of 99.12% compared with PROPHET and 40.21% compared with Epidemic) and the lowest message replication (63.60% lower than PROPHET and 60.84% lower than Epidemic). This work is under submission to an international journal.

Designing Adaptive Replication Schemes in Distributed Content Delivery Networks

Participants : Mathieu Leconte, Marc Lelarge, Laurent Massoulié.

In a paper published at the ITC'15 conference, we address the problem of content replication in large distributed content delivery networks, composed of a data center assisted by many small servers with limited capabilities and located at the edge of the network. We aim at optimizing the placement of contents on the servers to offload the data center as much as possible. We model the sub-system constituted by the small servers as a loss network, each loss corresponding to a request to the data center. Based on large system / storage behavior, we obtain an asymptotic formula for the optimal replication of contents and propose adaptive schemes to attain it by reacting to losses, as well as faster algorithms which can react before losses occur. We show through simulations that our adaptive schemes outperform significantly standard replication strategies both in terms of loss rates and adaptation speed.

Vehicular Network under a Social Perception

Participants : Felipe D. Cunha, Aline Carneiro Viana, Raquel A. F. Mini, Antonio A.f. Loureiro.

Vehicular Mobility is strongly influenced by the speed limits, destinations, traffic conditions, period of the day, and direction of the public roads. At the same time, the driver's behavior produces great influences in vehicular mobility. People tend to go to the same places, at the same day period, through the same trajectories, which le ad them to the appearance of driver's daily routines. These routines lead us to the study of mobility in VANETs under a social perspective and to investigate how effective is to explore social interactions in this kind of network. In this work, we thus characterize and evaluate social properties of a realistic vehicular trace found in literature. Our aim is to study the vehicles' mobility in accordance to social behaviors. Social metrics are computed and the obtained results are compared to random graphs. With our analysis, we could verify the existence of regularity and common interests among the drivers in vehicular networks.

After having identified routine in vehicles mobility patterns and their correlation with the period of the day, we then leverage the identified social aspects to design a Socially Inspired Broadcast Data Dissemination for VANETs. We claim that protocols and applications designed for Vehicular Ad Hoc Networks need to adapt to vehicles routines in order to provide better services. With this issue in mind, we designed a data dissemination solution for these networks that considers the daily road traffic variation of large cities and the relationship among vehicles. The focus of our approach is to select the best vehicles to rebroadcast data messages according to social metrics, in particular, the clustering coefficient and the node degree. Moreover, our solution is designed in such a way that it is completely independent of the perceived road traffic density. Simulation results show that, when compared to related protocols, our proposal provides better delivery guarantees, reduces the network overhead and possesses an acceptable delay.

Design and Analysis of an Efficient Friend-to-Friend Content Dissemination System

Participants : Kanchana Thilakarathna, Aline Carneiro Viana, Aruna Seneviratne, Henrik Petander.

In this work, we focus on dissemination of content for delay tolerant applications/services, (i.e. content sharing, advertisement propagation, etc.) where users are geographically clustered into communities. Due to emerging security and privacy concerns, majority of users are becoming more reluctant to interact with strangers and are only willing to share information/content with the users who are previously identified as friends. As a result, despite its promise, opportunistic communications systems have not been widely adopted. In addition, in this environment, opportunistic communication will not be effective due to the lack of known friends within the communication range. We thus propose a novel architecture which combines the advantages of distributed decentralized storage and opportunistic communications. The proposed system addresses the trust and privacy concerns of opportunistic communications systems, and enables the provision of efficient distributed mobile social networking services. We exploit the fact that users will trust their friends, and the friends will help in disseminating content by temporarily storing and forwarding content. This can be done by replicating content on friends' devices who are likely to consume that content and provide the content to other friends when the device has access to low cost networks. The fundamental challenge then is to minimize the number of replicas, to ensure high and timely availability. We provide a formal definition of this content replication problem, and show that it is NP hard. Then, we propose a community based greedy heuristic algorithm with novel dynamic centrality metrics that replicates the content on a minimum number of friends' devices, and maximizes the availability of content. Using both real world and synthetic traces, we validate effectiveness of the proposed scheme. In addition, we demonstrate the practicality of the the proposed system, through an implementation on Android smartphones. This work is under submission in an international transaction.

Telling Apart Social and Random Relationships in Dynamic Networks

Participants : Pedro Olmo Vaz de Melo, Aline Carneiro Viana, Marco Fiore, Katia Jaffrès-Runser, Frédéric Le Mouël, Antonio A. F. Loureiro, Lavanya Addepalli, Guangshuo Chen.

Recent studies have analyzed data generated from mobile individuals in urban regions, such as cab drivers or students in large campuses. Particular attention has been paid to the dynamics of user movement, whose real-world complexity cannot be fully captured through synthetic models. Indeed, understanding user mobility is of fundamental importance when designing new communication protocols that exploit opportunistic encounters among users. In this case, the problem mainly lies in correctly forecasting future contacts. To that end, the regularity of daily activities comes in handy, as it enforces periodic (and thus predictable) space-time patterns in human mobility. Although human behavior is characterized by an elevated rate of regularity, random events are always possible in the routines of individuals. Those are hardly predictable situations that deviate from the regular pattern and are unlikely to repeat in the future.

We argue that the ability to accurately spot random and social relationships in dynamic networks is essential to network applications that rely on a precise description of human routines, such as recommendation systems, forwarding strategies and opportunistic dissemination protocols. We thus propose a strategy to analyze users' interactions in mobile networks where users act according to their interests and activity dynamics. Our strategy, named Random rElationship ClASsifier sTrategy (RECAST), allows classifying users' wireless interactions, separating random interactions from different kinds of social ties. To that end, RECAST observes how the real system differs from an equivalent one where entities' decisions are completely random. We evaluate the effectiveness of the RECAST classification on five real-world user contact datasets collected in diverse networking contexts. Our analysis unveils significant differences among the dynamics of users' wireless interactions in the datasets, which we leverage to unveil the impact of social ties on opportunistic routing. We show that, for such specific purpose, the relationships inferred by classifier are more relevant than, e.g., self-declared friendships on Facebook. This work was published in 2015 at the Performance Evaluation Elsevier Journal [9] .