IMDEA Networks Institute Digital Repository
Not a member yet
1915 research outputs found
Sort by
A Cell-free Networking System with Visible Light
Get PDFLED luminaries are now deployed densely in indoor areas to provide uniform illumination. Visible Light Communication (VLC) can also benefit from this dense LED infrastructure. In this paper, we propose DenseVLC, a cell-free massive MIMO networking system enabled by densely distributed LEDs, that forms different beamspots to simultaneously serve multiple receivers. This is a cell-free system, as there is no notion of autonomous cells and transmitters cooperate to jointly serve the users.
Given a power budget for communication, DenseVLC assigns the power budget among the distributed LEDs to optimize the system throughput and user fairness.
We formulate an optimization problem to derive the optimal policy for the power allocation. Our insights from the optimal policies allow us to simplify DenseVLC's system design and propose a heuristic algorithm that can reduce the complexity by 99.96%. Besides, we propose a novel synchronization method using non-line-of-sight VLC to synchronize all the transmitters that will form a beamspot to serve the same receiver.
We implement DenseVLC with off-the-shelf devices, solve practical challenges in the system design, and evaluate it with extensive and realistic experiments in a system of 36 transmitters and 4 receivers in an area of 3mX3m.
Our results show that DenseVLC can improve the average system throughput by 45%, or improve the average power efficiency by 2.3 times, while maintaining the requirement for uniform illumination. Finally, we demonstrate that DenseVLC is robust against blockage.pu
Stability Under Adversarial Injection of Dependent Tasks
Get PDFIn this work, we consider a computational model of a distributed system formed by a set of servers in which jobs, that are continuously arriving, have to be executed. Every job is formed by a set of dependent tasks (i. e., each task may have to wait for others to be completed before it can be started), each of which has to be executed in one of the servers. The arrival and properties of jobs are assumed to be controlled by a bounded adversary, whose only restriction is that it cannot overload any server. This model is a non-trivial generalization of the Adversarial Queuing Theory model of Borodin et al. and, like that model, focuses on the stability of the system: whether the number of jobs pending to be completed is bounded at all times. We show multiple results of stability and instability for this adversarial model under different combinations of the scheduling policy used at the servers, the arrival rate, and the dependence between tasks in the jobs.TRUEpu
The Lockdown Effect: Implications of the COVID-19 Pandemic on Internet Traffic
Get PDFDue to the COVID-19 pandemic, many governments imposed lock-downs that forced hundreds of millions of citizens to stay at home. The implementation of confinement measures increased Internet traffic demands of residential users, in particular, for remote working, entertainment, commerce, and education, which, as a result, caused traffic shifts in the Internet core.
In this paper, using data from a diverse set of vantage points (one ISP, three IXPs, and one metropolitan educational network), we examine the effect of these lockdowns on traffic shifts. We find that the traffic volume increased by 15-20% almost within a week---while overall still modest, this constitutes a large increase within
this short time period. However, despite this surge, we observe that the Internet infrastructure is able to handle the new volume, as most traffic shifts occur outside of traditional peak hours. When looking directly at the traffic sources, it turns out that, while hypergiants still contribute a significant fraction of traffic, we see (1) a higher increase in traffic of non-hypergiants, and (2) traffic increases in applications that people use when at home, such as Web conferencing, VPN, and gaming. While many networks see increased traffic demands, in particular, those providing services to residential users, academic networks experience major overall decreases. Yet, in these networks, we can observe substantial increases when considering applications associated to remote working and lecturing.TRUEpu
Design And Development Of A Worldwide-Scale Measurement Methodology And Its Application In Network Measurements And Online Advertising Auditing
Get PDFOnline advertising has evolved into a key component of the Internet we know today. It is a very complex ecosystem that accomplishes to reach billions of users in a short period of time. It has global coverage, and it is able to target specific audiences based on demographic, geographic, and behavioral aspects. The capabilities offered by the online advertising ecosystem have opened a new era in research that has attracted the interest of the scientific community. This thesis leverages the nature of online advertising and builds a novel methodology capable of inserting JavaScript code into an ad that runs every time it is displayed on a user’s device. This methodology opens up new measurement opportunities. Specifically, this methodology is applied for two different purposes in this thesis: (1) Performing network measurements from the end-user perspective, and (2) Auditing the transparency of the online advertising ecosystem from the advertisers’ perspective. In the context of Internet measurements, this methodology is implemented in a solution referred to as AdTag. Its design - including technical, deployability, and economic factors – and its potential to analyze a wide range of aspects of Internet connectivity from the browser are discussed and evaluated. Several experiments are performed that prove the ability of AdTag to reach millions of nodes in a short period of time. Furthermore, the possibility of selecting the measurement nodes based on its geographical location is also demonstrated. In this thesis, we showcase the utility of AdTag to conduct network measurements in two specific use cases. First, we study the DNS infrastructure, one of the most critical Internet systems. Our analysis addresses issues such us grasping the real DNS infrastructure configured by the ISPs, and understanding the end-users DNS choices, whether they use private ISPs’ resolvers or establish third-party DNS resolvers, to improve security and web performance. Harnessing the scale offered by the online advertising ecosystem, two ad campaigns have been launched, triggering more than 3M DNS lookups, which allow the identification and study of more than 76k recursive DNS resolvers supporting more than 25k eyeball ASes in 178 countries. The data analysis provides new insights into the DNS infrastructure, such as user preferences towards third-parties. Our results indicate that 13% of users use third-party DNS providers (such as Google, OpenDNS, Level 3, and Cloudflare). Besides, this research detects different deployment decisions of many ISPs that provide both mobile and fixed access networks to separate the DNS infrastructure that serves each access technology type. The second considered use case consists of analyzing the browser market landscape with active measurements. We leverage AdTag to develop an active measurement platform to obtain the brand and the version of the device receiving the ad. We prove that the landscape picture obtained with our methodology is very similar to that offered by state-of-the-art techniques based on passive measurements. However, our solution presents some advantages over passive solutions: the ability to conduct geographically and demographically targeted measurements and its accessibility to a larger group of scientists and practitioners. The performance, accuracy, and capabilities of this methodology are analyzed through real experiments that, in total, produced more than 6M measurements. The lack of transparency in the online advertising ecosystem motivates the second part of this thesis. In particular, we have developed Q-Tag, a novel methodology that serves to audit reported quality metrics so that advertisers can obtain trustable information about the real performance of their advertising campaigns. The first version of Q-Tag was deployed in Google AdWords. The results reveal that AdWords seems to provide incomplete information to advertisers. In particular, they show that: (i) AdWords did not report 57% of the publishers where ad impressions from our campaigns were delivered, (ii) AdWords reports a large fraction of contextually significant impressions based on (undisclosed) criteria other than publisher’s theme, (iii) higher CPM investment does not lead to impressions being delivered to more popular publishers, (iv) AdWords does not offer default control of frequency cap (limit of impressions per user), (v) about 10% of ad impressions in two of the campaigns were delivered to IPs from Data Centers. The second version of Q-Tag was developed to measure the viewability metric. This standard metric serves to assess whether an ad impression was viewed or not by a user. Q-Tag has been deployed in production by a Demand Side Platform (DSP) to measure the viewability rate of the ad campaigns. Taking advantage of the infrastructure of this DSP, the performance of Q-Tag has been compared with a commercial solution. Both techniques report a similar overall viewability rate of 50% (i.e.,, 50% of the ad impressions meet the viewability standard and thus are considered viewed). However, Q-Tag is able to measure the viewability metric in 93% of the ads served by the DSP, unlike 74% of the ads measured by the commercial solution. In summary, the research conducted in this thesis showcases the potential of the proposed large-scale ad-based measurement. It offers a wider range of possibilities beyond those presented in this thesis. A methodology that can unravel different aspects of the Internet infrastructure and performance from the user perspective as well as provide an independent tool for advertisers to measure the quality of their advertising campaigns.Telematics EngineeringUniversidad Carlos III de Madrid, Spainpu
Optimal strategies for floating anchored information with partial infrastructure support
Get PDFFloating Content (FC) is a communication paradigm to locally share ephemeral content without direct support from infrastructure. It is based on constraining the opportunistic replication of content in a way that strikes a balance between minimizing resource usage and maximizing content availability among the intended recipients. However, existing approaches to management of FC schemes are unfit for realistic scenarios with non-uniform user distributions, resulting in heavy overdimensioning of resources allocated to FC. In this work, we propose a new version of FC, called Cellular Floating Content (CFC), which optimizes the use of bandwidth and memory by adapting the content replication and storage strategies to the spatial distribution of users, and to their mobility patterns. The main idea underlying our approach is to partition users into small “local communities”, and to optimally weight their contributions to the FC paradigm according to their specific mobility features, and to the resources required to achieve a target performance level. We characterize numerically the properties of the optimal strategies in a variety of mobility patterns and traffic conditions, showing the accuracy of our approach, and the significant savings it enables in the amount of resources necessary to run FC, which in a realistic setup can be as high as 27% with respect to traditional FC dimensioning strategies.pu
RL-Cache: Learning-Based Cache Admission for Content Delivery
Get PDFContent delivery networks (CDNs) distribute much of the Internet content by caching and serving the objects requested by users. A major goal of a CDN is to maximize the hit rates of its caches, thereby enabling faster content downloads to the users. Content caching involves two components: an admission algorithm to decide whether to cache an object and an eviction algorithm to determine which object to evict from the cache when it is full. In this paper, we focus on cache admission and propose a novel algorithm called RL-Cache that uses model-free reinforcement learning (RL) to decide whether or not to admit a requested object into the CDN’s cache. Unlike prior approaches that use a small set of criteria for decision making, RL-Cache weights a large set of features that include the object size, recency, and frequency of access. We develop a publicly available implementation of RL-Cache and perform an evaluation using production traces for the image, video, and web traffic classes from Akamai’s CDN. The evaluation shows that RL-Cache improves the hit rate in comparison with the state of the art and imposes only a modest resource overhead on the CDN servers. Further, RL-Cache is robust enough that it can be trained in one location and executed on request traces of the same or different traffic classes in other locations of the same geographic region. The paper also reports extensive analyses of the RL-Cache sensitivity to its features and hyperparameter values. The analyses validate the made design choices and reveal interesting insights into the RL-Cache behavior.pu
A Machine Learning approach to 5G Infrastructure Market optimization
Get PDFIt is now commonly agreed that future 5G Networks will build upon the network slicing concept. The ability to provide virtual, logically independent "slices" of the network will also have an impact on the models that will sustain the business ecosystem. Network slicing will open the door to new players: the infrastructure provider, which is the owner of the infrastructure, and the tenants, which may acquire a network slice from the infrastructure provider to deliver a specific service to their customers. In this new context, how to correctly handle resource allocation among tenants and how to maximize the monetization of the infrastructure become fundamental problems that need to be solved. In this paper, we address this issue by designing a network slice admission control algorithm that (i) autonomously learns the best acceptance policy while (ii) it ensures that the service guarantees provided to tenants are always satisfied. The contributions of this paper include: (i) an analytical model for the admissibility region of a network slicing-capable 5G Network, (ii) the analysis of the system (modeled as a Semi-Markov Decision Process) and the optimization of the infrastructure providers revenue, and (iii) the design of a machine learning algorithm that can be deployed in practical settings and achieves close to optimal performance.pu
A Machine Learning-based Framework for Optimizing the Operation of Future Networks
Get PDFThe fifth generation of mobile networks (5G) and beyond are not only sophisticated and difficult to manage, but must also satisfy a wide range of stringent performance requirements and adapt quickly to changes in traffic and network state. Advances in machine learning and parallel computing underpin new powerful tools that have the potential to tackle these complex challenges. In this paper, we develop a general machine
learning- based framework that leverages artificial intelligence to forecast future traffic demands and characterize traffic features. This enables to exploit such traffic insights to improve the performance of critical network control mech- anisms, such as load balancing, routing, and scheduling. In contrast to prior works that design problem-specific machine learning algorithms, our generic approach can be applied to different network functions, allowing to re-use existing control mechanisms with minimal modifications. We explain how our framework can orchestrate ML to improve two different network mechanisms. Further, we undertake validation by implementing one of these, i.e., mobile backhaul routing, using data collected by a major European operator and demonstrating a 3x reduction of the packet delay, compared to traditional approaches.TRUEpu
Time-based indoor positioning and context information using commodity WiFi chipsets
Get PDFIn the last years indoor localization and applications that use positioning information have attracted a lot of attention from the research community as well as the industry. The achieved accuracy of an indoor localization system is the key to enable certain applications, such as navigation. In a scenario where the system runs exclusively on commodity hardware such as smartphones and even without installing any mobile app in the mobile device, location information may be exploited not only for navigation, but also for the benefit of the network itself or for the investigation of physical behaviors.
In indoor areas, communication technologies such as Wireless Fidelity (WiFi) are gaining popularity due to the ever increasing availability and deployment of APs that can be used simultaneously as an infrastructure for networking and positioning. Despite being the most used technology for tracking devices in indoor environments, the research community has focused more intensively on approaches such as the Received Signal Strength Indicator (RSSI) or combining WiFi signals with inertial sensors as found in smartphones. These approaches are error prone, dependent on frequent calibrations, or they require specialized hardware or the user must be interested to run a specific application designed to run on the smartphone. Using only commodity WiFi chipsets, this thesis investigates location systems that are based on Time-of-Flight (ToF) echo technique, and do not require any calibration and user intervention, and it explores ranging and location data to improve network management and infer user behaviour. ToF technique has been modestly exploited for indoor localization, whereas ToF is successfully used in Global Positioning System (GPS) in outdoor environments. The reason is that WiFi ToF measurements, mainly extracted from commodity chipsets, suffer from extensive device-related noise which makes it challenging to differentiate between direct path from non-direct path signal components when estimating the ranges. Existing multipath mitigation techniques tend to fail at identifying the direct path when the devicerelated Gaussian noise is in the same order of magnitude, or larger than the multipath noise. In order to address this challenge, we first propose in this thesis a new method for filtering ranging measurements, extracted from a commodity WiFi chipset, that is better suited for the inherent large noise as found in WiFi radios. Our proposed filter combines
statistical learning and robust statistics and it does not require specialized hardware,
the intervention of the user, or cumbersome on-site manual calibration. This makes the method we propose as the first contribution of the present work particularly suitable for indoor localization in large-scale deployments using existing legacy WiFi infrastructures. We build and investigate a multi WiFi ToF-based APs localization system, where these filtered timing signals are used as ranging inputs of the multi-lateration problem for positioning. We then deploy and evaluate our system for indoor mobile tracking scenarios in many multipath-rich environments, across multiple testbeds which cover different surfaces and further test it in Microsoft indoor localization competitions, demonstrating that, despite these challenges, our technique can achieve distance accuracy comparable to
other approaches proposed by the start of the art but does not share their aforementioned shortcomings.
The deployment of a multi-APs positioning system is targeted to specific areas, such as large offices and shopping malls. In order to bring indoor positioning also to homes and small businesses which typically have a single AP, the next step we envision towards a hybrid single-AP positioning system is a deep inspection and interaction of the Fine Time Measurements (FTM), a type of ToF echo technique standardized recently, and Channel State Information (CSI) for ranging and Angle-Of-Arrival (AOA) for angle estimates, respectively. We exploit Physical Layer (PHY) information to detect the number of paths and their directions and we use this information to derive a new method for filtering ranging measurements obtained with the FTM protocol. We achieve sub-meter distance estimation accuracy eliminating the adverse effect of multipath in FTM using calibrated inputs from CSI. We then evaluate the system in multipath-rich environments, demonstrating the capability of the combination of AOA estimation and the proposed FTM refinement approach to achieve reasonable positioning accuracy in areas comparable to typical flat sizes just using a commodity smartphone as target device.
All the contributions described so far and all the extracted location data may be exploited also in the network core to better allocate network resources based on the expected link performance. Thus, we take advantage of positioning data and more in general context information, to enable reliable mobile communications via advanced resource management policies and adaptive traffic engineering strategies. Of particular interest for this thesis is to investigate usage of positioning data in industrial environments, where the presence of metallic objects challenge the reliability of wireless communication. With the advent of the fourth Industrial revolution (Industry 4.0) such harsh environments
have attracted high attention and, for this reason, of particular interest for this thesis is also the deployment of our multi-APs system in such industrial environment. The latter, due to blockage and strong reflections, is notorious for being adverse to wireless communication, impacting on the signal quality. We propose to exploit the knowledge of location to derive context information to dynamically allocate wireless resources in time and space to target devices. We exploit the spatial geometry of the APs and a statistical model that maps the user position’s spatial distribution to an angle error distribution to derive a hypothesis test to declare if the link is in Line-Of-Sight (LOS) or Non-Line-Of-Sight (NLOS). In order to avoid changes to the client side and operate with a single interface radio, we use the same wireless network both for positioning and scheduling. We experimentally show that context information applied to wireless resources protocol help increasing the network throughput in the aforementioned industrial-like scenario. Finally, taking advantage of ranging information extracted from FTM, we propose to estimate the device movement without any access to physical inertial sensors in the mobile. The idea is to infer the movements of the mobile through radio measurements, a concept we call ”virtual inertial sensors”. We propose a method for estimating the user walking speed and a novel method for the rotation of a mobile device. We evaluate and demonstrate the proposed approaches with experiments, and we compare the rotation method with a possible solution that explores CSI measurements. While FTM works with only one single antenna, it achieves better performance than a CSI based estimator that exploits four antennas and multiple sub-carriers at the AP, but are yet limited by the typical one single WiFi antenna at the smartphone side. This new concept of virtual inertial sensors can be leveraged by location systems and sensing mechanisms to improve localization accuracy, infer user behavior, and design better and more secure communication.
In conclusion, in this thesis we experimentally demonstrate that ToF data, extracted from commodity WiFi chipsets, can be used for tracking devices and provide context information in indoor environments without the need for environmental calibration and installations of mobile apps, but only with standard-compliant measurements performed from the WiFi infrastructure.Telematics EngineeringUniversidad Carlos III de Madrid, Spainpu