Decentralized compute markets claim to be an alternative to the cloud, yet the systems community lacks hard evidence on whether they can run modern workloads end to end. We put this claim to the test by deploying a Retrieval-Augmented Generation (RAG) pipeline on the Akash and Golem networks, two popular decentralized computing markets. Our analysis reveals a striking paradox: supply is abundant, with over 10,000 CPU cores and 50 TiB of memory available, yet utilization remains minimal (about 0.3% on Akash). When mapped carefully, several RAG pipeline stages execute correctly and up to 3 × cheaper than AWS; others degrade under node churn and network bottlenecks. We also observe that decentralization in practice is already hybrid, with centralized gateways and offchain services masking blockchain complexity. Taken together, our findings suggest that decentralized markets are neither hype nor drop-in cloud replacements, but an underutilized substrate whose viability depends on better scheduling, reliability mechanisms, and trust primitives.

Immersive telepresence has the potential to transform real-time communication through highly interactive and engaging experiences. Despite recent advances in reducing communication and computation costs, existing systems largely overlook packet loss, which can severely degrade the quality of experience (QoE). Recovering lost immersive content is considerably more challenging than in 2D video due to the complexity of dense 3D representations. Recovery must be both accurate and timely while minimizing the communication and computation overhead it incurs. To address these challenges, we present LITE, the first loss-resilient immersive telepresence system. LITE incorporates three key design principles: (1) leveraging semantic communication to transmit compact motion and audio semantics, which can be reconstructed into the remote user's immersive representation and voice, enabling fast semantic-level recovery and remaining robust to congestion-control-induced rate reductions under loss; (2) fusing audio and motion semantics via a lightweight multimodal model to achieve accurate, real-time recovery of motion semantics; and (3) encoding audio semantics from multiple past frames into succinct neural redundancy to enable robust recovery. We prototype LITE using a well-known parametric facial motion representation and extensively evaluate its performance across diverse networks. Our results demonstrate that LITE improves QoE by up to 109% compared with existing schemes, while sustaining real-time streaming at 30 frames per second and preserving high visual fidelity (structural similarity index measure above 0.9, where 1 indicates perfect similarity).

Over a quarter of the global population remains without access to the Internet in 2026, primarily due to cost and limited infrastructure in developing regions. This gap is becoming more consequential as access to information increasingly depends on AI systems that presume reliable, high-bandwidth connectivity. In response, we present WebFM, a low-cost, scalable Web and AI access solution that builds on widely deployed infrastructure in developing regions: FM radio for downlink broadcasting and SMS for personalized uplink. Prior work on web-over-FM is limited to simulations, short-range experiments, and lossless links, leaving the effects of geographically induced signal variability largely unexplored. WebFM addresses this gap through several system-level innovations to effectively transmit Web content and enable LLM interactions over sound over FM radio, in a reliable and compressed form. For example, we introduce a loss-resilient file format for encoding webpages and modify Android to leverage built-in FM tuners, allowing deployment across a wide range of devices without requiring root access. We enable ChatGPT interactions by allowing users to submit prompts over SMS and receive model responses via FM broadcast. Finally, we preemptively transmit popular pages and a subset of their internal pages while the server is idle, improving user experience during peak hours. We deployed WebFM at an FM radio station in Cameroon for six weeks with 30 participants. Our evaluation shows a sustained downlink throughput of 10 kbps, less than 20% loss for most transmissions with signal strength above -90 dBm, and strong user engagement across both web browsing and ChatGPT interactions.

English is the predominant language on the web, powering nearly half of the world's top ten million websites. Support for multilingual content is nevertheless growing, with many websites increasingly combining English with regional or native languages in both visible content and hidden metadata. This multilingualism introduces significant barriers for users with visual impairments, as assistive technologies like screen readers frequently lack robust support for non-Latin scripts and misrender or mispronounce non-English text, compounding accessibility challenges across diverse linguistic contexts. Yet, large-scale studies of this issue have been limited by the lack of comprehensive datasets on multilingual web content. To address this gap, we introduce LangCrUX, the first large-scale dataset of 120,000 popular websites across 12 languages that primarily use non-Latin scripts. Leveraging this dataset, we conduct a systematic analysis of multilingual web accessibility and uncover widespread neglect of accessibility hints. We find that these hints often fail to reflect the language diversity of visible content, reducing the effectiveness of screen readers and limiting web accessibility. We finally propose Kizuki, a language-aware automated accessibility testing extension to account for the limited utility of language-inconsistent accessibility hints.

The rise of generative artificial intelligence (GenAI), powered by large language models, has led to the emergence of real-time, voice-based conversational applications that enable dynamic, multi-modal interactions for everyday tasks such as checking the weather or planning a trip. These human-to-GenAI calling applications blend speech processing, generative intelligence, and real-time communication, presenting new challenges in latency optimization, network infrastructure design, and resilience under load. Despite their growing popularity, little is known about the operational characteristics and performance of these applications. This paper conducts an empirical measurement of six human-to-GenAI calling applications from Google, Meta, Microsoft, and OpenAI, focusing on their input/output modalities, network behavior, latency metrics, and robustness. Our findings reveal key design choices and performance bottlenecks in these emerging applications. For example, the conversational latency often reaches several seconds, far exceeding the typical sub-second delays of human-to-human voice communication and potentially impairing interactivity. Moreover, voice-based GenAI traffic is inherently asymmetric: the uplink, carrying real-time human speech, benefits from streaming-based transmission, while the typically large downlink GenAI responses are better served through batch-based delivery.

Recently, various Mobile Network Aggregators (MNAs) have emerged, leveraging the coverage of a few base operators to offer global connectivity. These MNAs reap the benefits of network softwarization and virtualization, such as eSIM technology or control/data-plane separation. This paper explores a new type of MNA — a thick MNA — that uses multiple base operators from different regions to provide eSIM profiles and employs public internet gateways outside the base operators' home countries. Specifically, we analyze Airalo, a thick MNA operating in 219 countries. Unlike other MNAs that our community has scrutinized, we show that Airalo often decouples the geographical location of the public internet gateway from the native country of the base operator via IPX Hub Breakout (IHBO). To map Airalo's underlying infrastructure, we ran web-based measurements that 14 volunteers performed while traveling and using an Airalo eSIM on their personal devices. We further dive into Airalo's performance by running device-based measurements (speedtest, traceroute, etc.) in 10 countries. We find that IHBO, despite offering some latency benefits over traditional Home-Routed Roaming, still results in inflated latency and degraded bandwidth compared to non-roaming traffic, introducing new challenges for user experience, content access, and privacy.

The growing demand for reliable Internet access during air travel has made in-flight connectivity (IFC) a critical service for commercial airlines. Traditional IFC systems rely on geostationary (GEO) satellites, but their high latency and limited bandwidth hinder user experience. Emerging Low Earth Orbit (LEO) constellations, such as Starlink, promise better network performance. This paper presents the first empirical comparison of IFC performance across GEO and LEO networks, using data from 25 flights operated by 7 airlines. Measurements collected via instrumented Android devices cover key metrics including latency, throughput, and CDN responsiveness. We observe that while GEO systems rely on static and often distant Internet gateways, Starlink clients connect to different gateways that align more closely with flight trajectories, enhancing end-to-end performance through shorter satellite paths. However, DNS-based content filtering on Starlink-equipped flights often affects user geolocation, introducing unnecessary terrestrial delays. Finally, our case study of TCP performance from 2 Starlink flights shows that the BBR congestion control algorithm delivers up to 35× higher throughput than Cubic and Vegas, but with significantly higher retransmissions due to aggressive bandwidth probing.

With the rising demand for cost-effective and privacy-preserving deep learning and visualization services, service providers are increasingly turning to in-browser solutions. General-purpose computations, leveraging the graphics processing unit (GPU), are foundational to executing algorithms that power these services. Web graphics library (WebGL) is a widely adopted GPU-access application programming interface (API) designed for multidimensional rendering in browsers, while WebGPU is a newer API developed with compute-specific capabilities. Although WebGPU is a promising standard, its performance has not been systematically evaluated for general-purpose computation. This paper investigates WebGPU and WebGL for accelerating client-side computation in web browsers. By benchmarking key computational GPU kernels for 16 PolyBench and 2 CHStone functions, we measure the performance of WebGPU and WebGL across varying input sizes and algorithmic complexities. Our results show that: 1) both WebGPU and WebGL exhibit poorer performance than central processing unit (CPU)-based execution for small input data due to setup and CPU-GPU synchronization overheads, but they outperform CPU execution as the input data size increases; 2) WebGL performs better than WebGPU for small inputs, except for CPU-driven loop functions, due to its lower initial setup overhead; and 3) WebGPU outperforms WebGL for large inputs through optimized GPU thread utilization and achieves better performance for loop-driven algorithms across all input sizes by minimizing CPU-GPU data exchange. Overall, our results indicate that WebGPU is a competitive option for enhancing the execution performance of large-scale web applications.

While internet usage is slowly catching up globally, it is still unclear how the web experience differs in developing and developed countries. On the one hand, the web has a notoriously large inertia, with many webpages still relying on unencrypted HTTP, deprecated web features, or old and buggy libraries. On the other hand, developing countries are expected to leapfrog and directly adopt the newest technologies by learning from the prior mistakes of more developed countries. Anecdotal evidence suggests that webpages in developing and developed regions differ significantly. In this work, we test this hypothesis by measuring differences in web development practices across the two groups of countries, using multiple dimensions: webpages' size, complexity, security, privacy, quality, technology adoption, and accessibility. Concretely, we collect the largest dataset to date that compares web development practices across developed and developing regions -- 200,000 webpages across 20 countries -- which we aim to open source along with this publication. Our findings reveal that webpages in developing regions are generally smaller and simpler, utilizing fewer requests - an adaptation that improves the performance over slower network conditions common in these areas. However, these sites are less optimized in other critical aspects: they frequently employ inefficient image formats, include unnecessary JavaScript or CSS code, or lack responsive image designs. Notably, our security assessment shows developing regions lagging in HTTPS adoption and vulnerability mitigation, possibly due to lower awareness of best practices.

The web experience in developing regions remains subpar, primarily due to the growing complexity of modern webpages and insufficient optimization by content providers. Users in these regions typically rely on low-end devices and limited bandwidth, which results in a poor user experience as they download and parse webpages bloated with excessive third-party CSS and JavaScript (JS). To address these challenges, we introduce the Mobile Application Markup Language (MAML), a flat layout-based web specification language that reduces computational and data transmission demands, while replacing the excessive bloat from JS with a new scripting language centered on essential (and popular) web functionalities. Last but not least, MAML is backward compatible as it can be transpiled to minimal HTML/JavaScript/CSS and thus work with legacy browsers. We benchmark MAML in terms of page load times and sizes, using a translator which can automatically port any webpage to MAML. When compared to the popular Google AMP, across 100 testing webpages, MAML offers webpage speedups by tens of seconds under challenging network conditions thanks to its significant size reductions. Next, we run a competition involving 25 university students porting 50 of the above webpages to MAML using a web-based editor we developed. This experiment verifies that, with little developer effort, MAML is quite effective in maintaining the visual and functional correctness of the originating webpages.

Due to the widespread adoption of "work-from-home" policies, videoconferencing applications (e.g., Zoom) have become indispensable for remote communication. However, they often lack immersiveness, leading to "Zoom fatigue" and degrading communication efficiency. The recent debut of Apple Vision Pro, a mobile headset that supports "spatial personas", offers an immersive telepresence experience. In this paper, we conduct a first-of-its-kind in-depth and empirical study to analyze the performance of immersive telepresence with FaceTime, Webex, Teams, and Zoom on Vision Pro. We find that only FaceTime provides a truly immersive experience with spatial personas, whereas others still operate 2D personas. Our measurements reveal that (1) FaceTime delivers semantic data to optimize bandwidth consumption, which is even lower than that of 2D personas for other applications, and (2) it employs visibility-aware optimizations to reduce rendering overhead. However, the scalability of FaceTime remains limited, with a simple server-allocation strategy that potentially leads to high network delay for users.

This study examines Airalo, a "thick" Mobile Network Aggregator. We analyze its unique use of multiple base operators for eSIM provisioning and its decoupling of internet gateways from operators' home countries. Our methodology combines measurements from volunteers in 24 countries using a mix of personal and instrumented Android devices, providing a comprehensive analysis of Airalo's operational model and impact on global mobile connectivity.

Immersive telepresence has the potential to revolutionize remote communication by offering a highly interactive and engaging user experience. However, state-of-the-art exchanges large volumes of 3D content to achieve satisfactory visual quality, resulting in substantial Internet bandwidth consumption. To tackle this challenge, we introduce MagicStream, a first-of-its-kind semantic-driven immersive telepresence system that effectively extracts and delivers compact semantic details of captured 3D representation of users, instead of traditional bit-by-bit communication of raw content. To minimize bandwidth consumption while maintaining low end-to-end latency and high visual quality, MagicStream incorporates the following key innovations: (1) efficient extraction of user's skin/cloth color and motion semantics based on lighting characteristics and body keypoints, respectively; (2) novel, real-time human body reconstruction from motion semantics; and (3) on-the-fly neural rendering of users' immersive representation with color semantics. We implement a prototype of MagicStream and extensively evaluate its performance through both controlled experiments and user trials. Our results show that, compared to existing schemes, MagicStream can drastically reduce Internet bandwidth usage by up to 1195X while maintaining good visual quality.

In the new industrial Internet, a wide variety of industrial applications are expected to rely on high-performance data communication between a multitude of sensors and actuators that are deployed on a large scale. Publish-subscribe-based communication model is well-suited to handle such large-scale data gathering and dissemination among data sources and sinks. To support publish-subscribe-based data delivery, the newly standardized Segmented Routing over IPv6 (SRv6) can provide non-disruptive network programming primitives for building and maintaining network-efficient, shareable data distribution trees within the network. We study optimal algorithms for setting up different types of multicasting in the SRv6-capable network. In particular, we show, both theoretically and experimentally, that splitting multicast streams into multiple sub-streams, as well as using end-to-end application-layer coding without any network participation can provide significant benefits in terms of multicast throughput compared to traditional single stream multicasting.

Location tags are designed to track personal belongings. Nevertheless, there has been anecdotal evidence that location tags are also misused to stalk people. Tracking is achieved locally, e.g., via Bluetooth with a paired phone, and remotely, by piggybacking on location-reporting devices which come into proximity of a tag. This paper studies the performance of the two most popular location tags (Apple's AirTag and Samsung's SmartTag) through controlled experiments - with a known large distribution of location-reporting devices - as well as in-the-wild experiments - with no control on the number and kind of reporting devices encountered, thus emulating real-life use-cases. We find that both tags achieve similar performance, e.g., they are located 55% of the times in about 10 minutes within a 100~m radius. It follows that real time stalking to a precise location via location tags is impractical, even when both tags are concurrently deployed which achieves comparable accuracy in half the time. Nevertheless, half of a victim's exact movements can be backtracked accurately (10m error) with just a one-hour delay, which is still perilous information in the possession of a stalker.

The Domain Name System (DNS) is a key component of Internet-based communication and its privacy has been neglected for years. Recently, DNS over HTTPS has improved the situation by fixing the issue of in-path middleboxes. Further progress has been made with proxy-based solutions such as Oblivious DoH, which separate a user's identity from their DNS queries. However, these solutions rely on non-collusion between DNS resolvers and proxy networks. This paper instead proposes PDNS, a new DNS extension that uses Private Information Retrieval to allow DNS resolvers to operate on blind queries, thereby eliminating any privacy leaks.

Free and open-source media centers are experiencing a boom in popularity for the convenience they offer users seeking to remotely consume digital content. Kodi is today’s most popular home media center, with millions of users worldwide. Kodi’s popularity derives from its ability to centralize the sheer amount of media content available on the Web, both free and copyrighted . Researchers have been hinting at potential security concerns around Kodi, due to add-ons injecting unwanted content as well as user settings linked with security holes. Motivated by these observations, this article conducts the first comprehensive analysis of the Kodi ecosystem: 15,000 Kodi users from 104 countries, 11,000 unique add-ons, and data collected over 9 months. Our work makes three important contributions. Our first contribution is that we build “crawling” software ( de-Kodi ) which can automatically install a Kodi add-on, explore its menu, and locate (video) content. This is challenging for two main reasons. First, Kodi largely relies on visual information and user input which intrinsically complicates automation. Second, the potential sheer size of this ecosystem (i.e., the number of available add-ons) requires a highly scalable crawling solution. Our second contribution is that we develop a solution to discover Kodi add-ons. Our solution combines Web crawling of popular websites where Kodi add-ons are published (LazyKodi and GitHub) and SafeKodi , a Kodi add-on we have developed which leverages the help of Kodi users to learn which add-ons are used in the wild and, in return, offers information about how safe these add-ons are, e.g., do they track user activity or contact sketchy URLs/IP addresses. Our third contribution is a classifier to passively detect Kodi traffic and add-on usage in the wild. Our analysis of the Kodi ecosystem reveals the following findings. We find that most installed add-ons are unofficial but safe to use. Still, 78% of the users have installed at least one unsafe add-on, and even worse, such add-ons are among the most popular. In response to the information offered by SafeKodi, one-third of the users reacted by disabling some of their add-ons. However, the majority of users ignored our warnings for several months attracted by the content such unsafe add-ons have to offer. Last but not least, we show that Kodi’s auto-update, a feature active for 97.6% of SafeKodi users, makes Kodi users easily identifiable by their ISPs. While passively identifying which Kodi add-on is in use is, as expected, much harder, we also find that many unofficial add-ons do not use HTTPS yet, making their passive detection straightforward. 1

Home IoT (Internet of Things) deployments are vulnerable to local adversaries, compromising a LAN, and remote adversaries, compromising either the accounts associated with IoT devices or third-party devices like mobile phones used to control the IoT. There is, however, a fundamental difference between an attacker and a legitimate IoT user: the physical interaction with the device (e.g., via a mobile app) used to operate the IoT. Such physical interactions can be used to build frictionless authentications. However, their integration with IoT requires each vendor to independently adopt them, which is both complex and expensive. We instead design and build FIAT, the first third-party mechanism to automatically authorize IoT traffic by learning recurring traffic and validating human actions behind unpredictable traffic. FIAT does not require modification of the IoT devices or apps, as it operates passively on network traffic. Our evaluation shows that FIAT achieves high accuracy with minimal impact on the user experience.

The increasing complexity of JavaScript (JS) in modern mobile web pages has become a performance bottleneck for low-end mobile phone users, especially in developing regions. In this paper we propose SlimWeb, a novel approach that automatically derives lightweight versions of mobile web pages on-the-fly by eliminating non-essential JavaScript that does not impact the core page content and interactive functionality. SlimWeb consists of a JavaScript classification service powered by a supervised Machine Learning (ML) model that provides insights into each JavaScript element embedded in a web page. SlimWeb aims to improve the web browsing experience by predicting the class of each element, such that essential elements are preserved and non-essential elements are blocked by the browsers using the service. We motivate SlimWeb’s core design via a preference survey where 306 users overwhelmingly preferred having faster page load times over fetching various categories of non-essential JavaScript. We evaluate SlimWeb across 500 popular web pages in a developing region on real cellular networks, along with a user experience study with 20 real-world users and a usage willingness survey of 588 users. Evaluation results show that SlimWeb achieves 50% reduction in page load time compared to the original pages, and more than 30% reduction compared to competing solutions, while achieving high similarity scores to the original pages measured via a qualitative evaluation study with 62 users. SlimWeb improves the overall user experience metric (defined by Google Lighthouse combining first contentful paint, time to interactive, speed index) by more than 60% compared to the original pages, while maintaining 90-100% of the visual and functional components of most pages.

Social virtual reality (VR) has the potential to gradually replace traditional online social media, thanks to recent advances in consumer-grade VR devices and VR technology itself. As the vital foundation for building the Metaverse, social VR has been extensively examined by the computer graphics and HCI communities. However, there has been little systematic study dissecting the network performance of social VR, other than hype in the industry. To fill this critical gap, we conduct an in-depth measurement study of five popular social VR platforms: AltspaceVR, Horizon Worlds, Mozilla Hubs, Rec Room, and VRChat. Our experimental results reveal that all these platforms are still in their early stage and face fundamental technical challenges to realize the grand vision of Metaverse. For example, their throughput, end-to-end latency, and on-device computation resource utilization increase almost linearly with the number of users, leading to potential scalability issues. We identify the platform servers' direct forwarding of avatar data for embodying users without further processing as the main reason for the poor scalability and discuss potential solutions to address this problem. Moreover, while the visual quality of the current avatar embodiment is low and fails to provide a truly immersive experience, improving the avatar embodiment will consume more network bandwidth and further increase computation overhead and latency, making the scalability issues even more pressing.

To quickly create interactive web pages, developers heavily rely on (large) general-purpose JavaScript libraries. This practice bloats web pages with complex unused functions dead code which are unnecessarily downloaded and processed by the browser. The identification and the elimination of these functions is an open problem, which this paper tackles with Muzeel, a black-box approach requiring neither knowledge of the code nor execution traces. While the state-of-the-art solutions stop analyzing JavaScript when the page loads, the core design principle of Muzeel is to address the challenge of dynamically analyzing JavaScript after the page is loaded, by emulating all possible user interactions with the page, such that the used functions (executed when interactivity events fire) are accurately identified, whereas unused functions are filtered out and eliminated. We run Muzeel against 15,000 popular web pages and show that half of the 300,000 JavaScript files used in these pages have at least 70% of unused functions, accounting for 55% of the files' sizes. To assess the impact of dead code elimination on Mobile Web performance, we serve 200 Muzeel-ed pages to several Android phones and browsers, under variable network conditions. Our evaluation shows that Muzeel can speed up page loads by 25-30% thanks to a combination of lower CPU and bandwidth usage. Most importantly, we show that such savings are achieved while maintaining the pages' visual appearance and interactive functionality.

Due to the recent "work from home" trend, recent years have seen a growing research interest in understanding existing commercial videoconferencing systems in terms of their performance and architecture. One important question left unanswered that we tackle in this paper is: what is the performance of videoconferencing in the wild? Answering this generic question is challenging because it requires, ideally, a world-wide testbed composed of diverse devices (mobile, desktop), operating systems (Windows, MacOS, Linux) and network accesses (mobile and WiFi). In this paper, we present such a testbed that we develop to evaluate videoconferencing performance in the wild via automation for Android and Chromium-based browsers. We deploy our testbed via 85 distinct devices worldwide and collect performance metrics from 58 hours' worth of more than 2,000 videoconferencing sessions from 37 unique countries in the world. This, to the best of our knowledge, is the largest collection of videoconferencing performance data in the wild.

Online malware scanners are one of the best weapons in the arsenal of cybersecurity companies and researchers. A fundamental part of such systems is the sandbox that provides an instrumented and isolated environment (virtualized or emulated) for any user to upload and run unknown artifacts and identify potentially malicious behaviors. The provided API and the wealth of information in the reports produced by these services have also helped attackers test the efficacy of numerous techniques to make malware hard to detect. The most common technique used by malware for evading the analysis system is to monitor the execution environment, detect the presence of any debugging artifacts, and hide its malicious behavior if needed. This is usually achieved by looking for signals suggesting that the execution environment does not belong to a native machine, such as specific memory patterns or behavioral traits of certain CPU instructions. In this paper, we show how an attacker can evade detection on such analysis services by incorporating a Proof-of-Work (PoW) algorithm into a malware sample. Specifically, we leverage the asymptotic behavior of the computational cost of PoW algorithms when they run on some classes of hardware platforms to effectively detect a non bare-metal environment of the malware sandbox analyzer. To prove the validity of this intuition, we design and implement Scramblesuit, a framework to automatically (i) implement sandbox detection strategies, and (ii) embed a test evasion program into an arbitrary malware sample. We perform a comprehensive evaluation of Scramblesuit across a wide range of: 1) COTS architectures (ARM, Apple M1, i9, i7 and Xeon), 2) malware families, and 3) online sandboxes (JoeSandbox, Sysinternals, C2AE, Zenbox, Dr.Web VX Cube, Tencent HABO, YOMI Hunter). Our empirical evaluation shows that a PoW-based evasion technique is hard to fingerprint, and reduces existing malware detection rate by a factor of 10. The only plausible counter-measure to Scramblesuit is to rely on bare-metal online malware scanners, which is unrealistic given they currently handle millions of daily submissions.

Since the outbreak of the COVID-19 pandemic, videoconferencing has become the default mode of communication in our daily lives at homes, workplaces and schools, and it is likely to remain an important part of our lives in the post-pandemic world. Despite its significance, there has not been any systematic study characterizing the user-perceived performance of existing videoconferencing systems other than anecdotal reports. In this paper, we present a detailed measurement study that compares three major videoconferencing systems: Zoom, Webex and Google Meet. Our study is based on 62 hours’ worth of more than 1.1K videoconferencing sessions, which were created with a mix of emulated videoconferencing clients deployed in the cloud, as well as real mobile devices running from a residential network over two separate periods with nine months apart. We find that the existing videoconferencing systems vary in terms of geographic scope and resource provisioning strategies, which in turns determine streaming lag experienced by users. We also observe that streaming rate can change under different conditions (e.g., available bandwidth, number of users in a session, mobile device status), which affects user-perceived streaming quality. Beyond these findings, our measurement methodology enables reproducible benchmark analysis for any types of comparative or longitudinal study on available videoconferencing systems.

The average US household currently hosts more than 10 Internet of Things (IoT) devices [2]. Many research papers [5, 8] have demonstrated critical security concerns of the IoT, often due to lack of best practices like partial usage of HTTPS, or old ciphers. Even when best security practices are implemented, the IoT is still vulnerable to many attacks. Intruders can penetrate the home WiFi and directly control some IoT devices. They can compromise the account associated with an IoT device, mostly relying on username and password, or of third-party services like IFTTT [4]. They can also compromise the devices where IoT apps run, i.e., mostly mobile phones [1].

Since the outbreak of the COVID-19 pandemic, videoconferencing has become the default mode of communication in our daily lives at homes, workplaces and schools, and it is likely to remain an important part of our lives in the post-pandemic world. Despite its significance, there has not been any systematic study characterizing the user-perceived performance of existing videoconferencing systems other than anecdotal reports. In this paper, we present a detailed measurement study that compares three major videoconferencing systems: Zoom, Webex and Google Meet. Our study is based on 48 hours' worth of more than 700 videoconferencing sessions, which were created with a mix of emulated videoconferencing clients deployed in the cloud, as well as real mobile devices running from a residential network. We find that the existing videoconferencing systems vary in terms of geographic scope, which in turns determines streaming lag experienced by users. We also observe that streaming rate can change under different conditions (e.g., number of users in a session, mobile device status, etc), which affects user-perceived streaming quality. Beyond these findings, our measurement methodology can enable reproducible benchmark analysis for any types of comparative or longitudinal study on available videoconferencing systems.

The current Internet content delivery model assumes strict mapping between a resource and its descriptor, e.g., a JPEG file and its URL. Content Distribution Networks (CDNs) extend it by replicating the same resources across multiple locations, and introducing multiple descriptors. The goal of this work is to build Web-LEGO, an opt-in service, to speedup webpages at client side. Our rationale is to replace the slow original content with fast similar or equal content. Further, we perform a reality check of this idea both in term of the prevalence of CDN-less websites, availability of similar content, and user perception of similar webpages via millions of scale automated tests and thousands of real users. Then, we devise Web-LEGO, and address natural concerns on content inconsistency and copyright infringements. The final evaluation shows that Web-LEGO brings significant improvements both in term of reduced Page Load Time (PLT) and user-perceived PLT. Specifically, CDN-less websites provide more room for speedup than CDN-hosted ones, i.e., 7x more in the median case. Besides, Web-LEGO achieves high visual accuracy (94.2%) and high scores from a paid survey: 92% of the feedback collected from 1,000 people confirm Web-LEGO's accuracy as well as positive interest in the service.

Distributed Virtual Private Networks (dVPNs) are new solutions aiming to solve the trust-privacy concern of a VPN's central authority by leveraging a distributed architecture. In this paper, we discuss the requirements of a successful dVPN system and we present VPN-Zero: a dVPN system with strong privacy guarantees that provides traffic accounting and has minimal performance impact on its users. VPN-Zero guarantees that a dVPN node only carries traffic it has “allowlisted”, without revealing its allowlist or knowing the traffic it tunnels. This is achieved via three main innovations: (a) an attestation mechanism which leverages TLS to certify a user visit to a specific domain, (b) a zero-knowledge proof to certify that some incoming traffic is authorized (e.g., falls in a node's allowlist, without disclosing the target domain), and (c) a dynamic chain of VPN tunnels to both increase privacy and guarantee service continuation while traffic certification is in place. The paper demonstrates VPN-Zero functioning when integrated with two production systems: BitTorrent's Distributed Hash Table and ProtonVPN. Early evaluation results show that the median setup time of VPN-Zero is about 10 seconds.

Accurate web measurement is critical for understanding and improving security and privacy online. Such measurements implicitly assume that automated crawls generalize to typical web user experience. But anecdotal evidence suggests the web behaves differently when seen via well-known measurement endpoints or measurement automation frameworks, for various reasons. Our work improves the state of web privacy and security by investigating how key measurements differ when using naive crawling tool defaults vs. careful attempts to match “real” users across the Tranco top 25k web domains. We find web privacy and security measurements significantly affected by vantage point and browser configuration. We conclude that unless researchers ensure their web measurement tools match real world user experience, the research community is likely missing important signals systematically. For example, we find browser configuration alone causing shifts in 19% of known ad and tracking domains encountered and altering the loading frequency of up to 10% of distinct JavaScript code units executed. We find network vantage point having similar, though less dramatic, effects on the same web metrics. To ensure reproducibility, we carefully document our methodology and publish both our code and collected data.

The median webpage has increased in size by more than 80% in the last 4 years. This extra complexity allows for a rich browsing experience, but it hurts the majority of mobile users which still pay for their traffic. This has motivated several data-saving solutions, which aim at reducing the complexity of webpages by transforming their content. Despite each method being unique, they either reduce user privacy by further centralizing web traffic through data-saving middleboxes or introduce web compatibility (Web-compat) issues by removing content that breaks pages in unpredictable ways.

The last three decades have seen much evolution in web and network protocols: amongst them, a transition from HTTP/1.1 to HTTP/2 and a shift from loss-based to delay-based TCP congestion control algorithms. This paper argues that these two trends come at odds with one another, ultimately hurting web performance. Using a controlled synthetic study, we show how delay-based congestion control protocols (e.g., BBR and CUBIC + Hybrid Slow Start) result in the underestimation of the available congestion window in mobile networks, and how that dramatically hampers the effectiveness of HTTP/2. To quantify the impact of such finding in the current web, we evolve the web performance toolbox in two ways. First, we develop Igor, a client-side TCP congestion control detection tool that can differentiate between loss-based and delay-based algorithms by focusing on their behavior during slow start. Second, we develop a Chromium patch which allows fine-grained control on the HTTP version to be used per domain. Using these new web performance tools, we analyze over 300 real websites and find that 67% of sites relying solely on delay-based congestion control algorithms have better performance with HTTP/1.1.

Free and open source media centers are currently experiencing a boom in popularity for the convenience and flexibility they offer users seeking to remotely consume digital content. This newfound fame is matched by increasing notoriety—for their potential to serve as hubs for illegal content—and a presumably ever-increasing network footprint. It is fair to say that a complex ecosystem has developed around Kodi, composed of millions of users, thousands of “add-ons”—Kodi extensions from 3rd-party developers—and content providers. Motivated by these observations, this paper conducts the first analysis of the Kodi ecosystem. Our approach is to build “crawling” software around Kodi which can automatically install an addon, explore its menu, and locate (video) content. This is challenging for many reasons. First, Kodi largely relies on visual information and user input which intrinsically complicates automation. Second, no central aggregators for Kodi addons exist. Third, the potential sheer size of this ecosystem requires a highly scalable crawling solution. We address these challenges with de-Kodi, a full fledged crawling system capable of discovering and crawling large cross-sections of Kodi’s decentralized ecosystem. With de-Kodi, we discovered and tested over 9,000 distinct Kodi addons. Our results demonstrate de-Kodi, which we make available to the general public, to be an essential asset in studying one of the largest multimedia platforms in the world. Our work further serves as the first ever transparent and repeatable analysis of the Kodi ecosystem at large.

Recent advances in cloud computing have simplified the way that both software development and testing are performed. Unfortunately, this is not true for battery testing for which state of the art test-beds simply consist of one phone attached to a power meter. These test-beds have limited resources, access, and are overall hard to maintain; for these reasons, they often sit idle with no experiment to run. In this paper, we propose to share existing battery testing setups and build BatteryLab, a distributed platform for battery measurements. Our vision is to transform independent battery testing setups into vantage points of a planetary-scale measurement platform offering heterogeneous devices and testing conditions. In the paper, we design and deploy a combination of hardware and software solutions to enable BatteryLab's vision. We then preliminarily evaluate BatteryLab's accuracy of battery reporting, along with some system benchmarking. We also demonstrate how BatteryLab can be used by researchers to investigate a simple research question.

Today's webpages development cycle consists of constant iterations with the goal to improve user retention, time spent on site, and overall quality of experience. Big companies like Google, Facebook, Amazon, etc. invest a lot of time and money to perform online testing. The prohibitive costs of these approaches are an entry barrier for smaller players. Further, the lack of a substantial user-base can be problematic to ensure statistical significance within a reasonable duration. In this paper we propose Kaleidoscope, an automated tool to evaluate Web features at a large scale, quickly, accurately, and at a reasonable price. Kaleidoscope can test two crucial user-perceived Web features - the style and page loading. As far as we know, it is the first testing tool to replay page loading by controlling visual changes on a webpage. Kaleidoscope allows to concurrently load a webpage in two versions (e.g., different fonts, with vs without ads) that are shown to a participant side-by-side. Further, Kaleidoscope also allows a participant to interact with each webpage version and provide feedback, e.g., respond to a questionnaire previously prepared by an "experimenter". Kaleidoscope supports both voluntary and paid testers from FigureEight, a popular crowdsourcing platform. Using hundreds of FigureEight testers, we validate that Kaleidoscope matches the accuracy of trusted in-lab tests while providing results about 12x faster (and arguably at a lower cost) than A/B testing. Finally, we showcase how to use Kaleidoscope's page loading feature to study the user-perceived page load time (uPLT) of a webpage.

Personalized live video streaming is an increasingly popular technology that allows a broadcaster to share videos in real time with worldwide viewers. Compared to video-on-demand (VOD) streaming, experimenting with personalized live video streaming is harder due to its intrinsic live nature, the need for worldwide viewers, and a more complex data collection pipeline. In this paper, we make several contributions to both experimenting with and understanding today's commercial live video streaming services. First, we develop LIME (Live video MEasurement platform), a generic and holistic system allowing researchers to conduct crowd-sourced measurements on both commercial and experimental live streaming platforms. Second, we use LIME to perform, to the best of our knowledge, a first study of personalized 360° live video streaming on two commercial platforms, YouTube and Facebook. During a 7-day study, we have collected a dataset from 548 paid Amazon Mechanical Turk viewers from 35 countries who have watched more than 4,000 minutes of 360° live videos. Using this unique dataset, we characterize 360° live video streaming performance in the wild. Third, we conduct controlled experiments through LIME to shed light on how to make 360° live streaming (more) adaptive in the presence of challenging network conditions.

Today's HTTP adaptive streaming applications are designed to provide high levels of Quality of Experience (QoE) across a wide range of network conditions. The adaptation logic in these applications typically needs an estimate of the future network bandwidth for quality decisions. This estimation, however, is challenging in cellular networks because of the inherent variability of bandwidth and latency due to factors like signal fading, variable load, and user mobility. In this paper, we exploit machine learning (ML) techniques on a range of radio channel metrics and throughput measurements from a commercial cellular network to improve the estimation accuracy and hence, streaming quality. We propose a novel summarization approach for input raw data samples. This approach reduces the 90th percentile of absolute prediction error from 54% to 13%. We evaluate our prediction engine in a trace-driven controlled lab environment using a popular Android video player (ExoPlayer) running on a stock mobile device and also validate it in the commercial cellular network. Our results show that the three tested adaptation algorithms register improvement across all QoE metrics when using prediction, with stall reduction up to 85% and bitrate switching reduction up to 40%, while maintaining or improving video quality. Finally, prediction improves the video QoE score by up to 33%.

There has been a growing interest in measuring and optimizing the power efficiency of mobile apps. Traditional power evaluations rely either on inaccurate software-based solutions or on ad-hoc testbeds composed of a power meter and a mobile device. This demonstration presents BatteryLab, our solution to share existing battery testing setups to build a distributed platform for battery measurements. Our vision is to transform independent battery testing setups into vantage points of a planetary-scale measurement platform offering heterogeneous devices and testing conditions. We demonstrate BatteryLab functionalities by investigating the energy efficiency of popular websites when loaded via both Android and iOS browsers. Our demonstration is also live at https://batterylab.dev/.

Free web proxies promise anonymity and censorship circumvention at no cost. Several websites publish lists of free proxies organized by country, anonymity level, and performance. These lists index hundreds of thousands of hosts discovered via automated tools and crowd-sourcing. A complex free proxy ecosystem has been forming over the years, of which very little is known. In this article, we shed light on this ecosystem via a distributed measurement platform that leverages both active and passive measurements. Active measurements are carried out by an infrastructure we name ProxyTorrent, which discovers free proxies, assesses their performance, and detects potential malicious activities. Passive measurements focus on proxy performance and usage in the wild, and are accomplished by means of a Chrome extension named Ciao. ProxyTorrent has been running since January 2017, monitoring up to 230K free proxies. Ciao was launched in March 2017 and has thus far served roughly 9.7K users and generated 14TB of traffic. Our analysis shows that less than 2% of the proxies announced on the Web indeed proxy traffic on behalf of users; further, only half of these proxies have decent performance and can be used reliably. Every day, around 5%--10% of the active proxies exhibit malicious behaviors, e.g., advertisement injection, TLS interception, and cryptojacking, and these proxies are also the ones providing the best performance. Through the analysis of more than 14TB of proxied traffic, we show that web browsing is the primary user activity. Geo-blocking avoidance—allegedly a popular use case for free web proxies—accounts for 30% or less of the traffic, and it mostly involves countries hosting popular geo-blocked content.

Streaming over the wireless channel is challenging due to rapid fluctuations in available throughput. Encouraged by recent advances in cellular throughput prediction based on radio link metrics, we examine the impact on Quality of Experience (QoE) when using prediction within existing algorithms based on the DASH standard. By design, DASH algorithms estimate available throughput at the application level from chunk rates and then apply some averaging function. We investigate alternatives for modifying these algorithms, by providing the algorithms direct predictions in place of estimates or feeding predictions in place of measurement samples. In addition, we explore different prediction horizons going from one to three chunk durations. Furthermore, we induce different levels of error to ideal prediction values to analyse deterioration in user QoE as a function of average error.

While developing mobile apps is becoming easier, testing and characterizing their behavior is still hard. On the one hand, the de facto testing tool, called "Monkey," scales well due to being based on random inputs, but fails to gather inputs useful in understanding things like user engagement and attention. On the other hand, gathering inputs and data from real users requires distributing instrumented apps, or even phones with pre-installed apps, an expensive and inherently unscaleable task. To address these limitations we present CHIMP, a system that integrates automated tools and large-scale crowdsourced inputs. CHIMP is different from previous approaches in that it runs apps in a virtualized mobile environment that thousands of users all over the world can access via a standard Web browser. CHIMP is thus able to gather the full range of real-user inputs, detailed run-time traces of apps, and network traffic. We thus describe CHIMP»s design and demonstrate the efficiency of our approach by testing thousands of apps via thousands of crowdsourced users. We calibrate CHIMP with a large-scale campaign to understand how users approach app testing tasks. Finally, we show how CHIMP can be used to improve both traditional app testing tasks, as well as more novel tasks such as building a traffic classifier on encrypted network flows.

Free web proxies promise anonymity and censorship circumvention at no cost. Several websites publish lists of free proxies organized by country, anonymity level, and performance. These lists index hundreds of thousand of hosts discovered via automated tools and crowd-sourcing. A complex free proxy ecosystem has been forming over the years, of which very little is known. In this paper we shed light on this ecosystem via ProxyTorrent, a distributed measurement platform that leverages both active and passive measurements. Active measurements discover free proxies, assess their performance, and detect potential malicious activities. Passive measurements relate to proxy performance and usage in the wild, and are collected by free proxies users via a Chrome plugin we developed. ProxyTorrent has been running since January 2017, monitoring up to 180,000 free proxies and totaling more than 1,500 users over a 10 months period. Our analysis shows that less than 2% of the proxies announced on the Web indeed proxy traffic on behalf of users; further, only half of these proxies have decent performance and can be used reliably. Around 10% of the working proxies exhibit malicious behaviors, e.g., ads injection and TLS interception, and these proxies are also the ones providing the best performance. Through the analysis of more than 2 Terabytes of proxied traffic, we show that web browsing is the primary user activity. Geo-blocking avoidance is not a prominent use-case, with the exception of proxies located in countries hosting popular geo-blocked content.

The functioning of mobile apps involves a large number of protocols and entities, with the Domain Name System (DNS) acting as a predominant one. Despite being one of the oldest Internet systems, DNS still operates with semi-obscure interactions among its stakeholders: domain owners, network operators, operating systems, and app developers. The goal of this work is to holistically understand the dynamics of DNS in mobile traffic along with the role of each of its stakeholders. We use two complementary (anonymized) datasets: traffic logs provided by a European mobile network operator (MNO) with 19M customers, and traffic logs from 5,000 users of Lumen, a traffic monitoring app for Android. We complement such passive traffic analysis with active measurements at four European MNOs. Our study reveals that 10k domains (out of 198M) account for 87% of total network flows. The time to live (TTL) values for such domains are mostly short (< 1min), despite domain-to-IPs mapping tends to change on a longer time-scale. Further, depending on the operators recursive resolver architecture, end-user devices receive even smaller TTL values leading to suboptimal effectiveness of the on-device DNS cache. Despite a number of on-device and in-network optimizations available to minimize DNS overhead, which we find corresponding to 10% of page load time (PLT) on average, we have not found wide evidence of their adoption in the wild.

360-degree videos are becoming increasingly popular on commercial platforms. In this position paper, we propose a holistic research agenda aiming at improving the performance, resource utilization efficiency, and users' quality of experience (QoE) for 360° video streaming on commodity mobile devices. Based on a Field-of-View (FoV) guided approach that fetches only portions of a scene that users will see, our proposed research includes the following: robust video rate adaptation with incremental chunk upgrading, big-data-assisted head movement prediction and rate adaptation, novel support for multipath streaming, and enhancements to live 360° video broadcast. We also show preliminary results demonstrating promising performance of our proof-of-concept 360° video streaming system on which our proposed research are being prototyped, integrated, and evaluated.

As researchers, we are aware of how hard it is to obtain access to vantage points in the Internet. Experimentation platforms are useful tools, but they are also: 1) paid, either via a membership fee or by resource sharing, 2) unreliable, nodes come and go, 3) outdated, often still run on their original hardware and OS. While one could build yet-another platform with up-to-date and reliable hardware and software, it is hard to imagine one which is free. This is the goal of this paper: we set out to build FreeLab, a free experimentation platform which also aims to be reliable and up-to-date. The key idea behind FreeLab is that experiments run directly at its user machines, while traffic is relayed by free vantage points in the Internet (web and SOCKS proxies, and DNS resolvers). FreeLab is thus free of access by design and up-to-date as far as its users maintain their experimenting machines. Reliability is a key challenge due to the volatile nature of free resources, and the introduction of errors (path inflation, header manipulation, bandwidth shrinkage) caused by traffic relays.

Tremendous effort has gone into the ongoing battle to make webpages load faster. This effort has culminated in new protocols (QUIC, SPDY, and HTTP/2) as well as novel content delivery mechanisms. In addition, companies like Google and SpeedCurve investigated how to measure "page load time" (PLT) in a way that captures human perception. In this paper we present Eyeorg [12], a platform for crowdsourcing web quality of experience measurements. Eyeorg overcomes the scaling and automation challenges of recruiting users and collecting consistent user-perceived quality measurements. We validate Eyeorg's capabilities via a set of 100 trusted participants. Next, we showcase its functionalities via three measurement campaigns, each involving 1,000 paid participants, to 1) study the quality of several PLT metrics, 2) compare HTTP/1.1 and HTTP/2 performance, and 3) assess the impact of online advertisements and ad blockers on user experience. We find that commonly used, and even novel and sophisticated PLT metrics fail to represent actual human perception of PLT, that the performance gains from HTTP/2 are imperceivable in some circumstances, and that not all ad blockers are created equal.

The rapid growth of server virtualization has ignited a wide adoption of software-based virtual switches, with significant interest in speeding up their performance. In a similar trend, software-defined networking (SDN), with its strong reliance on rule-based flow classification, has also created renewed interest in multi-dimensional packet classification. However, despite these recent advances, the performance of current software-based packet classifiers is still limited, mostly by the low parallelism of general-purpose CPUs. In this paper, we explore how to accelerate packet classification using the high parallelism and latency-hiding capabilities of graphic processing units (GPUs). We implement GPU-accelerated versions for both linear and tuple search, currently deployed in virtual switches, and also introduce a novel algorithm called Bloom search. These algorithms are integrated with high-speed packet I/O to build GSwitch, a GPU-accelerated software switch, and also to extend Open vSwitch. Our experimental evaluation indicates that, under realistic rule sets, GSwitch is at least 7 × faster than an equally-priced CPU classifier. We also show that our GPU-accelerated Open vSwitch outperforms the classic Open vSwitch implementation by a factor of 10, on average.

A significant fraction of Internet traffic is now encrypted and HTTPS will likely be the default in HTTP/2. However, Transport Layer Security (TLS), the standard protocol for encryption in the Internet, assumes that all functionality resides at the endpoints, making it impossible to use in-network services that optimize network resource usage, improve user experience, and protect clients and servers from security threats. Re-introducing in-network functionality into TLS sessions today is done through hacks, often weakening overall security.

Consider a network of prosumers of media content in which users dynamically create and request content objects. The request process is governed by the objects' popularity, which may vary across network regions and over time. In order to meet user requests, content objects can be stored and transported over the network, characterized by the capacity and efficiency of its storage and transport resources. The energy-efficient dynamic content distribution problem aims at finding the evolution of the network configuration, in terms of the placement and routing of content objects over time, that meets user requests, satisfies network resource capacities and minimizes overall energy use. We present 1) an information-centric linear programming formulation for the energy efficient dynamic content distribution problem that captures multicasting and caching over the network, per-object system dynamics, and delivery deadlines; 2) an offline solution that characterizes the minimum energy use achievable with global knowledge of user requests and network resources; and 3) an efficient distributed online solution that allows network nodes to make caching decisions based on their local estimate of the global energy benefit. Using a custom-built content distribution network simulator as well as a real prototype implementation in an information-centric networking testbed, we show the significant energy savings that can be obtained via the efficient and lightweight cache cooperation induced by our service and energy aware distributed online solution with respect to state of the art approaches.

Netflix and Hulu are leading Over-the-Top (OTT) content service providers in the US and Canada. Netflix alone accounts for 29.7% of the peak downstream traffic in the US in 2011. Understanding the system architectures and performance of Netflix and Hulu can shed light on the design of such large-scale video streaming platforms, and help improving the design of future systems. In this paper, we perform extensive measurement study to uncover their architectures and service strategies. Netflix and Hulu bear many similarities. Both Netflix and Hulu video streaming platforms rely heavily on the third-party infrastructures, with Netflix migrating that majority of its functions to the Amazon cloud, while Hulu hosts its services out of Akamai. Both service providers employ the same set of three content distribution networks (CDNs) in delivering the video contents. Using active measurement study, we dissect several key aspects of OTT streaming platforms of Netflix and Hulu, e.g., employed streaming protocols, CDN selection strategy, user experience reporting, etc. We discover that both platforms assign the CDN to a video request without considering the network conditions and optimizing the user-perceived video quality. We further conduct the performance measurement studies of the three CDNs employed by Netflix and Hulu. We show that the available bandwidths on all three CDNs vary significantly over the time and over the geographic locations. We propose a measurement-based adaptive CDN selection strategy and a multiple-CDN-based video delivery strategy that can significantly increase users' average available bandwidth.

Internet users are interested in content regardless of its location; however, the current client/server architecture still requires requests to be directed to a specific server. Information-centric networking (ICN) is a recent vein that relaxes this requirement through the use of name-based forwarding, where forwarding decisions are based on content names instead of IP addresses. Despite previous name-based forwarding strategies have been proposed, almost none have actually built a content router. To fill this gap, in this paper we design and prototype a content router called Caesar for high-speed forwarding on content names. Caesar introduces several innovative features, including (i) a longest-prefix matching algorithm based on a novel data structure called prefix Bloom filter; (ii) an incremental design which allows for easy integration with existing protocols and network equipment;(iii) a forwarding scheme where multiple line cards collaborate in a distributed fashion; and (iv) support for offloading packet processing to graphics processing units (GPUs). We build Caesar as an enterprise router, and show that every line card sustains up to 10 Gbps using a forwarding table with more than 10 million content prefixes. Distributed forwarding allows the forwarding table to grow even further, and to scale linearly with the number of line cards at the cost of only a few microseconds in the packet processing latency. GPU offloading, in turn, trades off a few milliseconds of latency for a large speedup in the forwarding rate.

The rapid growth of server virtualization has ignited a wide adoption of software-based virtual switches, with significant interest in speeding up their performance. In a similar trend, software-defined networking (SDN), with its strong reliance on rule-based flow classification, has also created renewed interest in multi-dimensional packet classification. However, despite these recent advances, the performance of current software-based packet classifiers is still limited, mostly by the low parallelism of general-purpose CPUs. In this paper, we explore how to accelerate packet classification using the high parallelism and latency-hiding capabilities of graphic processing units (GPUs). We implement GPU-accelerated versions for both linear and tuple search, currently deployed in virtual switches, and also introduce a novel algorithm called Bloom search. These algorithms are integrated with high-speed packet I/O to build GSwitch, a GPU-accelerated software switch. Our experimental evaluation shows that GSwitch is at least 7x faster than an equally-priced CPU classifier and is able to reach 10 Gbps with minimum-sized packets and a rule set containing 128K OpenFlow entries with 512 different wildcard patterns.

On-path caching has been proved to be effective in shortening the content retrieval time and educing the traffic workload imposed on the underlying network. In this paper, we study the effectiveness of on-path caching for the content distribution in the wireless network environment with GPSR as the underlying routing protocol. We extend the general cache network model introduced in [14] to account for GPSR's loopy paths. We further develop the performance models to take into account the effect of wireless link loss, different error recovery strategies, and the GPSR routing induced long and loopy paths. The performance models allow us to assess the average content availability, the average number of messages per content retrieval, and the average content retrieval latencies for an on-path caching enabled content distribution network over MANETs. The results show that on-path caching can significantly improve the content availability, reduce the average number of required messages to acquire a content object, and shorten the content retrieval latency. In addition, the study shows that the on-path caching is able to mitigate the negative impact caused by the long and loopy paths due to the GPSR routing.

As wireless nodes become more powerful, a large volume of diverse content, such as images or videos, become available in a mobile ad hoc network (MANET). It follows that a MANET can largely benefit from the support for "range queries", i.e., a query for content whose attributes are within a requested range. Supporting range queries over MANET is challenging due to node mobility and adverse medium conditions. In this work, we argue that Information- Centric Networking (ICN), a novel networking paradigm that decouples resources from the host they reside on, can enable efficient support for range queries in MANET. Accordingly, we design and evaluate ERMINE, a solution for "Enabling Range queries in MANET with Information-centric NEtworking".

Consider a network of prosumers of media content in which users dynamically create and request content objects. The request process is governed by the objects' popularity and varies across network regions and over time. In order to meet user requests, content objects can be stored and transported over the network, characterized by the capacity and energy efficiency of the storage and transport resources. The energy efficient dynamic in-network caching problem aims at finding the evolution of the network configuration, in terms of the content objects being cached and transported over each network element at any given time, that meets user requests, satisfies network resource capacities and minimizes overall energy use. We provide 1) an information-centric optimization framework for the energy efficient dynamic in-network caching problem, 2) an offline solution, EE-OFD, based on an integer linear program (ILP) that obtains the maximum efficiency gains that can be achieved with global knowledge of user requests and network resources, and 3) an efficient fully distributed online solution, EEOND, that allows network nodes to make local caching decisions based on their current estimate of the global energy benefit. Our solutions take into account the network heterogeneity, in terms of capacity, energy efficiency and content popularity, and adapt to changing network conditions minimizing overall energy use.

BitTorrent is both the dominant Peer-to-Peer (P2P) protocol for file-sharing and a nightmare for ISPs due to its network agnostic nature. Many solutions exist to localize BitTorrent traffic relying on cooperation between ISPs and the trackers. Recently, BitTorrent users have been abandoning the trackers in favor of Distributed Hash Tables (DHTs). Despite DHTs are complex heterogeneous systems, DHT-based traffic localization is also possible; however, it is unclear how it performs. The goal of this work is to measure DHT-based traffic localization in the wild. We run multiple experiments involving up to five commercial ISPs and a maximum duration of one month, collecting about 400 GB of BitTorrent traffic. Then, we perform an extensive analysis with the following goals: understand the impact of system parameters, verify accuracy of the measurements, estimate the localization benefits.

Netflix is the leading provider of on-demand Internet video streaming in the US and Canada, accounting for 29.7% of the peak downstream traffic in US. Understanding the Netflix architecture and its performance can shed light on how to best optimize its design as well as on the design of similar on-demand streaming services. In this paper, we perform a measurement study of Netflix to uncover its architecture and service strategy. We find that Netflix employs a blend of data centers and Content Delivery Networks (CDNs) for content distribution. We also perform active measurements of the three CDNs employed by Netflix to quantify the video delivery bandwidth available to users across the US. Finally, as improvements to Netflix's current CDN assignment strategy, we propose a measurement-based adaptive CDN selection strategy and a multiple-CDN-based video delivery strategy, and demonstrate their potentials in significantly increasing user's average bandwidth.

Social virtual worlds such as Second Life (SL) are digital representations of the real world where human-controlled avatars evolve and interact through social activities. Understanding the characteristics of virtual worlds can be extremely valuable in order to optimize their design. In this paper, we perform an extensive analysis of SL. We exploit standard avatar capabilities to monitor the virtual world, and we emulate avatar behaviors in order to evaluate user experience. We make several surprising observations. We find that 30% of the regions are never visited during the six-day monitoring period, whereas less than 1% of the regions have large peak populations. Moreover, the vast majority of regions are static, i.e., objects are seldom created or destroyed. Interestingly, we show that avatars interact similarly to humans in real life, gathering in small groups of 2-10 avatars. We also show that user experience is poor. Most of the time, avatars have an incorrect view of their neighbor avatars, and inconsistency can last several seconds, impacting interactivity among avatars.

Second Life (SL) is a virtual world where people interact and socialize through virtual avatars. Avatars behave similarly to their human counterparts in real life and naturally define a social network. However, not only human-controlled avatars participate in the social network. Automated avatars called bots are common, difficult to identify and, when malicious, can severely detract from the user experience of SL. In this paper we study the SL social network and the role of bots within it. Using traces of avatars in a popular SL region, we analyze the social graph formed by avatar interactions. We find that it resembles natural networks more than other online social networks, and that bots have a fundamental impact on the SL social network. Finally, we propose a bot detection strategy based on the importance of the social connections of avatars in the social graph.

Applications such as Second Life require massive deployment of servers worldwide to support a large number of users. We investigate experimentally how Peer-to-Peer (P2P) communication could help cut the deployment cost and increase the scalability of Social Virtual Worlds such as Second Life. We design and build a communication infrastructure that distributes the management of the virtual world among user resources using a structured P2P network. Our communication infrastructure is implemented on the top of Kad, the P2P network that supports millions of eMule users. We then use avatar and object traces collected on Second Life to perform a realistic emulation of P2P Second Life over the Internet. We show that, despite using a standard P2P solution, P2P Second Life is mostly consistent, persistent and scalable. However, the latency avatars experience to recover from an inconsistent view of the virtual world can become disturbing for very large numbers of participants and objects. We analyze and discuss this limitation and give recommendation on how to design P2P Social Virtual Worlds.

Social virtual worlds such as Second Life are digital representations of the real world where human-controlled avatars evolve and interact through social activities. Understanding the characteristics of existing virtual worlds can be extremely valuable to optimize their design. In this work we perform the first extensive analysis of Second Life. We have crawled around 13000 Regions over one month, and gathered information about objects, avatars, and server state. The analysis of our traces shows several surprising results. We find that 30% of the Regions are never visited during a six day period, whereas only few Regions have large peak populations. Moreover, the vast majority of Regions are static, i.e., objects are seldom created or destroyed. Interestingly, avatars interact similarly to humans in real life, gathering in small groups, visiting the same places and meeting the same avatars again, showing a highly predictable behavior. Based on these observations, we discuss several techniques to enhance Second Life or other similar social virtual worlds.

A Networked Virtual Environment (NVE) is a digital world where multiple participants interact via virtual characters called avatar. A popular application for NVEs is Second Life (SL)[5].

We propose Push-to-Peer, a peer-to-peer system to cooperatively stream video. The main departure from previous work is that content is proactively pushed to peers, and persistently stored before the actual peer-to-peer transfers. The initial content placement increases content availability and improves the use of peer uplink bandwidth. Our specific contributions are: (i) content placement and associated pull policies that allow the optimal use of uplink bandwidth; (ii) performance analysis of such policies in controlled environments such as DSL networks under ISP control; (iii) a distributed load balancing strategy for selection of serving peers.