Search

The highly dynamic wireless communication environment poses a challenge for many applications (e.g., adaptive multimedia streaming services). Providing accurate TP can significantly improve performance of these applications. The scheduling algorithms in cellular networks consider various PHY metrics, (e.g., CQI) and throughput history when assigning resources for each user. This article explains how AI can be leveraged for accurate TP in cellular networks using PHY and application layer metrics. We present key architectural components and implementation options, illustrating their advantages and limitations. We also highlight key design choices and investigate their impact on prediction accuracy using real data. We believe this is the first study that examines the impact of integrating network-level data and applying a deep learning technique (on PHY and application data) for TP in cellular systems. Using video streaming as a use case, we illustrate how accurate TP improves the end user's QoE. Furthermore, we identify open questions and research challenges in the area of AI-driven TP. Finally, we report on lessons learned and provide conclusions that we believe will be useful to network practitioners seeking to apply AI.

Large data transfers can result in significant congestion and performance degradation for interactive end-user applications such as web browsing and streaming. While there are existing TCP congestion control algorithms for delivery of large volume data (e.g., LEDBAT, TCP-LP), our results show that these protocols are not effective in cellular networks due to variability in radio channel conditions and the use of cellular schedulers in base stations. We propose Legilimens, an agile TCP variant for cellular downlink transfers, which not only retains desirable properties of existing approaches, but also exploits the properties of the cellular schedulers to estimate load and capacity and addresses the challenges in cellular networks. As a result, Legilimens is able to deliver traffic using only the spare capacity on the downlink. We conduct extensive evaluations of Legilimens in multiple settings—in a large cellular network for real-world performance, on the PhantomNet emulator for controlled experiments, and ns-3 simulator for scaled experiments—all of which demonstrate that Legilimens is superior to existing protocols in transferring large volumes of data without interfering with regular user traffic. Compared to existing low-priority protocols, Legilimens improves the throughput of background flows by 2x on average (up to 5x) without degrading the performance of foreground flows across all the three testbeds.

This special issue of Internet Computing includes articles that tackle both the technical challenges and vast opportunities of AI-Powered 5G services. The first two papers deal with the use of AI and Machine Learning for improved resource utilization in 5G cellular networks, and provisioning in the 5G Core, respectively. The third paper presents a cautionary perspective by examining AI/ML for 5G and beyond through an adversarial lens, and finally, the last paper covers the study of a scalable 5G-based solution for real-time air quality monitoring that has the potential to improve the lives of millions impacted by air quality issues.

Distributed clouds offer a choice of data center locations for providers to host their applications. In this paper, we consider distributed clouds that host virtual desktops which are then accessed by users through remote desktop protocols. Virtual desktops have different levels of latency-sensitivity, primarily determined by the actual applications running and affected by the end users’ locations. In the scenario of mobile users, even switching between 3G and WiFi networks affects the latency-sensitivity. We design VMShadow, a system to automatically optimize the location and performance of latency-sensitive VMs in the cloud. VMShadow performs black-box fingerprinting of a VM’s network traffic to infer the latency-sensitivity and employs both ILP and greedy heuristic based algorithms to move highly latency-sensitive VMs to cloud sites that are closer to their end users. VMShadow employs a WAN-based live migration and a new network connection migration protocol to ensure that the VM migration and subsequent changes to the VM’s network address are transparent to end-users. We implement a prototype of VMShadow in a nested hypervisor and demonstrate its effectiveness for optimizing the performance of VM-based desktops in the cloud. Our experiments on a private as well as the public EC2 cloud show that VMShadow is able to discriminate between latency-sensitive and insensitive desktop VMs and judiciously moves only those that will benefit the most from the migration. For desktop VMs with video activity, VMShadow improves VNC’s refresh rate by 90% by migrating virtual desktop to the closer location. Transcontinental remote desktop migrations only take about 4 min and our connection migration proxy imposes 13 μs overhead per packet.

Network functions virtualization is an emerging technology that can significantly improve the flexibility of network service provisioning and offer potential cost savings. However, it is critical that service providers offer high reliability and availability of the network and services when moving from proprietary hardware appliances to virtualized network functions on commodity servers. In a network consisting of physical appliances, providers can deploy redundant hardware and extra capacity to handle failures, although this is quite expensive in practice. Virtualization of network functions lead to more challenges for resiliency, but also bring new opportunities to address these challenges in a more cost-effective manner. In this article, we explain the resiliency requirements for virtual network functions in order to provide carrier grade services, summarize the existing solutions in the literature, highlight several research challenges, and present a concrete case study to demonstrate how to decompose a type of virtual router and thus enhance its resiliency.

Traffic migration is a common procedure performed by operators during planned maintenance and unexpected incidents to prevent/reduce service disruptions. However, current practices of traffic migration often couple operators' intentions (e.g. device upgrades) with network setups (e.g. load-balancers), resulting in poor re-usability and substantial operational complexities. Our study of 205 Methods of Procedure (MOPs) from a major U.S. carrier suggests that generalizing traffic migration with a unified model is feasible. Such generalization along with SDN's automation capability is key to scalable and flexible management of traffic, especially for virtualized network functions with unprecedented scale, heterogeneity, and fast iteration. In this paper, we propose Egret, a generic traffic migration system that simplifies traffic management for physical and virtual network functions. Egret (1) hides intricate implementation details from operators with generic intention-based interfaces, and (2) modularizes common traffic migration procedures to enable plug-and-play by developers and vendors. Leveraging a novel mask-based abstraction of traffic migration jobs, Egret can further simplify reverse traffic migration and enable job interleaving.

In this demo, we showcase how to stream 8K 360° video to commodity mobile devices without pre-processing or viewpoint prediction using our MEC-VR system. Users can freely select 360° video from YouTube and immediately become immersed in the video of 8K resolution using a Samsung GearVR compatible smartphone. Our system can support zoom in and zoom out while saving up to 80% bandwidth.

Volumetric videos offer six degree-of-freedom (DoF) as well as 3D rendering, making them highly immersive, interactive, and expressive. In this paper, we design Nebula, a practical and resource-efficient volumetric video streaming system for commodity mobile devices. Our design leverages edge computing to reduce the computation burden on mobile clients. We also introduce various optimizations to lower the perceived "motion-to-photon" delay, to dynamically adapt to the fluctuating network bandwidth, and to reduce the system's resource consumption while maintaining a high QoE.

Network functions virtualization (NFV) promises to bring significant flexibility and cost savings to networking. These improvements are predicated on being able to run many virtualized network elements on a server, which leads to a fundamental question on how scalable an NFV platform can be. With this in mind, the authors build an experimental platform with commonly used NFV technologies. They evaluate the NFVs performance and scalability and, based on their demonstrated improvements, discuss best practices for achieving optimum NFV performance on commodity hardware. They also reveal the limitations on NFV scalability and propose a new architecture to address them.

Previous studies of peer-to-peer (P2P) video-on-demand (VoD) are performed separately from studies utilizing adaptive bit rate video since the techniques seemingly tackle orthogonal goals. Additionally, previous policies used by P2P VoD do not account for viewer abandonment of video during download and playback. Through analysis, we show that the popularity of a P2P swarm and seed staying time significantly affects the achievable per-receiver download rate. Specifically, we identify conditions under which popularity affects swarm efficiency, contradicting a typical misconception in previous work, and we show that abandonment under these previous policies significantly increases playback interruptions. In light of these observations, we propose Joint-Family, a protocol that supports HTTP-based adaptive bitrate streaming for on-demand videos using P2P techniques. Joint-Family accounts for user video viewing behavior, such as abandonment, to improve the quality of experience for the viewer. Peers in Joint-Family simultaneously participate in multiple swarms to exchange chunks of different bitrates. Joint-Family takes advantage of abandonment by converting peers to "partial seeds" to increase system capacity. Joint-Family adopts chunk, bitrate, and peer selection policies that minimize occurrence of interruptions in the presence of abandonment while delivering high quality video and improving the total capacity of the system. Using traces from a large-scale commercial VoD service, we compare Joint-Family with existing approaches for P2P VoD and show that viewers in Joint-Family enjoy higher playback rates with minimal interruption, irrespective of video popularity.

IPTV service providers offering Video-on-Demand currently use servers at each metropolitan office to store all the videos in their library. With the rapid increase in library sizes, it will soon become infeasible to replicate the entire library at each office. We present an approach for intelligent content placement that scales to large library sizes (e.g., 100 Ks of videos). We formulate the problem as a mixed integer program (MIP) that takes into account constraints such as disk space, link bandwidth, and content popularity. To overcome the challenges of scale, we employ a Lagrangian relaxation-based decomposition technique combined with integer rounding. Our technique finds a near-optimal solution (e.g., within 1%–2%) with orders of magnitude speedup relative to solving even the linear programming (LP) relaxation via standard software. We also present simple strategies to address practical issues such as popularity estimation, content updates, short-term popularity fluctuation, and frequency of placement updates. Using traces from an operational system, we show that our approach significantly outperforms simpler placement strategies. For instance, our MIP-based solution can serve all requests using only half the link bandwidth used by least recently used (LRU) or least frequently used (LFU) cache replacement policies. We also investigate the tradeoff between disk space and network bandwidth.

Flare is a practical system for streaming 360-degree videos on commodity mobile devices. It takes a viewport-adaptive approach, which fetches only portions of a panoramic scene that cover what a viewer is about to perceive. We conduct an IRB-approved user study where we collect head movement traces from 130 diverse users to gain insights on how to design the viewport prediction mechanism for Flare. We then develop novel online algorithms that determine which spatial portions to fetch and their corresponding qualities. We also innovate other components in the streaming pipeline such as decoding and server-side transmission. Through extensive evaluations (~400 hours' playback on WiFi and ~100 hours over LTE), we show that Flare significantly improves the QoE in real-world settings. Compared to non-viewport-adaptive approaches, Flare yields up to 18x quality level improvement on WiFi, and achieves high bandwidth reduction (up to 35%) and video quality enhancement (up to 4.9x) on LTE.

Flare is a practical system for streaming 360-degree videos on smartphones. It takes a viewport-adaptive approach, which fetches only portions of a panoramic scene that cover what a viewer is about to perceive. Flare consists of a novel framework for the end-to-end streaming pipeline, introduces innovative streaming algorithms, and brings numerous system-level optimizations. In our demo, we will show that Flare substantially outperforms traditional viewport-agnostic streaming algorithms in terms of the video quality. We will also invite the audience to use Flare to watch attractive 360-degree videos.

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. We find that by applying accurate prediction to three algorithms, user QoE can improve up to 55% depending on the algorithm in use. Furthermore having longer horizon positively affects QoE metrics. Accurate predictions have the most significant impact on stall performance by completely eliminating them. Prediction also improves switching behaviour significantly and longer prediction horizons enable a client to promptly reduce quality and avoid stalls when the throughput drops for a relatively long time that can deplete the buffer. For all algorithms, a 3-chunk horizon strikes the best balance between different QoE metrics and, as a result, achieving highest user QoE. While error-induced predictions significantly lower user QoE in certain situations, on average, they provide 15% improvement over DASH algorithms without any prediction.

As Augmented Reality (AR) ties closely to the physical world, its users looking at overlapped scenes are likely to be in the vicinity of each other, which naturally enables the collaboration and interaction among them. In this paper, we propose CARS (Collaborative Augmented Reality for Socialization), a framework that leverages the social nature of human beings to improve the user-perceived Quality of Experience (QoE) for AR, especially the end-to-end latency. CARS takes advantage of the unique feature of AR to support intelligent sharing of information between nearby users when it is feasible. It brings various benefits at the user, application and system levels, e.g., reduction of end-to-end latency and reuse of networking and computation resources. We demonstrate the efficacy of CARS through a preliminary evaluation based on a prototype implementation.

Despite its widespread adoption and popularity, the Hypertext Transfer Protocol (HTTP) suffers from fundamental performance limitations. SPDY, a recently proposed alternative to HTTP, tries to address many of the limitations of HTTP (e.g., multiple connections, setup latency). In this paper, we perform a detailed measurement study to understand the benefits of using SPDY over cellular networks. Through careful measurements conducted over 4 months, we provide a detailed analysis of the performance of HTTP and SPDY, how they interact with the various layers, and their implications on Web design. Our results show that unlike in wired and 802.11 networks, SPDY does not clearly outperform HTTP over cellular networks. We identify negative interactions between the protocols used for Web access (HTTP/SPDY over TCP) and cellular radio resource management as the underlying cause. Overall performance suffers when devices go through a cellular radio state promotion after an idle period, and the consequent increase in latency. This impacts SPDY more because of the use of a single TCP connection. We conclude that a viable solution has to account for these unique cross-layer dependencies to achieve improved Web performance over cellular networks.

Network function virtualization was recently proposed to improve the flexibility of network service provisioning and reduce the time to market of new services. By leveraging virtualization technologies and commercial off-the-shelf programmable hardware, such as general-purpose servers, storage, and switches, NFV decouples the software implementation of network functions from the underlying hardware. As an emerging technology, NFV brings several challenges to network operators, such as the guarantee of network performance for virtual appliances, their dynamic instantiation and migration, and their efficient placement. In this article, we provide a brief overview of NFV, explain its requirements and architectural framework, present several use cases, and discuss the challenges and future directions in this burgeoning research area.

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.

Connected cars are a rapidly growing segment of Internet of Things (IoT). While they already use cellular networks to support emergency response, in-car WiFi hotspots and infotainment, there is also a push towards updating their firmware over-the-air (FOTA). With millions of connected cars expected to be deployed over the next several years, and more importantly persist in the network for a long time, it is important to understand their behavior, usage patterns, and impact --- both in terms of their experience, as well as other users. Using one million connected cars on a production cellular network, we conduct network-scale measurements of over one billion radio connections to understand various aspects including their spatial and temporal connectivity patterns, the network conditions they face, use and handovers across various radio frequencies and mobility patterns. Our measurement study reveals that connected cars have distinct sets of characteristics, including those similar to regular smartphones (e.g. overall diurnal pattern), those similar to IoT devices (e.g. mostly short network sessions), but also some that belong to neither type (e.g. high mobility). These insights are invaluable in understanding and modeling connected cars in a cellular network and in designing strategies to manage their data demand.

We consider the problem of managing Firmware Over-The-Air (FOTA) updates for cars at a massive scale over a cellular network. This problem is constrained by factors like large number of cars, a need to perform updates quickly and securely, ability to do the update anywhere and any time, and delivering the often large update without harming the network. We present a scheduling approach for managing large FOTA downloads in a cellular network that combines historical network load with car usage and location analytics. The proposed approach uses a combination of heuristics and an optimized scheduler and applies them to schedule the admission of cars to download FOTA. Using real network data from a large cellular provider that includes a million cars and nearly a billion radio-level network connections, we show that our scheduling approach is feasible and practical. We also show that it is possible to manage FOTA even when device and network models use high levels of aggregation over time. Simulation results show that our method improves over random uncontrolled approaches by (i) reducing median download startup delay by 48%, (ii) reducing the number of cars that don't complete the update by 10% and most importantly, (iii) reducing the load in busy cells by up to 37%.

The availability of reliable predictions for cellular throughput would offer a fundamental change in the way applications are designed and operated. Numerous cellular applications, including video streaming and VoIP, embed logic that attempts to estimate achievable throughput and adapt their behaviour accordingly. We believe that providing applications with reliable predictions several seconds into the future would enable profoundly better adaptation decisions and dramatically benefit demanding applications like mobile virtual and augmented reality. The question we pose and seek to address is whether such reliable predictions are possible. We conduct a preliminary study of throughput prediction in a cellular environment using statistical machine learning techniques. An accurate prediction can be very challenging in large scale cellular environments because they are characterized by highly fluctuating channel conditions. Using simulations and real-world experiments, we study how prediction error varies as a function of prediction horizon, and granularity of available data. In particular, our simulation experiments show that the prediction error for mobile devices can be reduced significantly by combining measurements from the network with measurements from the end device. Our results indicate that it is possible to accurately predict achievable throughput up to 8 sec in the future where 50th percentile of all errors are less than 15% for mobile and 2% for static devices.

Despite much recent progress, Web page latencies over cellular networks remain much higher than those over wired networks. Proxies that execute Web page JavaScript (JS) and push objects needed by the client can reduce latency. However, a key concern is the scalability of the proxy which must execute JS for many concurrent users. In this paper, we propose to scale the proxies, focusing on a design where the proxy's execution is solely to push the needed objects and the client completely executes the page as normal. Such redundant execution is a simple, yet effective approach to cutting network latencies, which dominate page load delays in cellular settings. We develop whittling, a technique to identify and execute in the proxy only the JS code necessary to identify and push the objects required for the client page load, while skipping other code. Whittling is closely related to program slicing, but with the important distinction that it is acceptable to approximate the program slice in the proxy given the client's complete execution. Experiments with top Alexa Web pages show NutShell can sustain, on average, 27\% more user requests per second than a proxy performing fully redundant execution, while preserving, and sometimes enhancing, the latency benefits.

Consensus is a fundamental problem in distributed computing. In this poster, we ask the following question: can we partially offload the execution of a consensus algorithm to the network to improve its performance? We argue for an affirmative answer by proposing a network-assisted implementation of the Raft consensus algorithm. Our approach reduces consensus latency, is failure-aware, and does not sacrifice correctness or scalability. In order to enable Raft-aware forwarding and quick response, we use P4-based programmable switches and offload partial Raft functionality to the switch. We demonstrate the efficacy of our approach and performance improvements it offers via a prototype implementation.

The current and the new generation of Internet of Things (IoT) devices present several challenges, among them the software update of legacy and new devices using wireless connections. In this paper, we study a problem of scheduling massive Firmware-Over-The-Air updates for millions of connected cars. We model this problem as a new generic problem called Time- and Machine-Dependent Scheduling Problem (TMDSP) that resembles project scheduling problems with variable-intensity activities. In the TMDSP, jobs, machine utilization, and production rates vary over time. In each period, a given job can be executed by a subset of machines with different capacities and production rates. The objective is to deploy a feasible schedule with minimum total completion time. We explore solving the problem using several approaches among than Biased Random-Key Genetic Algorithm (BRKGA), Iterated Local Search (ILS), Simulating Annealing (SA), two variations of Tabu Search (TB), and traditional genetic algorithms (TGA), in addition to a Mixed Integer Programming (MIP) model. We test the proposed approaches on a synthetic benchmark and very large real instances in IoT space. While using a commercial solver and the MIP model, we are able to solve only a few real instances (with, at most, 1,976 cars and 9,116 sectors); on the other hand, BRKGA presents significantly better results when compared to ILS, MIP solver and a simple multi-start heuristic.

Modern Web-enabled multimedia conferencing systems relays the source video flows to all call participants through media conferencing servers. This solution works well for small conference groups. However, as the conference size increases, the outbound bandwidth of the conferencing server becomes the bottleneck, which limits the scalability of the system and eventually degrades conference QoS. In this paper, we re-architect the multimedia conferencing service using Software-Defined Networks (SDN) that provides enhanced scalability and service orchestration. The main idea of our approach is to decouple the data plane and the control plane of the conferencing service. We advertise the group-based traffic forwarding capability of commodity OpenFlow switches as OpenFlow Selective Forwarding as a Service (OFSFaaS). The conferencing server now only handles control traffic, and dynamically establish media channels using OpenFlow switches through OFSFaaS. Preliminary prototype evaluations show that OFSFaaS eliminates the outbound bottleneck at conferencing servers and scales efficiently. OFSFaaS can be subscribed to by a wide range of multimedia services.

Compared with using only a single wireless path such as WiFi, leveraging multipath (e.g., WiFi and cellular) can dramatically improve users' quality of experience (QoE) for mobile video streaming. However, Multipath TCP (MPTCP), the de-facto multipath solution, lacks the support to prioritize one path over another. When applied to video streaming, it may cause undesired network usage such as substantial over-utilization of the metered cellular link. In this paper, we propose MP-DASH, a multipath framework for video streaming with the awareness of network interface preferences from users. The basic idea behind MP-DASH is to strategically schedule video chunks' delivery and thus satisfy user preferences. MP-DASH can work with a wide range of off-the-shelf video rate adaptation algorithms with very small changes. Our extensive field studies at 33 locations in three U.S. states suggest that MP-DASH is very effective: it can reduce cellular usage by up to 99% and radio energy consumption by up to 85% with negligible degradation of QoE, compared with off-the-shelf MPTCP.

AT&T is embarking on an exciting journey to revolutionize its network by transforming itself into a software company running the largest and most intelligent programmable cloud on the planet. Indeed, the network of Domain 2.0 (D2) will be intelligent software systems and applications operating on general-purpose commodity hardware [1]. This transformation will not only drive down CAPEX, OPEX and help to configure our network with less human intervention, but it will also create significant opportunities to scale and monetize existing and new intelligent services. This transformation will enable AT&T's D2 to establish a new services ecosystem equivalent in concept to the application ecosystem adopted by the Apple iOS and Android. D2 will facilitate mass marketing existing and new services, and lower the barrier to entry for enterprise and small business customers to create new innovative services.

As an important component of the virtual reality (VR) technology, 360-degree videos provide users with panoramic view and allow them to freely control their viewing direction during video playback. Usually, a player displays only the visible portion of a 360 video. Thus, fetching the entire raw video frame wastes bandwidth. In this paper, we consider the problem of optimizing 360 video delivery over cellular networks. We first conduct a measurement study on commercial 360 video platforms. We then propose a cellular-friendly streaming scheme that delivers only 360 videos' visible portion based on head movement prediction. Using viewing data collected from real users, we demonstrate the feasibility of our approach, which can reduce bandwidth consumption by up to 80% based on a trace-driven simulation.

IoT drives the future of Connected Cars including smart cars and it will transform the way we interact with our vehicles. With the emergence of millions of connected cars in the horizon, the wireless infrastructure needed to support this capability has to scale efficiently. To better understand the impact on the resource utilization of the wireless core infrastructure, we provide a detailed statistical model of the control plane/signaling interactions in connected cars. Specifically, our model is based on a 40K sample data set spanning more than 2100 IoT vehicles collected over 20 hours from a national telecommunications provider. The control plane model quantifies the additional load that the infrastructure (e.g., MME) needs to handle compared to an average busy hour LTE traffic model. We identify the heavy hitters of the control plane events and run real experiments based on our models in a testbed to characterize the resource utilization for supporting total event loadings using a real world high performance virtualized MME. No personally identifiable information (PII) was gathered or used in conducting this study. To the extent any data was analyzed, it was anonymous and/or aggregated data.

A primary design decision in HTTP/2, the successor of HTTP/1.1, is object multiplexing. While multiplexing improves web performance in many scenarios, it still has several drawbacks due to complex cross-layer interactions. In this paper, we propose a novel multiplexing architecture called TM3 that overcomes many of these limitations. TM3 strategically leverages multiple concurrent multiplexing pipes in a transparent manner, and eliminates various types of head-of-line blocking that can severely impact user experience. TM3 works beyond HTTP over TCP and applies to a wide range of application and transport protocols. Extensive evaluations on LTE and wired networks show that TM3 substantially improves performance e.g., reduces web page load time by an average of 24% compared to SPDY, which is the basis for HTTP/2. For lossy links and concurrent transfers, the improvements are more pronounced: compared to SPDY, TM3 achieves up to 42% of average PLT reduction under losses and up to 90% if concurrent transfers exist.

The service provider edge is responsible for connecting customers using standard protocols such as IP and BGP to the service providers internal network while enforcing service specific policies and service guarantees. Today this function is performed by the Provider Edge Router (PE). The specialized nature of the PE, however, restricts operational flexibility and their monolithic design impacts reliability. In this paper, we propose a new edge router architecture following SDN and NFV principals called EdgePlex. EdgePlex is a distributed system where the functions performed by a traditional edge router are decomposed and assigned to different elements in the system. A key aspect of our design is the use of a sandboxed environment (through the use of virtual machines) per customer. This gives EdgePlex the ability to isolate customers from one-another and independently move customers within and across EdgePlex platforms. We describe the architecture and a prototype implementation of EdgePlex. We perform detailed experiments using this prototype and show that EdgePlex is able to saturate the server in terms of throughput while having acceptable latency and jitter overheads. Our results lead us to believe that the EdgePlex design not only addresses the limitations of existing routers, but is also viable and can meet performance demands of production networks.

When traffic arrives from the network for an idled mobile device, the network executes device activation procedures to wake the device up. Current device activation mechanisms are ill suited to support the expected growth of machine-to-machine (M2M) devices and traffic. We propose an adaptive device activation architecture for LTE/EPC cellular networks that adapts to network conditions and M2M application requirements to realize scalable device activation without increasing the resources used for this purpose. Our evaluation shows that our adaptive approach enables the network to handle M2M applications with a large number of devices without negatively impacting existing human-to-human (H2H) and human-to-machine (H2M) traffic.

Existing video streaming algorithms use various estimation approaches to infer the inherently variable bandwidth in cellular networks, which often leads to reduced quality of experience (QoE). We ask the question: "If accurate bandwidth prediction were possible in a cellular network, how much can we improve video QoE?". Assuming we know the bandwidth for the entire video session, we show that existing streaming algorithms only achieve between 69%-86% of optimal quality. Since such knowledge may be impractical, we study algorithms that know the available bandwidth for a few seconds into the future. We observe that prediction alone is not sufficient and can in fact lead to degraded QoE. However, when combined with rate stabilization functions, prediction outperforms existing algorithms and reduces the gap with optimal to 4%. Our results lead us to believe that cellular operators and content providers can tremendously improve video QoE by predicting available bandwidth and sharing it through APIs.

Today's web page download process is ill suited to cellular networks resulting in high page load times and radio energy usage. While there have been notable prior attempts at tackling the challenge with assistance from proxies (cloud), achieving a responsive and energy efficient browsing experience remains an elusive goal. In this paper, we make a fresh attempt at addressing the challenge by proposing PARCEL. PARCEL splits functionality between the mobile device and the proxy based on their strengths, and in a manner distinct from both traditional browsers and existing cloud-heavy approaches. We conduct extensive evaluations over an operational LTE network using a prototype implementation of PARCEL. Our results show that PARCEL reduces page load times by 49.6%, and radio energy consumption by 65% compared to traditional mobile web browsers. Further, our results show PARCEL continues to perform well under client interactions, owing to its judicious functionality split.

The popularity of smartphones and smartphone applications means that data is the dominant traffic type in current mobile networks. In this paper we present our work on a systematic investigation into facets of the LTE/EPC architecture that impact the performance of TCP as the predominant transport layer protocol used by applications on mobile networks. We found that (1) load increase in a cell causes dramatic bandwidth reduction on UEs and significantly degrades TCP performance, (2) seamless handover causes significant TCP losses while lossless handover increases TCP segments' delay.

Distributed clouds offer a choice of data center locations to application providers to host their applications. In this paper we consider distributed clouds that host virtual desktops(VDs) which are then accessed by their users through remote desktop protocols. VDs have different sensitivities to latency, primarily determined by the types of applications running (games or video players are more sensitive to latency) and the end users' locations. We design VMShadow, a system to automatically optimize the location and performance of latency-sensitive VDs in the cloud. VMShadow performs black-box fingerprinting of a VM's network traffic to infer its latency-sensitivity and employs a greedy heuristic based algorithm to move highly latency-sensitive VMs to cloud sites that are closer to their end users. VMShadow employs WAN-based live migration and a new network connection migration protocol to ensure that the VM migration and subsequent changes to the VM's network address are transparent to end-users. We implement a prototype of VMShadow in a nested hypervisor and demonstrate its effectiveness for optimizing the performance of VM-based desktops in the cloud. Our experiments on a private and the public EC2 cloud show that VMShadow is able to discriminate between latency-sensitive and insensitive desktop applications and judiciously move only those VMs that will benefit the most. For desktop VMs with video activity, VMShadow improves VNC's refresh rate by 90%. Further our connection migration proxy, which utilizes dynamic rewriting of packet headers, imposes a rewriting overhead of only 13μs per packet. Trans-continental VM migrations take about 4 minutes.

In recent years, there has been growing interest in both industry and academia in augmenting mobile web browsing with support from the cloud [4, 1, 3, 16, 18]). These efforts are motivated by the goals of lowering costs of data transfer, and reducing web latencies and device energy consumption. While these efforts have adopted different approaches to cloud-based browsing, there isn't a systematic understanding of the rich design space due to the proprietary nature of many of the solutions. In this paper, we take a step towards obtaining a better understanding by evaluating an extreme point in the design space that involves cloud support for most browsing functionality including execution of JavaScript (JS), and for compaction of data (e.g., image transcoding and compression). Our study is conducted in the context of Cloud Browser (CB), a popular commercially available browser that embodies this design point. Our results indicate that CB does not provide clear benefits over Direct (a device-based browser) either in energy or download time. For e.g. while CB decreases the download time compared to Direct for 38.87% of pages, it increases it by as much as 29.8s for other pages. Similarly while CB decreases the total energy by up to 20.77J compared to Direct for 52.7% of the pages, it increases it by up to 21.31J for other pages. Interestingly, even though CB does JS execution in the cloud, it increases the CPU and network energy for close to 50% of the pages. Overall our study indicates that cloud-based browsing is not always a win, and there are important trade-offs that must be carefully considered when moving functionality to the cloud.

Despite its widespread adoption and popularity, the Hypertext Transfer Protocol (HTTP) suffers from fundamental performance limitations. SPDY, a recently proposed alternative to HTTP, tries to address many of the limitations of HTTP (e.g., multiple connections, setup latency). With cellular networks fast becoming the communication channel of choice, we perform a detailed measurement study to understand the benefits of using SPDY over cellular networks. Through careful measurements conducted over four months, we provide a detailed analysis of the performance of HTTP and SPDY, how they interact with the various layers, and their implications on web design. Our results show that unlike in wired and 802.11 networks, SPDY does not clearly outperform HTTP over cellular networks. We identify, as the underlying cause, a lack of harmony between how TCP and cellular networks interact. In particular, the performance of most TCP implementations is impacted by their implicit assumption that the network round-trip latency does not change after an idle period, which is typically not the case in cellular networks. This causes spurious retransmissions and degraded throughput for both HTTP and SPDY. We conclude that a viable solution has to account for these unique cross-layer dependencies to achieve improved performance over cellular networks.

We propose Joint-Family, a protocol that combines peer-to-peer (P2P) and adaptive bitrate (ABR) streaming for video-on-demand (VoD). While P2P for VoD and ABR have been proposed previously, they have not been studied together because they attempt to tackle problems with seemingly orthogonal goals. We motivate our approach through analysis that overcomes a misconception resulting from prior analytical work, and show that the popularity of a P2P swarm and seed staying time has a significant bearing on the achievable per-receiver download rate. Specifically, our analysis shows that popularity affects swarm efficiency when seeds stay “long enough”. We also show that ABR in a P2P setting helps viewers achieve higher playback rates and/or fewer interruptions. We develop the Joint-Family protocol based on the observations from our analysis. Peers in Joint-Family simultaneously participate in multiple swarms to exchange chunks of different bitrates. We adopt chunk, bitrate, and peer selection policies that minimize occurrence of interruptions while delivering high quality video and improving the efficiency of the system. Using traces from a large-scale commercial VoD service, we compare Joint-Family with existing approaches for P2P VoD and show that viewers in Joint-Family enjoy higher playback rates with minimal interruption, irrespective of video popularity.

We present VMShadow, a system that automatically optimizes the location and performance of applications based on their dynamic workloads. We prototype VMShadow and demonstrate its efficacy using VM-based desktops in the cloud as an example application. Our experiments on a private cloud as well as the EC2 cloud, using a nested hypervisor, show that VMShadow is able to discriminate between location-sensitive and location-insensitive desktop VMs and judiciously moves only those that will benefit the most from the migration. For example, VMShadow performs transcontinental VM migrations in ~ 4 mins and can improve VNC's video refresh rate by up to 90%.

Peer-to-Peer (P2P) systems have evolved from being used for file sharing to delivering streaming video on demand (VoD). The policies adopted in P2P VoD, however, have not taken user viewing behavior - that users abandon videos - into account. We show that abandonment can result in increased interruptions and wasted resources. As a result, we reconsider the set of policies to use in the presence of abandonment. Our goal is to balance the conflicting needs of delivering videos without interruptions while minimizing wastage. We find that an Earliest-First chunk selection policy in conjunction with the Earliest-Deadline peer selection policy allows us to achieve high download rates. We take advantage of abandonment by converting peers to “partial seeds”; this increases capacity. We minimize wastage by using a playback lookahead window. We use analysis and simulation experiments using real-world traces to show the effectiveness of our approach.

We empirically study the exponential potential function (EPF) approach to linear programming (LP), as applied to optimizing content placement in a video-on-demand (VoD) system. Even instances of modest size (e.g., 50 servers and 20k videos) stretch the capabilities of LP solvers such as CPLEX. These are packing LPs with block-diagonal structure, where the blocks are fractional uncapacitated facility location (UFL) problems. Our implementation of the EPF framework allows us to solve large instances to 1% accuracy 2000x faster than CPLEX, and scale to instances much larger than CPLEX can handle on our hardware. Starting from the packing LP code described by Bienstock [4], we add many innovations. Our most interesting one uses priority sampling to shortcut lower bound computations, leveraging fast block heuristics to magnify these benefits. Other impactful changes include smoothing the duals to obtain effective Lagrangian lower bounds, shuffling the blocks after every round-robin pass, and better ways of searching for OPT and adjusting a critical scale parameter. By documenting these innovations and their practical impact on our testbed of synthetic VoD instances designed to mimic the proprietary instances that motivated this work, we aim to give a head-start to researchers wishing to apply the EPF framework in other practical domains.

We propose a novel, scalable video on demand delivery approach that attempts to take advantage of multicast in hierarchical distribution networks. Using a typical IPTV network as an example, we assume that each level of the network has the ability to store videos and attempt to pre-populate different portions of videos at each of these levels. Pre-population of portions of videos, enable the requests from one level to be batched in next higher level and thus enables the use of multicast as the delivery mechanism. We formulate the problem of identifying what portions of videos to pre-populate and where, as an optimization problem and propose two approaches, DWaFiL and WaFiL. While both these approaches use the predicted demand for videos to determine the placement solution, DWaFiL also takes users' session duration into account. We overcome errors in demand predictions and short term popularity fluctuations using a small LRU cache in each location. Using traces from large scale commercial VoD deployment, we show that our approach is able to aggregate requests at each of the levels, reduce load on individual nodes, and outperform popular alternatives such as caching and replication in terms of scalability and server bandwidth. For example, our proposed scheme achieves up to ~27% aggregation in a single level and reduces aggregated bytes transfer from all nodes by ~25%. Finally, we analyse the dependency between different parameters and the performance of our scheme by experimenting with different parameter values.

Virtualized cloud-based services can take advantage of statistical multiplexing across applications to yield significant cost savings. However, achieving similar savings with real-time services can be a challenge. In this paper, we seek to lower a provider's costs for real-time IPTV services through a virtualized IPTV architecture and through intelligent time-shifting of selected services. Using Live TV and Video-on-Demand (VoD) as examples, we show that we can take advantage of the different deadlines associated with each service to effectively multiplex these services. We provide a generalized framework for computing the amount of resources needed to support multiple services, without missing the deadline for any service. We construct the problem as an optimization formulation that uses a generic cost function. We consider multiple forms for the cost function (e.g., maximum, convex and concave functions) reflecting the cost of providing the service. The solution to this formulation gives the number of servers needed at different time instants to support these services. We implement a simple mechanism for time-shifting scheduled jobs in a simulator and study the reduction in server load using real traces from an operational IPTV network. Our results show that we are able to reduce the load by ∼ 24% (compared to a possible ∼ 31%). We also show that there are interesting open problems in designing mechanisms that allow time-shifting of load in such environments.

We investigate how consumers view content using Video on Demand (VoD) in the context of an IP-based video distribution environment. Users today can use interactive stream control functions such as skip, replay, fast-forward, pause, and rewind to control their viewing. The use of these functions can place additional demands on the distribution infrastructure (servers, network, and set top boxes) and can be challenging to manage with a large subscriber base. A model of user interaction provides insight into the impact of stream control on server and bandwidth requirements, client responsiveness, etc.We capture the activity users in a natural setting, viewing video at home. We first develop a model for the arrival process of requests for content. We then develop two stream control models that accurately capture user interaction. We show that stream control events can be characterized by a finite state machine and a sojourn time model, parametrized for major periods of usage (weekend and weekday). Our semi-Markov (SM) model for the sojourn time in each stream control state uses a novel technique based on a polynomial fit to the logarithm of the Inverse CDF. A second constrained model(CM) uses a stick-breaking approach familiar in machine learning to model the individual state sojourn time distributions. The SM model seeks to preserve the sojourn time distribution for each state while the CM model puts a greater emphasis on preserving the overall session duration distribution. Using traces across a period of 2 years from a large-scale operational IPTV environment, we validate the proposed model and show that we are able to faithfully predict the workload presented to a video server. We also provide a synthetic trace developed from the model enabling researchers to also study other problems of interest. We also use the techniques to model consumer viewing of video content recorded on their personal Digital Video Recorder (DVR).

As content providers adopt peer-to-peer approaches for content sharing and distribution, they face new challenges in guaranteeing privacy to their clients. Participating peers can glean information from their communication with other peers, such as their identities or the shared data and use this information for malicious purposes. We present Nemor, a protocol that allows a requesting peer and a corresponding serving peer to communicate anonymously with each other and from other participating peers, while protecting the identity of the content being exchanged. Nemor relies on a trusted intermediary, such as a provider-managed tracker, to identify a potential serving peer. A peer in Nemor joins one or more trees. Using a combination of a random walk, a probabilistic jump from one tree to another and constrained flooding, the requesting and serving peer dynamically construct an overlay path between them. A key differentiator of Nemor is the integrated design of a congestion avoidance mechanism that yields significant performance benefits without compromising on anonymity. Using experimental results from PlanetLab and simulations with traces from an operational VoD system, we demonstrate that Nemor outperforms state of the art approaches like TOR and OneSwarm. Our results confirm that Nemor, while being resilient to attacks on anonymity, achieves high performance and scalability and is suitable for a range of applications, including distribution of large volume content, such as streaming video.

Cloud computing is a new infrastructure environment that delivers on the promise of supporting on-demand services in a flexible manner by scheduling bandwidth, storage and compute resources on the fly. IPTV services like Video On Demand (VoD) and Live broadcast TV requires substantial bandwidth and compute resources to meet the real time requirements and to handle the very bursty resource requirements for each of these services. To meet the needs of the bursts of requests, each with a deadline constraint for both VoD and LiveTV channel changes, we propose a resource provisioning framework that allows these services to co-exist on a common infrastructure by taking advantage of virtualization. We propose an optimal algorithm that provides the minimum number of servers needed to fulfill all requests for these services. We prove this optimality in a general setting for any number of services with general deadline constraints. By using real world data from an operational IPTV environment, our results show that anticipating and thereby enabling the delaying of VoD requests by up to 30 seconds gives significant resource savings even under conservative environmental assumptions. We also experiment with different scenarios (by varying the deadline constraints, changing the peak to average ratios of the constituent services) to compute the overall savings.

Service providers are evolving to provide more video content on-demand. Customers like to watch a variety of entertainment content of their choice and at their convenience. Serving this ever-increasing base of on-demand viewers requires careful provisioning by the providers to accommodate for both scale and interactivity. In this paper, we examine long-term usage patterns of hundreds of thousands of consumers of a nationwide IPTV service, and confirm that viewers are indeed migrating to what is called “time-shifted” viewing of television programming and movies using digital video recorders or on-demand viewing. We also examine the impact of such user's interactive control of their viewing experience using “stream control” functions (e.g., fast-forward, rewind, skip, replay, etc.) Through careful measurements on an IPTV server, we compute the load due to video streaming and handling these stream control events. We then extrapolate from these micro-benchmark measurements to predict the processing load imposed by users that would resort to using a “network-based” DVR capability if such a service were offered. We use both detailed trace-driven simulations and a simple operational-analysis based model to predict the capacity requirements of the server complex in a video-hub office to serve a large population of customers (e.g., a densely populated city like Mumbai). We provide insights on the number of requests serviced by the server, the average time to service these requests and the response time as perceived by the client.