Is the Web User Experience Improving?
Aruna Balasubramanian, Stony Brook University
Abstract: There is no question that browsers are critical to how users access Internet content today. The problem however is that we do not have a good understanding of whether browsing performance is improving. In fact, in this talk I argue that we dont even know how to measure Web performance.
There are a number of metrics used to objectively measure Web page load times, but these metrics do not measure user perception. This gets worse on mobile browsers, since users start to scroll the page even before the page is loaded, but most metrics only measure the latency of loading the first viewport. In the first part of the talk, I will discuss a new user-perception metric we define called uPLT. We conduct crowdsourcing studies to measure this uPLT metric both for desktop and mobile users. I will then describe our system WebGaze, whose goal is to explicitly improve the uPLT metric by leveraging eye gaze. Gaze tracking is commonly employed in the cognitive science literature to measure a user’s focused attention. We use eye gaze tracking to build a model of user attention and then prioritize bandwidth to objects that demand higher user attention.
In the second part of the talk I will switch gears to discuss performance. I will describe a tool we developed called WProf that lets us measure the main bottlenecks during the page load process. Using this tool, we find that, on mobile browsers, the performance bottleneck is the computational tasks. However, on desktop browsers, the performance bottleneck is the network. Our work further shows that the compute bottleneck on mobile browsers is exacerbated in lower-end mobile devices that are popular in the developing regions of the world. In effect, our study shows that web developers rely on the presence of high-end mobile devices to mask their decreased performance. But users who do not wish (or cannot afford) to constantly upgrade their devices are experiencing a slower web.
Biography: Aruna Balasubramanian is an Associate Professor at Stony Brook University. She received her Ph.D from the University of Massachusetts Amherst, where her dissertation won the UMass outstanding dissertation award and was the SIGCOMM dissertation award runner up. She works in the area of networked systems. Her current work consists of two threads: (1) significantly improving Quality of Experience of Internet applications, and (2) improving the usability, accessibility, and privacy of mobile systems. She is the recipient of the SIGMobile Rockstar award, a Ubicomp best paper award, a VMWare Early Career award, several Google research awards, and the Applied Networking Research Prize. She is passionate about improving the diversity in Computer Science and broadening participation. She leads the diversity committee at Stony Brook and is an active member of the N2Women group.
Redesigning Content Distribution Networks for Fun and Profit!
Theophilus Benson, Brown University
Abstract: Today, the internet addresses a broader array of applications (e.g., augmented reality or smart cars), traffic demands (e.g., live streaming), and infrastructure constraints (e.g., 5G) than ever. My work explores network state management and reconfiguration algorithms that quickly adapt to these growing requirements in a fine-grained and flexible manner.
In this talk, I will describe several results: First, I will present a novel global state layer that allows us to correlate events across the network and detect a novel set of problems. Second, I will illustrate how this global view allows us to improve performance by employing novel data-driven techniques. Finally, I will highlight the broader challenges for understanding the impact of data-driven web optimizations on client-side web performance.
Biography: Theophilus Benson is an Assistant Professor of Computer Science at Brown University. He earned his B.S. from Tufts, Ph.D. from U of Wisconsin – Madison, and post-doctorate from Princeton. Dr. Benson’s research focuses on improving the performance and availability of computer networks. His research was recognized by paper awards, including IMC, EuroSYS, ANRP. Dr. Benson received the NSF CAREER Award, NEC Faculty Award, Google Faculty Award, Facebook Faculty Award (X2), Faculty Research and Engagement Program (X2). Dr Benson was recently named to DARPA’s ISAT (Information Science and Technology) study group.
3DMs: Decentralised Data Delivery Markets
Alfonso de la Rocha, Protocal Labs
Abstract: Serving content globally at scale is a hard technical problem, as evidenced by the multiple decades of innovation and improvements in the Content Delivery Networks field. This challenge becomes even more interesting once we move away from centralized cloud infrastructure, which is typically managed and monitored by a single party, to a decentralized network that is permissionless, possibly anonymous, constantly changing (i.e. with high node churn) and lacking access to the convenience of a third party mediator system that facilitates the fair exchange of goods (e.g. credit providers). The benefits of moving to a decentralised setting are significant, however: cheaper storage and delivery, resilience against business failures (i.e., the network and all its components are not dependent on business decisions made by a single entity), independence from the personal data-driven business models, as well as a significantly lower barrier to entry for new players who want to contribute.
The aim of this talk is to introduce Decentralized Data Delivery Markets (3DMs) as a new field of research and innovation and one with rapidly increasing relevance. Decentralised storage networks, such as Filecoin, have reached unprecedented storage capacity commitments (of over 7EiB of cold storage in Jun 2021) and continue to grow rapidly. Filecoin is the first of its kind decentralised storage network, where any user can join and contribute storage capacity to the peer-to-peer network. There is an urgent need to complement decentralised storage with decentralised data delivery as these storage networks seek a solution to deliver the data stored in their network to end-users while meeting the expectations users have of the centralised services of today.
Biography: Before joining Protocol Labs, Alfonso worked as a blockchain expert at Telefónica R&D, where he was responsible for the design and development of core technology based on blockchains, distributed systems, and advanced cryptography. Alfonso’s involvement in research and development began at Universidad Politécnica de Madrid, where he worked on topics related to energy efficiency in data centers. His broad R&D experience also includes research into the compression efficiency of video coding standards at Ericsson Research and projects related to securing interdomain routing protocols at KTH Royal Institute of Technology in Stockholm.
Secure and Efficient Distributed Learning for Wireless Edge Networks
Diep N. Nguyen, University of Technology Sydney, Australia
Abstract: Unlike theoretical distributed learning (DL), DL over wireless edge networks faces the inherent dynamics/uncertainty of wireless connections and edge nodes, making DL less efficient or even inapplicable under the highly dynamic wireless edge networks (e.g., using mmW interfaces). This talk first discusses hurdles and users’ security-related risks of DL in wireless edge network. We then explore potential solutions by leveraging recent advances in coded computing, deep dueling neural network architecture, contract theory. By introducing coded structures/redundancy, a distributed learning task can be completed without waiting for straggling nodes. Unlike conventional coded computing that only optimizes the code structure, coded distributed learning over the wireless edge also requires to optimize the selection/scheduling of wireless edge nodes with heterogeneous connections, computing capability, and straggling effects. However, even neglecting the aforementioned dynamics/unce
rtainty, the resulting joint optimization of coding and scheduling to minimize the distributed learning time turns out to be NP-hard (the Knapsack problem). To tackle this and to account for the dynamics and uncertainty of wireless connections and edge nodes, we reformulate the problem as a Markov Decision Process and then design a novel deep reinforcement learning algorithm that employs the deep dueling neural network architecture to find the jointly optimal coding scheme and the best set of edge nodes for different learning tasks without explicit information about the wireless environment and edge nodes’ straggling parameters. Simulations show that the proposed framework reduces the average learning delay in wireless edge computing up to 66% compared with conventional DL approaches. The proposed framework here is also applicable to any distributed learning scheme with heterogeneous and uncertain computing nodes.
Biography: Diep N. Nguyen (Senior Member, IEEE) received the M.E. degree in electrical and computer engineering from the University of California at San Diego (UCSD) and the Ph.D. degree in electrical and computer engineering from The University of Arizona (UA). He is currently a Faculty Member of the Faculty of Engineering and Information Technology, University of Technology Sydney (UTS). Before joining the UTS, he was a DECRA Research Fellow at Macquarie University, and a member of Technical Staff at Broadcom, CA, USA, and ARCON Corporation, Boston, consulting the Federal Administration of Aviation on turning detection of UAVs and aircraft, U.S. Air Force Research Lab on antijamming. His current research interests include computer networking, wireless communications, and machine learning applications, with an emphasis on systems performance and security/privacy. He has received several awards from LG Electronics, UCSD, UA, U.S. National Science Foundation, and Australian Research Council. He is an Editor, Associate Editor of the IEEE TRANSACTIONS ON MOBILE COMPUTING, IEEE ACCESS, IEEE SENSORS JOURNAL, and IEEE OPEN JOURNAL OF THE COMMUNICATIONS SOCIETY (OJ-COMS).