Skip to main content

Role of optical network for split computing between edge and cloud in support of ultra low latency services

Details are subject to change.

  • Organisers

    Reza Nejabati, University of Bristol, UK
    Andrew Lord, 
    BT, UK

  • Subcommittee

    SC10 – Architecture, Control & Management of Optical Networks

  • Day & Time

    18.09.2022, 09:00 – 12:30

  • Location

    Room Samarkand/Osaka

  • Description

    Mobile applications are evolving rapidly, requiring accurate and highly sophisticated computational methods such as machine learning (ML) techniques. Their processing power requirements cannot be supported in a mobile device with reasonable latency and energy consumption.

    Currently these applications are designed based on hosting all computation in high-end remote cloud servers. Queries generated from users’ mobile devices are sent to the cloud for processing. In this approach, large amounts of data (e.g., images, video and audio) are uploaded to the server via multiple networks, resulting in high latency and energy costs.

    To overcome these issues, emerging solutions are designed to host all the computation at the edge computing servers close to users. The drawbacks of this approach include limited computing capability at the edge, network delay and its variation (wireless network) and complexity to handle mobility of users from edge to edge.

    Another recent technological solution is Split Computing. In this approach, computing tasks are split and executed between the mobile devices, the edge compute resource and the cloud.

    Realizing split computing requires advanced techniques for breaking a computational task , e.g. Deep Neural Network (DNN ) into head and tails ends for execution in mobile, edge and cloud. It also requires close orchestration between application and a network in order to provide low latency connectivity between different parts of a computing tasks split across mobile, edge and cloud data center. This becomes even more challenging when users are mobile and in highly dense scenarios.

    This workshop aims to discuss challenges and possible solutions as well as opportunities for optical technologies for realizing next generation edge computing based on split computing.

    The workshop includes a series of position talks from industry and academia followed by a panel discussion:

  • Programme

    Session 1: Use case, application, algorithms and tools (90 min, 10 min per speaker + 30 min discussion)
    Federated and split computing at the edge, Mahesh Sooriyabandar, Toshiba Europe, UK
    6G boosted split-computing, Joan Pujol Roig, Samsung R&D, Korea
    Optical system optimization trade-offs in low latency (and low power) DCI and edge, Loukas Paraschis, IEEE Communication Society, USA
    Complexity, Accuracy and Delay Tradeoff in Split Computing for Distributed Computer Vision, Marco Levorato, UC Irvine, USA
    Intelligent Cloud & Edge dynamic orchestration of demanding 6G services, Xenofon Vasilakos, University of Bristol, UK
    Transport SDN & Orchestration in support of split computingRamon Casellas, CTTC, Spain
    Panel Discussion

    Coffee Break (30 min)

    Session 2: Technologies (90 min, 10 min per speaker + 30 min discussion)
    Low power backhaul networking in support of extreme edge computing, Andy Reid, BT, UK
    Centralize what you can, distribute what you must? – Strategies for distributing compute functions, Jörg-Peter Elbers, ADVA, Germany
    Time to transport and time to compute, could it be time to care about time?, Sebastian Bigo, Nokia Bell Labs, France
    Can new Photonics Technologies Transform the Landscape of Edge Computing and Split Computing? Where is the balance?, Ben Yoo, UC Davis, USA
    Scaling AI at the Edge – scenarios in Telco, Automotive and Industry 4.0, Laurent Schares , IBM, USA
    Combining Edge and Central Cloud Compute: An Enabler for 6G Services, Anna Tzanakaki, NKUA, Greece 
    Panel Discussion