Description
Date depot: 24 mars 2021
Titre: Container Orchestration for the Edge Cloud
Directeur de thèse:
Olivier FOURMAUX (LIP6)
Encadrant :
Timur FRIEDMAN (LIP6)
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Non defini
Resumé: Objective
Cloud architectures typically are designed for on-demand access to computing and
storage resources located in centralized data centers. The ever-greater importance
of services such as content delivery and the Internet of Things (IoT) challenges this
infrastructure. Constraints such as a need for low latency, and benefits such as those
that arise when large data flows are not sent along backhaul links combine to create
an interest in deploying services on nodes that are close to the end user. Edge
computing is emerging as a distributed computing paradigm that promises to enjoy
considerable prominence, especially with the deployment of 5G networks, which
explicitly envisage this development. However, edge computing currently lacks
robust orchestration mechanisms to automate the management of the life cycles of
the applications that deliver services. There is no concrete, production-ready solution
such as those, like Kubernetes, that have existed for some years now in the
centralized datacenter. However, it is an active research and development field, with
ongoing work on edge-specific aspects of orchestration such as scheduling, task
offloading, and autonomy in case of non-connectivity.
This thesis will advance the state of the art with respect to orchestration for the edge
cloud. We will base our work on EdgeNet, a prototype orchestrator that is already
under development at LIP6, and that is in use in support of third-party applications
(scientific experiments and not-for-profit services). The existence of this tool will
allow us to test our innovations in a real-world environment. The key research
challenge will be to conceive of scheduling algorithms for the orchestrator that are
capable of efficiently finding near-optimal solutions to a constraint matrix made up of
edge requirements such as the desired geolocation of edge nodes, the quality of
edge node connectivity, and the resources available at the edge nodes.
Expected Results
We expect to be among the first in the scientific community to propose scheduling
algorithms that can be put into use for a new emerging class of edge orchestrators.
The algorithms will increase the efficiency of “multitenancy”, in which multiple
independent users deploy their services to edge devices that are owned by network
operators, as is envisaged with the roll-out of 5G. In the process, we expect to
develop software that can be contributed to the open-source code base that has
emerged around the Kubernetes orchestrator, which currently is the de facto industry
standard for data center orchestration. Our contributions will therefore be both
scientific and in the form of tools that we hope will gain wide usage.
International Impact
The developments around EdgeNet at LIP6 already enjoy considerable international
interest. The Silicon Valley company VMware has made charitable donations to the
Fondation Sorbonne Université in support of this work, and a leading research team
at the New York University Tandon School of Engineering is making use of EdgeNet
to deploy a cutting edge content distribution network based upon peer-to-peer
payments through cryptocurrency. This thesis will serve to deepen these contacts
around areas of mutual interest having to do with the edge cloud, with the potential
for joint work and, possibly, the generation of valuable intellectual property for
Sorbonne Université.
Doctorant.e: Senel Berat Can