Edge Computation and the Future of Transport - RA and PhD student positions available
7 August 2020
/ edge-computation
Edge compute infrastructure is an essential part of modern networks,
but existing approaches to managing such infrastructure are complex
and do not scale. Thanks to generous support from Rakuten Mobile, we're
looking to appoint a Research Associate and two PhD students to work on
a new research project to develop highly decentralised models for edge
computation, leveraging modern network transport protocols and new
programming models to manage edge compute infrastructure at scale.
One post-doctoral Research Associate position and two PhD studentships
are available, working with
Jeremy Singer and I at the University of Glasgow, in cooperation
with Rakuten Mobile in Tokyo:
The funding provides a stipend to cover living expenses, and will pay
fees at the UK student rate. The University
excellence bursaries can cover international student fees.
Please contact me for further details or
apply online.
The project builds on lessons learned building the
FRμIT testbed. That project explored whether massive aggregation
of low-cost, low-power, commodity infrastructure could form an
efficient and effective compute fabric for distributed applications. A
key outcome of the FRμIT testbed was an understanding that existing
large-scale systems management and DevOps tools can be effective in
data centre environments, but are not well-suited to managing the edge
compute infrastructure.
One of the key reasons for this is network heterogeneity.
A data centre network can be made uniform but an edge network, and
particularly an edge network that includes devices and resources in
a home environment, or systems with mobile or intermittent network
connectivity, will be heterogeneous.
To support ubiquitous edge computation, a management system must work
with devices using different types of access link, whether fixed or
mobile, and irrespective of link bandwidth, latency, or reliability.
It must work across IPv4, IPv6, and non-IP networks, through NATs and
in the presence of firewalls, using a range of transport protocols,
and supporting edge devices that only sometimes have connectivity.
This means that the traditional ways of accessing devices to be managed
will fail. The management system must rather leverage the ideas of peer
to peer and delay tolerant networks, gossip protocols, and named data
networks, to ensure that compute resources can receive updated software
and data, and disseminate results to the wider network. And this must
be done in a manner that is compatible with the current Internet, and
that can be deployed over-the-top of existing networks.
Developing such a system will require innovation in transport protocols
and overlay networks to ensure data can be delivered in a timely manner
where necessary, and eventually where possible; and it will require new
programming models, frameworks, and APIs to allow developers to manage
and make effective of such a heterogeneous, distributed, edge compute
fabric.
If you have experience or interest in modern Internet transport
protocols, peer-to-peer or delay-tolerant networking, named data
networking, or languages and programming models for large-scale
distributed systems, then please get in touch – this promises
to be an interesting project, with a great industrial partner.