Manycore Grand Challenges Workshop
Belfast, Thu 5 Apr 2018


After a discussion about historical grand challenges, we asked whether
we should have an application focus - for instance, the UK-based
Global Challenges Research Fund (GCRF) and Industrial Strategy
Challenge Fund (ISCF) are both application-centric.

With this approach, we try to solve real-world problems. Examples
might include:

* human rights
* self-driving cars, or adaptive traffic management
* modelling living processes, discrete event simulation
* robots and AI in real-time

Building such systems poses challenges to software engineers and
system architects. Some techniques, e.g. machine learning algorithms,
might require fundamental redesigns to avoid shared-memory
architectural assumptions.

The key to tackling such global challenges is the deployment issue -
getting buy-in from users and communities, perhaps realistic
deployment plans in developing countries for GCRF.

We noted that current EPSRC cross-ICT priorities are fuzzy and not
particularly technically focused. Systems research needs to be
perceived as relevant, but it doesn't naturally 'fit' into any current
priority.

One high-level challenge is PROGRAMMABILITY. This fits into EPSRC
priority "People at the heart of ICT". How do end-users add new
functions to devices? Possible technology includes end-user
programming, or example-driven programming, or programming IoT devices
by speech. One issue is that adaptive systems have evolving
requirements in typical user scenarios.


In general, there are several ways to map GCs - either start with
societal challenges and map down to manycore/systems domain. And/Or
start with systems and work up to society. Somehow, we need to connect
systems with society, and it is not clear how this should work.

The nice thing about the historical challenges (Longitude / Moonshot)
is the definite goal in mind. Success criteria were clear in
advance. Not so clear what the success criterion would be for a
'programmability' grand challenge.

We reflected on the contributions of the high-performance computing
(HPC) community to performance improvement. How much improvement is
due to Moore's Law, and how much to research innovation in HPC? We
contrasted HPC research efforts with the Apache Spark ecosystem, which
is massively popular despite its inefficiency and lack of a research
based approach. Apache infrastructure has excellent, user-friendly
tool support and a vibrant volunteer community.


Possible next steps

# ask EPSRC whether we can coordinate a workshop on grand challenges
  in systems

# manufacture a crisis - e.g. exascale- and then think what we would
  do to tackle this crisis.

# take the list of societal challenges, e.g. UN Sustainable
  Development Goals, and trace back to relate them to computing grand
  challenges.

# do something similar for the EPSRC Prosperity Outcomes {Healthy,
  Connected, Resilient, Productive} Nation - how do these relate to
  manycore?

# UKRI Industry-led efforts to identify challenges - follow their
  template, e.g. see docs at
https://www.gov.uk/government/news/industrial-strategy-challenge-fund-tell-us-what-to-support

# have a panel discussion at a future MaRIONet meeting