Stef Salvini

1
The HPC/NA Roadmap Activity Looking into the
Future of Numerical Advanced Computing
Stef Salvini OeRC, University of
Oxford stef.salvini_at_oerc.ox.ac.uk
2
The Main Players

• Prof. A.E. Trefethen, University of Oxford
• Prof. N.J. Higham, University of Manchester
• Prof. I.S. Duff, Rutherford Appleton Laboratory
• Prof. P.V. Coveney, University College London
• Sponsored by EPSRC

3
Talk Contents

• Background
• The HPC/NA Roadmap
• The Future Implementing the Road Map

4
HPC/NA Webpage
http//www.oerc.ox.ac.uk/research/hpc-na /roadmap
5
1. Background
6
Vision for the HPC/NA Roadmap

• The Grand Challenge is to provide
• software
• Reusable in different fields and applications
• High quality
• High performance
• Long-term supported
• Libraries, modules etc
• Hopefully, self-adapting to parallel modes,
  hardware, etc
• a community that allows
• communication of interdisciplinary knowledge
• development of appropriate skills.

7
Aims of the HPC/NA Roadmapping Activity

• Survey a range of applications and users to
  understand
• The role and limits of a common algorithmic base
• How this is currently delivered and how should be
  in the future
• Current applications requirements and limitations
• Their future requirements
• Road-blocks that limit the applications future
  exploitation
• Knowledge gap between algorithmic developments
  and deployment
• Practical issues in the way of algorithmic
  content delivery
• Computing languages
• Vehicles for delivery
• etc

8
HPC/NA Roadmap Activities

• Community Consultation
• Workshop 1 Oxford, Nov 2008
• Applications focus
• Workshop 2 Manchester, Dec 2008
• Algorithms/NA focus
• Workshop 3 London, Jan 2009
• Review of Roadmap V1, further user industry
  perspectives
• Background work
• considering DOE/DARPA/NSF workshops
• Discussions with applications outside of
  workshops
• Good participation
• 80 active contributors
• Application scientists, Users, Numerical
  analysts, Computer scientists
• Physics, Biology, Chemistry, Engineeering,
  Finance, Oceanography, etc

9
2. HPC/NA Roadmap
10
Theme 1 Cultural Issues

• Identify potential community players across
• application domains
• numerical analysis
• computer science
• Develop community models to
• Share algorithms, software and ideas
• Provide activities
• workshops,
• Training
• virtual meeting spaces.
• Engage internationally

11
Theme 2 Applications and Algorithms

• Identify exemplar applications to
• baseline models for communication and
  benchmarking
• Develop map of algorithms across application
  domains
• Indentify impact of algorithms across discipline
  groups
• Take mapping of DWARFs or similar on capability
  computing
• Develop map of developments internationally
• Collect information about ongoing related
  activities
• Discuss with international funding agencies what
  plans are in place in this area

12
Common Future Algorithmic Requirements

• Optimisation
• Scalable FFT
• Adaptive mesh refinement
• Adaptive algorithms
• Eigenvalue/eigenvector (all or few)
• Iterative implicit solvers
• Out of core algorithms to enable larger problems

13
Major Future Applications Issues

• Load balancing
• Meshes
• particle dynamics and computation of interactions
• Better software environments for complex
  application development
• Adaptive software to automatically meet
  architectural needs
• Use of novel architectures (in the immediate
  future)
• FPGAs
• GPUs
• IBM Cell
• Other....
• Coupling between different models and codes
• Error propagation
• Scalable I/O
• Visualisation

14
Theme 3 Software Challenges

• Abstractions for more effective application
  development
• Computer scientists contribution essential
• Code generation and adaptive software systems
• automatic delivery of efficient code for complex
  architectures
• Guidance on best practice for software
  engineering development
• Frameworks and tools for application developers
• Better reuse of algorithms
• Better understanding of usability issues for
  complex software systems

15
Theme 4 Sustainability

• Address the sustainability of
• application codes
• software libraries
• skills
• Models for sustainable HPC software that might
  include
• Long term funding
• Industrial translation
• Open community support

16
Theme 5 Knowledge Base

• This theme is concerned with the general issue of
  sharing of knowledge and knowledge creation.
• The recommended actions are
• Develop mechanisms for collecting information on
  existing
• Software
• Expertise
• dissemination
• Develop mechanism for continuing community input
• Develop appropriate education and training
• MScs
• DTCs
• Short courses and summer schools.
• Engage industry (internships?) to ensure industry
  needs are also met.

17
3. Implementation
18
Implementing the Roadmap
19
Implementing the Roadmap

• Develop EPSRC Network Proposal
• Keep momentum going with existing group
• Outreach to other groups
• Coordination of HPC-NA activities
• Bring together numerical analysts, computer
  scientists application researchers/developers
• Currently planned
• Please contact Prof. A.E. Trefethen, OeRC,
  University of Oxford
• Study Groups with Industry (cf. Smith Institute)

20
Implementing the Roadmap APACE

• APplication Advanced Computing Exchange
• http//apace.myexperiment.org/
• Community site built on same lines as
  myExperiment
• Will facilitate collection and exchange of
  information on
• numerical analysis algorithms
• definition of applications in terms of algorithms
• Expertise in applications and algorithms
• Global activity in development etc
• Build community groups

21
APACE Webpage
http//apace.myexperiment.org/
22
Thank You!
stef.salvini_at_oerc.ox.ac.uk http//www.oerc.ox.ac.u
k http//www.oerc.ox.ac.uk/research/hpc-na