Training and Education

1
Training and Education at EPCC Judy
Hardy j.hardy_at_epcc.ed.ac.uk
2
EPCC Activities

• Founded in 1990
• Aim
• To accelerate the effective exploitation of
  Novel Computing in Industry, Academia, and
  Commerce
• Vital statistics
• 65 staff
• 2.7M turnover (almost) all from external sources
• research councils
• government agencies
• industry
• European Commission

3
Training and education

• HPC Training
• HPCx users
• European visitor programme
• UK academia
• commercial HPC users
• MSc in HPC
• one of first in world
• significant numbers of overseas students during
  first 4 years
• supported by industry
• excellent practical grounding for future careers

4
Training - history

• JISC funding
• Provided seed capital to write some of our
  courses
• HPC services
• Tailor courses to machines
• Course development continued
• Application based courses written with JTAP
  funding
• Visitor programmes
• Small training modules
• MSc in High Performance Computing
• Regular training courses throughout the year

5
Courses

• Original Courses
• Introduction to HPC
• Decomposing the Potentially Parallel
• MPI Programming
• HPF Programming
• Scientific Visualisation
• Computational Methods in Physical Science
• Introduction to Genetic Algorithms
• Performance Optimisation on Sun Workstations

• T3D/T3E Service Courses
• Introduction to the Cray T3D
• Introduction to the Cray T3E
• MPI Programming on the T3D/T3E
• HPF Programming on the T3D/T3E
• Performance Optimisation for the Cray T3D/T3E

• Application Based Courses
• Simulation for Engineers
• Mesh Generation Computer Course
• Mesh Decomposition
• Modelling in the Geophysical Sciences
• Computational Fluid Dynamics
• HPC in Business

• HPCx Courses (EPCC/CCLRC)
• Using the HPCx service
• The DL_POLY molecular dynamics package
• Optimisation techniques for the Power4
  processor
• Improved Performance Scaling on HPCx

• MSc Core Courses
• Fundamental Concepts of HPC
• Practical Software Development
• Shared Memory Programming
• Message Passing Programming
• Parallel Decomposition
• Core Topics in HPC and e-Science

• MSc Advanced Courses
• Exploiting the Computational Grid
• Applied Numerical Algorithms
• Hardware, Compilers and Performance
  Programming
• Object Oriented Programming for HPC
• Scientific Visualisation
• Advanced Topics in HPC and e-Science

6
Training facilities

• Dedicated training room for 26 students
• Sun Ray terminals plus Sun Blade 2000 server
• UoE HPC service is standard HPC resource
• Sun Fire E15K server
• 52 CPUs _at_ 900 MHz
• HPCx national service
• cluster of IBM SMP nodes
• 1600 POWER4 processors _at_ 1.7GHz

7
Training courses

• Annual training commitment for HPCx national
  service
• 30 training days
• 12 different courses
• 4 locations
• HPCx users also have reserved places on EPCC
  courses
• the same content as MSccourses but delivered
  differently

• HPCx Courses
• Using the HPCx service
• The DL_POLY molecular dynamics package
• Optimisation techniques for the Power4
  processor
• Improved Performance Scaling on HPCx

8
MSc in HPC - background

• MSc builds on EPCCs established training history
• Initially funded by 5-year Masters Training
  Package from EPSRC
• provides 6 studentships a year plus some
  development/admin costs
• First student intake in 2001, now in its fourth
  year
• longer-term support from the School of Physics
• Supported by the following companies
• Sun
• NEC
• Rolls Royce
• AWE
• UK Met Office
• and Portland Group, LAM MPI, Quadstone Ltd,
  Edinburgh Petroleum Services, PALLAS

9
Aims of MSc

• To teach practical skills
• not just theory
• In areas relevant to EPCCs HPC activities
• in academic research
• in industry

• Hidden aims
• to get to know potential EPCC employees
• have employed an MSc graduate every year of the
  programme to date
• to expand our training activities
• to do accredited teaching at a postgraduate level
• increasing activity in undergraduate teaching in
  School of Physics
• have taken a PhD student this year for the first
  time

10
Structure
Computational Science
Physical Sciences
Computer Science
Theory
Simulation
Pure
Applied
11
Format

• First three years
• Courses dual use MSc/training
• MSc students, HPCx users, UK academics,visitors,
  commercial attendees
• Each course taught as an intensive three-day
  block
• From this year
• Courses dedicated to MSc students
• Taught conventionally over 10 teaching weeks
• Each course taught in half-day blocks
• Six courses in each of two semesters
• This allows sharing with different MSc programmes
• Maths Operational Research MSc
• Informatics MSc (HPC Specialism)

12
Course design and delivery

• Courses designed in half-day blocks
• can be delivered in either model
• long and thin semsterised courses for MSc
  students
• 2 or 3 day intensive training courses
• Strong practical focus
• lectures
• supporting practicals
• tutorials/group discussions
• case studies etc

13
MSc courses

• MSc Core Courses
• Fundamental Concepts of HPC
• Practical Software Development
• Shared Memory Programming
• Message Passing Programming
• Parallel Decomposition
• Core Topics in HPC and e-Science

• MSc Advanced Courses
• Exploiting the Computational Grid
• Applied Numerical Algorithms
• Hardware, Compilers and Performance
  Programming
• Object Oriented Programming for HPC
• Scientific Visualisation
• Advanced Topics in HPC and e-Science

14
Student applications
Year 2001-02 2002-03 2003-04 2004-05 OS
18/0 30/3 40/5 48/5 EU 4/2
7/1 17/4 15/4 UK 10/4 23/4
31/5 43/4 Total 32/6 60/8
88/14 106/13
Student numbers (applications/enrolled)
15
Student destinations
First Destinations Students
(2002 - 2004) further study 25 UK
employment 35 overseas employment 10
overseas student returned 5 seeking
employment/study 0 unknown 25
16
Summary

• Training is one of EPCCskey activities
• HPC training
• supports HPC users throughout UK and Europe
• MSc in HPC
• practical training for range of careers in
  academia and industry

17
Overview

• HPC Training at EPCC
• training courses
• facilities
• MSc in High Performance Computing
• motivation
• taught courses
• student statistics

18
Programming Languages

• C or Fortran
• Shared memory programming (OpenMP)
• Message passing programming (MPI)
• Fortran
• Data parallel programming (HPF)
• Java
• Object-oriented programming (threads, MPJ, JOMP,
  RMI)

19
Core courses

• Fundamental Concepts of HPC
• hardware, performance, programming paradigms,
  decomposition techniques
• Practical Software Development
• tools and techniques of software engineering
  process
• source code management, software design, software
  projects
• Shared Memory Programming
• using OpenMP (C or Fortran)
• Message Passing Programming
• using MPI (C or Fortran)
• Parallel Decomposition
• decomposition techniques plus data parallel
  programming (HPF)
• Core Topics in HPC and e-Science
• Guest Lecture series, languages, tools (CVS,
  make, ant,)

20
Advanced courses

• Exploiting the Computational Grid
• Grid and web services, GT3 (4?), OGSA-DAI
• Applied Numerical Algorithms
• numerical methods, standard algorithms/libraries
  (LAPACK, FFTW)
• Hardware, Compilers and Performance Programming
• H/W architecture, processors, caches, memory,
  inteconnects
• compiler optimisation strategies
• Object Oriented Programming for HPC
• OO analysis and design, Java (threads/MPJ/JOMP/RMI
  )
• Scientific Visualisation
• AVS/Express, VTK
• Advanced Topics in HPC and e-Science
• Guest Lectures, advanced topics (posix threads,
  FPGAs, single-sided comms, condor, databases)
• No options this year but may include options from
  other programmes
• eg computer networks (?) from Informatics