ARCHER Bulletin: 23 April 2014
In this issue:
- April ARCHER Technical Forum Webinar on Parallel IO
- Upcoming ARCHER Training courses
- Updates to default software modules
Note that all ARCHER training is free to all UK-based academic researchers regardless of their funding council affiliation.
ARCHER Technical Forum Webinar
The next in our monthly open web meetings will be at 3pm on Wednesday 30 April 2014. The meeting will consist of a short presentation followed by open discussion.
This month's presentation is from Charles Moulinec (STFC, Daresbury Laboratory):
Parallel IO in Code_Saturne
Code_Saturne is a multipurpose CFD software which has been ported on the fastest HPC machines. The solver has demonstrated very good speed-up for over 1.5 million processors of Argonne BG/Q, using a 13 billion cell mesh and on up to 72,000 cores of ARCHER, using a 0.9 billion cell mesh.
Ways around have been found to generate large meshes (on-the-fly mesh multiplication, for instance) and to postprocess data (co-processing). However, dumping the whole flow field for checkpoints on the disk seems unavoidable.
Two IO strategies exist in Code_Saturne, the former based on blocks of data collected from each process and the latter based on MPI-IO.
Comparison of IO performance observed on Lustre and GPFS systems will be presented, for different size of meshes.
Connecting to the meeting
You can join the meeting via the following link:
https://ca-sas.bbcollab.com/m.jnlp?sid=2011426&password=M.BF42A71BDAB6A68433FB5D77943F92
You will be asked to download the free Collaborate client after which you will be able to join the meeting. If you wish to test your connection in advance you can do so in the Collaborate Test Room at:
More details on the Technical Forum and on connecting to web meetings from mobile devices can be found on the ARCHER website at:
http://www.archer.ac.uk/community/techforum/
Upcoming ARCHER training
See: http://www.archer.ac.uk/training/
- Tools for Large-Scale Debugging and Profiling: 29 April - 1 May, Edinburgh
- Learn from the developers how to debug codes on ARCHER using DDT, and how to do parallel performance analysis with Scalasca and Vampir.
- Advanced OpenMP: 6-8 May, Oxford
- For those seeking to deepen their OpenMP understanding. Topics include the OpenMP memory model, performance tuning, OpenMP+MPI and OpenMP 4.0.
- Introduction to F95: 12-13 May, STFC Daresbury Laboratory
- An introduction to features of Fortran 90/95 relevant for numerical applications e.g. program control, subprograms, variables, I/O, arrays.
- Parallel Computing with R: 29-30 May, Swansea
- SPRINT provides easy access to HPC from the R language. This course introduces HPC followed by a detailed practical introduction to SPRINT.
- Programming the Xeon Phi: 4-5 June. Bristol
- Learn aboout vectorisation and parallel programming on the manycore Xeon Phi. Hands-on training from Intel, with case studies from EPCC.
- Hands-on Introduction to HPC: 30 June - 1 July, Edinburgh
- An introduction to HPC for the non-specialist. Learn about the key hardware and software building blocks and run real examples on ARCHER.
- Message-Passing Programming with MPI: 2-4 July, Edinburgh
- The Message-Passing Interace is the basic building block of almost all parallel programs. Learn MPI by writing codes in C or Fortran.
For more details or to register, see http://www.archer.ac.uk/training/.
Attendance is free for all academics.
Updates to default modules on ARCHER
The following modules will have their default version changed on Wednesday 30 April:
- vasp5 5.3.3 -> 5.3.5
