H5Part development and discussion

[Andreas Adelmann <andreas.adelmann AT psi.ch>]

Dear Colleagues this is a summary on discussions we had during my December visit at Berkley w.r.t. H5Part/Block/FED.

At the moment the LBL are very active in promoting, using and further developing H5Part and H5Block. 

The most significant activities are summarized in a) to c) and e).

a) performance improvements/tuning by Mark Howison. On December 19

he send some of his work (Updates to H5Part 1.4.0 from LBL) to the 

h5part AT lists.psi.ch.

b) together with Gunther Weber we could fix the Visit H5Block reader. I have a

MAC version of the new reader on my laptop. Now we can see particles and fields

together. We need two more H5Block function: 

H5BlockSetOrigin(H5PartFile *fileID,double xmin, double ymin, double zmin)

double * H5BlockGetOrigin(H5PartFile *fileID)

----> We should ASAP add the two features to H5Block. This requires additional small changes to Visit which will 

hopefully make their way into the next major release. 

----> We also should ASAP work in Mark's changes and MAKE A NEW RELEASE

c) Oliver Rübel, Prabhat and Gunther Weber  working on the FastBit implementation. There is a working prototype available and shows

three orders of magnitude improvement in speed on large data sets and complicated queries. The goal is to 

   implement FastBit support until June 2009 into H5Part/H5Block.     

   

d) There is a great interest and need for H5FED. We decided that we will revise LBL's H5FED contribution after the FastBit implementation is done. 

e) Janet Jacobsen will again (try to) promote/implement etc. H5Part/H5Block into IMPACTxx. She has our IMPACT-T implementation.

f) We agree to start writing a (peer reviewed) Paper on H5Part. I will take the lead on that.

g) Miscellaneous

1) A two day workshop on scalable parallel I/O with HDF featuring presentation from NERSC users, HPC consultants, and the HDF developers
will be held January 20-21 at NERSC's Oakland Scientific Computing Facility. You are invited to attend and encouraged to bring your
application, I/O kernels, and descriptions of how you're doing I/O at scale. The workshop includes a variety of discussion on how to get the
most out of parallel I/O using HDF. The workshop will conclude with a written synopsis of NERSC user priorities for the HDF developers and
NERSC consultants to address in future work.

More details and registration information is provided here: https://outreach.scidac.gov/hdf/

Here the LBL/H5Part team with LBL/Analytics (J. Shalf) will present our needs to the HDF5 folks. A PSI contribution (Achim!?) 

is welcome. 

2) Here an interesting paper to read: Automated analysis for detecting beams in laser wakefield simulations, Daniela M. Ushizima et.al (ICMLA08)

@inproceedings{Ushizima:ICMLA:2008, author = {Daniela Ushizima and Oliver R{\"u}bel and Prabhat and Gunther Weber and E. Wes Bethel and Cecilia Aragon and Cameron Geddes and Estelle Cormier-Michel and Bernd Hamann and Peter Messmer and Hans Hagen}, title = {{Automated Analysis for Detecting Beams in Laser Wakefield Simulations}}, booktitle = {2008 Seventh International Conference on Machine Learning and Applications, Proceedings of IEEE ICMLA'08}, year = {2008}, note = {LBNL-960E}, }

Attachment: Ushizima_ICMLA08.PDF
Description: Adobe PDF document

This is very interesting for the PSI-XFEL project and for beam dynamics in general.

Lets discuss this asap, best 

Andreas 

------
Dr. sc. math. Andreas (Andy) Adelmann
Staff Scientist
Paul Scherrer Institut WLGB/132 CH-5232 Villigen PSI
Phone Office: xx41 56 310 42 33 Fax: xx41 56 310 31 91
Phone Home: xx41 62 891 91 44
-------------------------------------------------------
Wednesday: ETH CAB F 10.1  xx41 44 632 82 76
============================================
The more exotic, the more abstract the knowledge, 

the more profound will be its consequences.
Leon Lederman 

============================================