Presentation is loading. Please wait.

Presentation is loading. Please wait.

HDF5 in SixTrack Regina Kwee-Hinzmann, Yngve I. Levinsen,

Similar presentations


Presentation on theme: "HDF5 in SixTrack Regina Kwee-Hinzmann, Yngve I. Levinsen,"— Presentation transcript:

1 HDF5 in SixTrack Regina Kwee-Hinzmann, Yngve I. Levinsen,
Kyrre N. Sjoebaek, Roderik Bruce CERN, 7/12/2016

2 Bugfixes 1st bug: data was not entirely correctly written out, incrementation was not done properly. in since svn r187 (or git 1412…) 2nd bug : orbit was not properly taken into account for some variables at collimator entrance. some minor fixes: closing, definition of variable type HDF5 in SixTrack R Kwee-Hinzmann

3 Validation tests no new validation tests since 2015 see run twin case to test the fixes, using identical random seed, 1 turn, 100 packs. compared all quantities of LPI file: all integer numbers were identical (pid, turn), all floats (s, x, y, xp, yp, dE/E type) were more precise, by at least 1 digit comparing to tracks2.dat run performance test: ~100 packs, 200 turns. factor 10 less disk space and 3 times faster (see later for details) HDF5 in SixTrack R Kwee-Hinzmann

4 Precision in tracks2.dat and tracks2.h5
test case: 1 turn, 100 packs, identical random seed compare output ascii dirrectly from SixTrack (custom writeout) hdf5 difference of each s-position in tracks2 files s-position HDF5 in SixTrack R Kwee-Hinzmann

5 Precision of dE/E For dE/E the mean is also more precise by 4 orders of magnitude. Some values went to the neighbouring bin. Probably again due precision. HDF5 in SixTrack R Kwee-Hinzmann

6 Precision of x, y, xp, yp when comparing the mean, hdf5 output is by at least an order more precise HDF5 in SixTrack R Kwee-Hinzmann

7 Performance Usual job with 100 Packs and 200 turns
have for ascii output a test sample of 1999 jobs (of each 50 packs) for hdf5 format 993 jobs (of each 100 packs). TRADITIONAL OUTPUT tracks2.dat min/max: 21.0/26.1 GB HS06 (benchmark quantity to measure CPU performance) min/max: 74k/216k seconds for total job, average:123.2k pool usage specification per job (50pcks): 50 GB HDF5 output tracks2.h5 min/max: 2.04/2.54 GB factor 10 better! HS06 min/max: 21k/73k seconds for total job average: 35k more than 3 times faster! pool per job (100pcks): not needed in test case. Never used any specification anymore (depends on your simulation). HDF5 in SixTrack R Kwee-Hinzmann

8 How to use it (status March 2015)
See delicate compilation instructions for latest hdf5 version at (it has to be configured such that SixTrack can use it) or just point to my hdf5 version in (default now for hdf5 flag) /afs/cern.ch/user/r/rkwee/public/hdf5/hdf The twiki also includes instructions where to get and how to use the adopted BeamLossPattern code (by Yngve). Status Dec 16: Alessio has now the code of Yngve’s beamLossPattern code to merge improvements made to the ascii version and make it read tracks2.h5. HDF5 in SixTrack R Kwee-Hinzmann


Download ppt "HDF5 in SixTrack Regina Kwee-Hinzmann, Yngve I. Levinsen,"

Similar presentations


Ads by Google