joomla templates

linksceem.eu

HIGHLIGHTER:
Wednesday, 22 Nov 2017

Synchrotron Radiation Data Analysis

E-mail Print

Many discoveries in various fields like biology, chemistry, medicine, environmental science, geology, archaeology and physics are made by ‘seeing’ things using light. The range of the light used can be from infrared to X-rays and beyond. In recent decades, synchrotron light has become an essential tool on studying matter using radiation from infrared to hard X-rays.

The experiments on a synchrotron light source produce massive amounts of data of the order of petabytes, for example tomography beamlines. There is a need for online and offline data analysis of such large amounts of data, what requires high performance computing (HPC).

Lots of algorithms currently implemented on CPUs could profit from being ported to GPUs: the efforts made in synchrotron-related research encourage continuing the work in this promising field by porting typical X-ray data analysis programs to GPUs and exploring the possibility of enabling these programs to run efficiently on multicore CPUs.
The HPC experts at the ESRF chose CPU/GPU-based heterogeneous clusters for carrying out online and offline analysis. They selected a number of applications that need to be ported to such heterogeneous systems which are as follows:

  1. PyHST -  PyHST (High Speed Tomography in python version): is the main volume reconstruction program used for a large number of tomography experiments at the ESRF. After porting the program to GPUs a speedup factor of about 40 has been achieved.
  2. BigDFT (Density Functional Theory) – a density functional theory program has shown a speedup factor up to seven running on GPUs. It has also been selected as one of the benchmark applications for the PRACE project
  3. PPM (Percolated Perpendicular Media) – a program for simulating the reflectivity of magnetic multilayer structures. It achieved a speedup ratio of more than 100 on GPUs.
  4. hadow3 – a ray tracing program. A part of this application, i.e. FRESNEL2D has been chosen for its promising acceleration performance. A typical computation can take days if not ported to GPUs. However, a speedup factor of approximately 37 was obtained on GPUs.
  5. XTLS – a program for simulating X-rays through nanofocusing optical elements called transfocators. This is similar to Shadow3 and simulates wave propagation through Fresnel lenses in 1D. A speed-up in excess of 95 was achieved on GPUs.
  6. PyFAI (python tool for fast azimuthal integration) – a program that performs azimuthal integration on huge stacks of powder diffraction images. The GPU version of PyFAI keeps pace with even modern high-throughput detectors.

The ESRF scientific software team produced a number of international (conference/journal) papers out of this valuable contribution to synchrotron radiation community.

 

Pradel A., Langer M., Maisey J., Geffard-Kuriyama D., Cloetens P., Janvier P. and Tafforeau P., 2009. Skull and brain of a 300 million-year-old chimaeroid fish revealed by synchrotron holotomography. PNAS, 106(13) :5224-5228.

 

Typical Image taken from a Powder Diffraction experiment using a 4Mpixel detector (2048 x 2048 pixels).

A trace configuration for Shadow 3 visualized by the in-build Beamline visualization tool.