Sites Inria

Version française


Bel Dumé – Technoscope - 21/12/2012

Testimonial from Katie Lewis, international user of SCOTCH

Katie Lewis

Katie Lewis of the Lawrence Livermore National Laboratory in the US uses SCOTCH in her everyday research on massively parallel 3D mesh/grid partitioning.

Katie discovered SCOTCH in 2009 after failing with all other software packages when it came to scaling different partitioning algorithms for peta- and exascale physics simulations. She has worked closely with François Pellegrini over the last three years on trivial questions, such as setting the correct compiler flags and obtaining useful output, as well as on much more complex problems like debugging out of memory issues in more than 8000 processors.

SCOTCH is ideal for scaling

“François Pellegrini has worked hard on parallel communication patterns that can scale to unprecedented numbers of processors, and I have helped things along by running the software on ever more processor counts, creating more parts and using larger problem sets than in previous experiments. Our collaboration has brought to light some areas for improvement, and François has been able to quickly fix any problems encountered along the way" explains Katie.

Katie tells us that she recently succeeded in using SCOTCH to partition a problem with more than three billion graph nodes into over one million parts. This is an order of magnitude greater than that possible with any other partitioning algorithm she has ever tried.

Katie says that she is unaware of any known problems that SCOTCH cannot handle and would now like to start pushing the limits of the software package by testing its static mapping algorithms. “I am not yet sure what challenges I will encounter in this new work,” she added.

Keywords: Massively parallel 3D mesh Grid partitioning Inria Bordeaux - Sud-Ouest Scotch software