Edikt mini-project

The Edikt project [4] funded two months FTE of effort, in the first quarter of 2011. The PETSc developers issued a patch for the bug in the interface to SUNDIALs, which meant that the micromag code could now call the CVODE library. The remaining functionality required for a parallel code, outlined in section 6.5, was implemented. Testing and benchmarking of up to 16 nodes (384 cores) of the XE6, phase 2b machine revealed a subtle bug in the construction of one of the matrices, and some poor scaling performance. Further work is required to fix the bug, and investigate the cause of the poor performance. The most likely cause of poor performance when using PETSc matrix data objects is the construction and population of the matrices. The bug and poor performance may not be unrelated.

Chris Maynard 2011-06-08