A water cooled and high compaction ratio cluster of 48 CPUs, is being developed at CFN for heavy calculus, and tests have been conducted with the first 8-CPU unit, housed in a 6-U, 21"rack. The code is an earlier simplified current filament grid adjusting a function describing the current profile. As ISTTOK has been equipped with a new inner set of MHD coils, these analog integrated signals are fed as a direct input to a Levenberg-Marquardt algorithm testing various function profiles. CPU time-consuming has been greatly reduced and the cluster should achieve real-time results.
The running code is written in Matlab and a parallelisation based on time-slices is used. Synchronization using MPI over MatLab have proved to be the best method. From the analysis of a large ISTTOK shots database, different function approaches are discussed. |