Efficient Processing of Multiple Contacts in MPP-DYNA
Complex models often contain more than just a few contact interfaces. The decomposition of such a
model can result in an uneven distribution of these contacts among the available processors. Some
contacts may lie wholly on a single processor, while others will be distributed across many or all of the
processors.
Some processors may have many contacts to handle, and others may have none. This variability can
cause inefficiencies which adversely impact scalability. I will show recent work on contact algorithms
in MPP-DYNA which addresses some of these issues.
Session24_Paper3.pdf
—
PDF document,
566Kb
