Charm++ parallel programming system

[Phil Miller <mille121 AT illinois.edu>]

On Thu, Mar 17, 2016 at 10:18 AM, Brunner, Robert Kraemer <rbrunner AT illinois.edu> wrote:

What is the state of fault tolerance support on Cray XE systems (in particular, Blue Waters) with respect to allowing user code to catch node failures. Can the runtime notify the user program that a node has failed, and allow the user program to handle the failure, and  perhaps to keep running, taking the loss of the node and any associated objects into account?

At a higher level, Charm++'s approach to fault tolerance is to make the failure transparent to the user code. The application will never see missing nodes and objects, because the runtime system takes responsibility for recovering them consistently.