NAMD checkpoint mechanism.

From: Michael Harney (mjharney915_at_gmail.com)
Date: Fri Jul 09 2010 - 15:38:46 CDT

Hello all,

I am a masters student and research assistant. Currently the research
group I am in is collecting information about checkpoint mechanism for
our research on checkpointing and IO contention. We have people who
are willing to help us collect data, but we need to know how NAMD
actually does its checkpointing. I cannot find a detailed description
of *how* it performs its checkpointing. I have looked through the
code, and I am new to CHARM++ so I may be misreading it, but my
understanding is that all it does on a checkpoint is sync up the
processes using a barrier and stores the checkpoint data in Chare
objects that resides in memory of one of the processes (which process
seems unimportant). Is my interpitation correct? Does the mechanism
only store the checkpoint in memory, or are checkpoint files written
to disk? If anyone can provide me the information, or even better,
point me to a source that describes NAMD's checkpoint mechanism in
greater detail, that would be extremely helpful.

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 05:23:05 CST