I think there's a good chance we can resolve the "lock" correction over email, unless you guys want more discussion on this. On Fri, Mar 21, 2014 at 3:01 PM, Wesley Bland <[email protected]> wrote:

I think we started this process during the last meeting. It's still not gone, but it's something we're working on. We should try to schedule a time where (at least) the three of us can get together and go over another draft of the text.

Thanks, Wesley

On Fri, Mar 21, 2014 at 1:56 PM, Jim Dinan <[email protected]> wrote:

...

On Tue, Mar 4, 2014 at 4:11 PM, Aurélien Bouteiller <[email protected]

...

wrote:

...

p 597, mid page. What is "the lock"? Note that WIN_LOCK/UNLOCK is not a mutex - it has to do with the beginning and ending of a passive target epoch.

This is a good comment; talking about a "lock" is informal and useful for discussion, but not good standard-eze. What we mean here is that a process has reserved resources associated with an RMA access epoch, and it cannot release those resources because of a failure. The particulars of when resources are reserved, what they are, and how they are released are only very loosely defined by the MPI spec, so we should try to describe this using fairly abstract language.

~Jim.

_______________________________________________ mpiwg-ft mailing list [email protected] http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-ft

_______________________________________________ mpiwg-ft mailing list [email protected] http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-ft