Jeff Squyres wrote:
It's half done, actually. But it was still going to be an option,
not necessarily the only way to do it:
http://www.open-mpi.org/hg/hgwebdir.cgi/jsquyres/shm-sysv/
On Mar 30, 2009, at 1:40 PM, Tim Mattox wrote:
I've been lurking on this conversation, and I am again left with the
impression
that the underlying shared memory configuration based on sharing a file
is flawed. Why not use a System V shared memory segment without a
backing file as I described in ticket #1320?
In some performance experiments so far, it appears that the start-up
performance of the two approaches are rather different. I don't
understand the results, but I note here that they're different.
Touching all that shared memory upon startup (which, as I noted, we
essentially do now anyhow) adds noticeably to the start-up time. (Less
so now with the single-queue changes. Whether the time is tolerable or
not is a matter of taste, etc.) Going from mmap to SysV speeds that
time up a lot in the tests I've done. But then (mysteriously) the time
to exchange first messages takes much longer.
I'm baffled. For now, I just point out that the performance
characteristics could be different.
