This all seems like a 6 of one half dozen of the other decision. Both
solutions suck because there are holes. So, it comes down to whether we
think the current code is worse than 1.2 or not. If they are the same
I'd be inclined to stay with what we have now for fear of inadvertantly
borking something else by rolling back the iof.
--td
Ralph Castain wrote:
I believe the changes all pretty much related to an attempt to fix the
iof_flush problem and correction of a different problem affecting the
reading of stdin. Unfortunately, the iof_flush problem still remains, albeit
perhaps in different form, and we now have a new problem in the stdin
behavior.
It's the old problem of hastily written spaghetti code to overly-ambitious
requirements, subsequently hacked by multiple people multiple times...so now
when you put your finger in one hole, a leak springs up somewhere else. :-)
Jeff and I are looking at this in more detail and will get back on it later.
The changes between 1.2 and 1.3 are not that big in terms of lines of code.
It just looks like they are suffering from the kid-and-the-dyke problem.
My motivation in proposing the rollback was simply that any attempt to
repair this new hole will quite likely open another one somewhere else. So
even if we can "fix" the duplicate stdin problem...did the kid really
improve the situation?
On 7/10/08 7:05 AM, "Terry Dontje" <terry.don...@sun.com> wrote:
I see that Jeff has updated the ticket saying that he is looking at the
code to see if he can generate a fix so the below may be superfluous.
Anyways, what were the issues fixed in 1.3? I really comes down to how
much more pain are we
giving our users by rolling back to 1.2 or not.
Note, I am assuming your comment of "most of our clusters do not..." is
referring to LANL's clusters. I do not believe this statement is
correct when you look at the OMPI community as a whole.
--td
Ralph Castain wrote:
I have been investigating Ticket #1135 - stdin is read twice if rank=0
shares the node with mpirun. Repairing this problem is going to be quite
difficult due to the rather terrible spaghetti code in the IOF, and the fact
that the IOF in the HNP actually rml.sends the IO to itself multiple times
as it cycles through the spaghetti.
Unfortunately, this problem -is- a regression from 1.2. Rather than spending
weeks trying to fix it, I see two approaches we could pursue. First, I could
repair the problem by essentially returning the IOF to its 1.2 state. This
will have to be done by hand as most of the differences are in function
calls to utilities that have changed due to the removal of the old NS
framework. However, there are a few places where the logic itself has been
modified - and the problem must stem from somewhere in there.
If I make this change, then we will be no better, and no worse, than 1.2.
Note that we currently advise people to read from a file instead of from
stdin to avoid other issues that were present in 1.2.
Alternatively, we could ship 1.3 as-is, and warn users (similar to 1.2) that
they should avoiding reading from stdin if there is any chance that rank=0
could be co-located with mpirun. Note that most of our clusters do not allow
such co-location - but it is permitted by default by OMPI.
We already plan to revisit the IOF at next week's technical meeting, with a
goal of redefining the IOF's API to a more reduced set that reflects a less
ambitious requirement. I expect to implement those changes fairly soon
thereafter, but that would be targeted to 1.4 - not 1.3.
Any thoughts on which way we should go?
Ralph
_______________________________________________
devel mailing list
de...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/devel
_______________________________________________
devel mailing list
de...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/devel
_______________________________________________
devel mailing list
de...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/devel
