[Pvfs2-developers] Strange behavior with level2 (MPI-IO.c)

[Thomas Ludwig]

Hi all,

Robert Latham wrote:
> On Thu, Mar 08, 2007 at 05:50:52PM +0100, Julian Martin Kunkel wrote:
>> We see a rather strange and wrong behavior with PVFS2 using a file view with 
>> MPI-IO using different levels :)
> 
> You guys do a great job finding the most excellent corner cases.

:-) as Julian said: "We see". We now have our Jumpshot extension running
quite well and we are about to prepare a talk and a report about it.

Look at the attached screen dump. It shows four time lines for the
client processes on the top. They do open, set_view, write, barrier. The
lower 4x4 time lines are from the four I/O servers (view is expanded
such that you do not have overlapping events). The green events are jobs
( dspace_..., BMI_send, FLOW, BMI_send). Within the flow you see the
TROVE_WRITE_LIST event (brown). It gives you the data in the info lines
that we added to jumpshot. They get updated by mouse-over events. The
mouse was over the top-most trove event when I made the dump. It says
131.072 (bytes). Well, I was just curious whether everything adds up
nicely - and it didn't. :-)

> Hope you don't find any more bugs,

Not very likely. :-)

> but if you do, keep the reports
> coming.

Definitely. :-)

Details on the project will follow next week. I produced lot's of screen
dumps and are about to put together some nice slides explaining everything.

Bye,

	Thomas
-------------- next part --------------
A non-text attachment was scrubbed...
Name: trove.png
Type: image/png
Size: 66241 bytes
Desc: not available
Url : http://www.beowulf-underground.org/pipermail/pvfs2-developers/attachments/20070310/e46f1f94/trove-0001.png