<font size=2 face="sans-serif">Really? A 4 year old BGP question!?
Do I remember it ... </font>
<br>
<br><font size=2 face="sans-serif">Paul Coffman is doing some ROMIO work.
Let's get him involved and see how this fits in to what he's doing
now on Q. It's good info for him anyway.</font>
<br>
<br><tt><font size=2>"Rob Latham" <robl@mcs.anl.gov> wrote
on 10/17/2013 10:32:40 AM:<br>
<br>
> From: "Rob Latham" <robl@mcs.anl.gov></font></tt>
<br><tt><font size=2>> To: mpich2-dev@mcs.anl.gov, </font></tt>
<br><tt><font size=2>> Cc: Bob Cernohous/Rochester/IBM@IBMUS</font></tt>
<br><tt><font size=2>> Date: 10/17/2013 10:33 AM</font></tt>
<br><tt><font size=2>> Subject: Re: [mpich2-dev] And yet another ROMIO
performance question</font></tt>
<br><tt><font size=2>> <br>
> On Wed, Sep 16, 2009 at 02:57:45PM -0500, Bob Cernohous wrote:<br>
> > Same customer noticed that the file system was doing much more
i/o than <br>
> > MPIIO on collective reads. <br>
> <br>
> Hi Bob. Did you ever get an acceptable answer to this 4 year
old<br>
> question? <br>
> <br>
> Have you ever had anyone respond to a 4 year old message? Let's
see<br>
> how well lotus notes handles /that/.<br>
> <br>
> Sorry it took me so long to give you an answer. By the time
I figured<br>
> out what was going on, I forgot you needed an answer. In my
meager<br>
> defense, my daughter was 6 weeks old back then.... <br>
> <br>
> What happens is two-phase implements data sieving. Normally,
this is<br>
> good. On Blue Gene, we turn on two-phase all the time (the hint
is<br>
> "enable", not "automatic"), even when accesses
are not overlapping.<br>
> <br>
> So, consider a file domain where the first and last bytes are written.<br>
> (e.g. a parallel-netcdf application is reading from a file with record<br>
> variables). ROMIO will issue one read for the entire file domain.
<br>
> <br>
> Writes are even worse: ROMIO will read in the entire file domain,<br>
> modify the regions, and write it out.<br>
> <br>
> Not sure the best way to fix this. Probably stick in a few peephole<br>
> optimizations, like if there's only one access in a file domain, just<br>
> go ahead and issue that one access. <br>
> <br>
> A "real" fix would have to build up a basic performance
model so we<br>
> could answer the question of where the tradeoff of "big request
with<br>
> extra wasted data" beats "many small requests, but no wasted
data".<br>
> <br>
> ==rob<br>
> <br>
> -- <br>
> Rob Latham<br>
> Mathematics and Computer Science Division<br>
> Argonne National Lab, IL USA<br>
> <br>
</font></tt>