Charm++ parallel programming system

[Tom Quinn <trq AT astro.washington.edu>]

I think the reason for no CmiFread() is simply that fread()s don't fail on 
Lustre nearly as often.  Furthermore, checkpoints are written about 10 
times more than they are read.  E.g. I'll checkpoint every hour for a 12 
hour queue run, and I'll only read one of them.

Tom Quinn       Astronomy, University of Washington
Internet:       
trq AT astro.washington.edu
Phone:          206-685-9009

On Thu, 27 Dec 2012, Phil Miller wrote:

> On Thu, Dec 27, 2012 at 1:33 PM, Phil Miller 
> <mille121 AT illinois.edu>
>  wrote:
> > On Thu, Dec 27, 2012 at 1:21 PM, Tom Quinn 
> > <trq AT astro.washington.edu>
> >  wrote:
> > > What with Lustre becoming more common, and its habit of failing with
> > > "EINTR", could the CmiFopen, CmiFwrite, and CmiFclose (and there probably
> > > needs to be a CmiFread), which handle this case, be made available to "C"
> > > code as well as C++?  I think all that needs to be done is add an 'extern
> > > "C"' in front of the declarations in converse.h and the definitions in
> > > util/pup_util.C.
> >
> > That all sounds reasonable to me. I'll have it pushed in a couple minutes.
>
> The exposure of the 3 existing functions to C code is done. I'm
> actually a bit curious to see why there's no matching read call, since
> that can certainly be interrupted. I'll follow up on this.