[Pvfs2-developers] MPI-io tests

[Murali Vilayannur]

Hi Phil,
Yeah..
I dont know nor can I comment on the recommended aspects of it since
O_CREAT only means create if it does not exist. So we have to get it to
work..
We had to workaround for our paper results by having only 1 process do the
create.
This used to work before and Sam's hunch is that letting concurrent create
requests through the request-scheduler may have broken things..
Also, do not attempt to do simultaneous unlink() of the same file.. bad
things happen with that as well..:(
Thanks,
Murali

>
> > We just the MPI part for starting up a lot of processes.. Sorry for the
> > incorrect phrasing in my emails. We dont use
> > MPI I/O. Used the posix interface directly.
> >
> > Simultaneous create problems with vfs is possibly due to the request
> > scheduler on server not serializing crdirents of the same component name.
> > I doubt you will even see that kind of pattern at the server with the mpi
> > i/o interface.
>
> Is this a bug that would show up if you have several VFS clients attempt
> to open the same file in O_CREAT mode at the same time?
>
> I know that is not a recommended access pattern, but there are some
> applications out there that do that.
>
> -Phil
>
>