[mpich-discuss] Is there only one ADIOI_xxx_OpenColl

[Ju JiaJia]

Thank for you reply. I found three collective open functions
 :Ad_opencoll.c (src\mpi\romio\adio\common):void ADIOI_GEN_OpenColl,
Ad_opencoll_failsafe.c (src\mpi\romio\adio\common):void
ADIOI_FAILSAFE_OpenColl and Ad_opencoll_scalable.c
(src\mpi\romio\adio\common):void ADIOI_SCALEABLE_OpenColl.
Thx.
On Tue, May 1, 2012 at 7:43 AM, Rajeev Thakur <thakur at mcs.anl.gov> wrote:

> See romio/adio/ad_nfs/ad_nfs_open.c and
> romio/adio/common/ad_opencoll_scalable.c . In general, do a recursive grep
> for Open in the adio directory and you will find all of them.
>
>
> On Apr 24, 2012, at 3:57 AM, Ju JiaJia wrote:
>
> > Hi all:
> >    I'm runing MPI app on two nodes with MPICH2. But without NFS, I copy
> the app file in which calls MPI_File_open to open a file in the current dir
> to the same location on all the nodes.
> > It failed in MPI_File_open when i run. I trace the code, It failed in
> MPI_File_open--->ADIO_Open----->ADIOI_OpenColl. ADIOI_OpenColl is a pointer
> to ADIOI_GEN_OpenColl.
> > in which one process opens to create the file, all others open, by
> turning off ADIO_CREATE. So i understand why my app failed.
> >
> > But in MPI_File_open, some comments like this:
> >     /* actual opens start here */
> >     /* generic open: one process opens to create the file, all others
> open */
> >     /* nfs open: everybody opens or else you'll end up with "file not
> found"
> >      * due to stupid nfs consistency semantics */
> >     /* scalable open: one process opens and broadcasts results to
> everyone */
> >     ADIOI_OpenColl(fd, rank, access_mode, error_code);
> >
> > I searched all the source code and found  ADIOI_GEN_OpenColl only.
> > So where is the other two open methods which mean nfs open and scalable
> open implemented?
> > Does it mean i can only run MPI app in which calls MPI_File_open when
> all nodes can access the unique copy of the app file?
> >
> > My Engllish is not good, i hope you can understand what i say.
> >
> > Any explanation will be appreciated.
> > _______________________________________________
> > mpich-discuss mailing list     mpich-discuss at mcs.anl.gov
> > To manage subscription options or unsubscribe:
> > https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss
>
> _______________________________________________
> mpich-discuss mailing list     mpich-discuss at mcs.anl.gov
> To manage subscription options or unsubscribe:
> https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss
>



-- 
How are you doing?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/mpich-discuss/attachments/20120501/426dec77/attachment.htm>