Peng,
For pio-bench, edit the code and add MPI_MODE_CREATE to all
MPI_File_open calls. For example, if it currently has MPI_MODE_RDWR, then do
MPI_MODE_CREATE | MPI_MODE_RDWR.
For b_eff_io, I don't know. To help isolate the problem, just run it on one
machine and point the files to the local file system instead of PVFS. If it
still fails, you will need to do some debugging of the benchmark code or
contact the developers.
Rajeev
> -----Original Message-----
> From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Peng Gu
> Sent: Friday, May 12, 2006 9:28 AM
> To: [email protected]
> Cc: Rob Ross; Rajeev Thakur; Jun Wang; Robert Latham
> Subject: pio-bench and b_eff_io on PVFS2?
>
> Hi,
>
> I have been running pio-bench and b_eff_io on top of a single node
> pvfs2-1.4.0 (one node assume the role of both pvfs2-server and
> pvfs2-client).
>
> However, both benchmarks failed to run smoothly.
>
> 1) b_eff_io
> Specifically, for b_eff_io, it always hangs after 20-30 minutes of
> execution, at the same time I got a lot of repetitive errors like the
> follows:
>
> # mpiexec -np 2 ./b_eff_io -MB 128 -MT 1024 -p $MNT_PVFS2
> [0] ERROR in deleting before opening: n=1, p=0,
> filename=/pvfs2/mnt_pvfs2/b_eff_io_data_i00_001_0, ierr_open=-524286,
> ierr_close=202429708
> [0] ERROR in File_open: n=1, p=0,
> filename=/pvfs2/mnt_pvfs2/b_eff_io_data_i00_001_0, ierr=-524286
> [0] ERROR in File_set_view: n=1, p=0, ierr=204002572
> [0] ERROR in File_read/write: n=1, p=0, ierr=204526860
> [0] ERROR in File_set_view: n=1, p=1, ierr=205575436
> [0] ERROR in File_read/write: n=1, p=1, ierr=206099724
> [0] ERROR in File_set_view: n=1, p=2, ierr=209769740
>
> In addition, sometimes I found that my hard drive ran out of space
> when b_eff_io hangs (before running b_eff_io, the free space is more
> than 8GB).
>
>
> 2) pio-bench
> For pio-bench, things became more interesting:
>
> In pio-bench test suites, there are many access patterns to be
> tested, including "Nested Strided", "Sequential", "Random Stride",
> "Segmented", "Simple Strided", "Tiled IO", "Flash-io", and
> "Unstructured Mesh".
>
> Most of the times I got errors as follows:
>
> #mpiexec -np 2 ./pio-bench
> File under test: /pvfs2/mnt_pvfs2/l_fc_c_9.0.031.tar
> Number of Processes: 2
> Sync: off
> Averaging: Off
>
> Now testing: Simple Strided (re-read)
> hints total time(s) MB/s buffer size total bytes
> cannot create file without MPI_MODE_CREATE
> cannot create file without MPI_MODE_CREATE
> cannot create file without MPI_MODE_CREATE
> file pio-bench.c, line 865: Error opening file on process 0, aborting.
> repetition 0 of test 0
> file pio-bench.c, line 378: MPI_File_open: Undefined dynamic
> error code
> cannot create file without MPI_MODE_CREATE
> file pio-bench.c, line 865: Error opening file on process 1, aborting.
> repetition 0 of test 0
> file pio-bench.c, line 378: MPI_File_open: Undefined dynamic
> error code
>
>
> However, "Flash-io" test always finish without complains. The results
> is shown as follows.
>
> # mpiexec -np 2 ./pio-bench
> File under test: /pvfs2/mnt_pvfs2/l_fc_c_9.0.031.tar
> Number of Processes: 2
> Sync: off
> Averaging: Off
>
> Now testing: Flash-io (write)
> hints total time(s) MB/s buffer size total bytes
> 2 0.1966 15.2570 1572864 3145728
>
> Any thoughts or clue will be appreciated!
>
> Peng
>
>
_______________________________________________
Pvfs2-users mailing list
[email protected]
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-users
