[MPICH] slow IOR when using fileview
Robert Latham
robl at mcs.anl.gov
Mon Jul 2 17:09:15 CDT 2007
On Mon, Jul 02, 2007 at 04:34:49PM -0500, Rajeev Thakur wrote:
> But I don't think the lock functions get called in the collective I/O case.
No, in this case the collective I/O case is falling back to the
independent methods (and thus calling the lock functions).
Let's take the collective write case. Assuming cb_write has not been
set to 'disabled', ROMIO will check for interleaving. In Wei-keng's
case, is there any interleaving? if not, interleave_count will be 0
ADIOI_GEN_WriteStridedColl will fall back to independent accesses.
For further confirmation you can run the test with testfs:
[3/8] ADIOI_TESTFS_SetInfo called on foo.view
[3/8] calling ADIOI_GEN_SetInfo
[3/8] ADIOI_TESTFS_Open called on foo.view
[3/8] ADIOI_TESTFS_SetInfo called on foo.view
[3/8] calling ADIOI_GEN_SetInfo
[3/8] ADIOI_TESTFS_WriteStridedColl called on foo.view
[3/8] calling ADIOI_GEN_WriteStridedColl
[3/8] ADIOI_TESTFS_WriteStrided called on foo.view
[3/8] calling ADIOI_GEN_WriteStrided
[3/8] ADIOI_TESTFS_ReadContig called on foo.view
[3/8] reading (buf = 0x2b37e3206010, loc = 3145728, sz = 524288)
[3/8] ADIOI_TESTFS_WriteContig called on foo.view
[3/8] writing (buf = 0x2b37e3206010, loc = 3145728, sz = 524288)
[3/8] ADIOI_TESTFS_ReadContig called on foo.view
[3/8] reading (buf = 0x2b37e3206010, loc = 3670016, sz = 524288)
[3/8] ADIOI_TESTFS_WriteContig called on foo.view
[3/8] writing (buf = 0x2b37e3206010, loc = 3670016, sz = 524288)
[3/8] ADIOI_TESTFS_Close called on foo.view
[3/8] ADIOI_TESTFS_SetInfo called on foo.noview
[3/8] calling ADIOI_GEN_SetInfo
[3/8] ADIOI_TESTFS_Open called on foo.noview
[3/8] ADIOI_TESTFS_WriteStridedColl called on foo.noview
[3/8] calling ADIOI_GEN_WriteStridedColl
[3/8] ADIOI_TESTFS_WriteContig called on foo.noview
[3/8] writing (buf = 0x7fffc77c1b00, loc = 3145728, sz = 1048576)
[3/8] ADIOI_TESTFS_Close called on foo.noview
==rob
--
Rob Latham
Mathematics and Computer Science Division A215 0178 EA2D B059 8CDF
Argonne National Lab, IL USA B29D F333 664A 4280 315B
More information about the mpich-discuss
mailing list