LU-11423 osc: Do not walk full extent list

It is only possible to merge with the extent immediately
before or immediately after the one we are trying to add,
so do not continue to walk the extent list after passing
that extent.

This has a significant impact when writing large sparse
files, where most writes create a new extent, and many
extents are too distant to be merged with their neighbors.

Writing 2 GiB of data randomly 4K at a time, we see an
improvement of about 15% with this patch.

mpirun -n 1 $IOR -w -t 4K -b 2G -o ./file -z
w/o patch:
write         285.86 MiB/s
w/patch:
write         324.03 MiB/s

Cray-bug-id: LUS-6523
Signed-off-by: Patrick Farrell <paf@cray.com>
Change-Id: I3da224762638aa71714cfc6dd1f0abac42e1f358
Reviewed-on: https://review.whamcloud.com/33227
Reviewed-by: Jinshan Xiong <jinshan.xiong@gmail.com>
Reviewed-by: Bobi Jam <bobijam@hotmail.com>
Tested-by: Jenkins
Tested-by: Maloo <hpdd-maloo@intel.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>

diff --git a/lustre/osc/osc_cache.c b/lustre/osc/osc_cache.c
index b9112ef..7524899 100644 (file)
--- a/lustre/osc/osc_cache.c
+++ b/lustre/osc/osc_cache.c
@@ -699,7 +699,7 @@ restart:
                pgoff_t ext_chk_end   = ext->oe_end   >> ppc_bits;
 
                LASSERT(sanity_check_nolock(ext) == 0);
-               if (chunk > ext_chk_end + 1)
+               if (chunk > ext_chk_end + 1 || chunk < ext_chk_start)
                        break;
 
                /* if covering by different locks, no chance to match */