Hisashi Hifumi
2009-Mar-05 08:25 UTC
[Ocfs2-devel] [PATCH] OCFS2: Pagecache usage optimization on OCFS2
Hi.
I introduced "is_partially_uptodate" aops for OCFS2.
A page can have multiple buffers and even if a page is not uptodate, some
buffers
can be uptodate on pagesize != blocksize environment.
This aops checks that all buffers which correspond to a part of a file
that we want to read are uptodate. If so, we do not have to issue actual
read IO to HDD even if a page is not uptodate because the portion we
want to read are uptodate.
"block_is_partially_uptodate" function is already used by ext2/3/4.
With the following patch random read/write mixed workloads or random read after
random write workloads can be optimized and we can get performance improvement.
I did a performance test using the sysbench.
#sysbench --num-threads=16 --max-requests=150000 --test=fileio --file-num=1
--file-block-size=8K --file-total-size=1G --file-test-mode=rndrw
--file-fsync-freq=0 --file-rw-ratio=0.5 run
-2.6.29-rc7
Test execution summary:
total time: 82.3114s
total number of events: 150000
total time taken by event execution: 989.1867
per-request statistics:
min: 0.0000s
avg: 0.0066s
max: 0.4400s
approx. 95 percentile: 0.0417s
Threads fairness:
events (avg/stddev): 9375.0000/257.65
execution time (avg/stddev): 61.8242/0.01
-2.6.29-rc7-patched
Test execution summary:
total time: 71.9416s
total number of events: 150000
total time taken by event execution: 845.7592
per-request statistics:
min: 0.0000s
avg: 0.0056s
max: 0.5892s
approx. 95 percentile: 0.0374s
Threads fairness:
events (avg/stddev): 9375.0000/263.46
execution time (avg/stddev): 52.8600/0.01
arch: ia64
pagesize: 16k
blocksize: 4k
Please merge following patch.
Thanks.
Signed-off-by: Hisashi Hifumi <hifumi.hisashi at oss.ntt.co.jp>
diff -Nrup linux-2.6.29-rc7.org/fs/ocfs2/aops.c linux-2.6.29-rc7/fs/ocfs2/aops.c
--- linux-2.6.29-rc7.org/fs/ocfs2/aops.c 2009-03-05 13:46:07.000000000 +0900
+++ linux-2.6.29-rc7/fs/ocfs2/aops.c 2009-03-05 13:50:59.000000000 +0900
@@ -1953,15 +1953,16 @@ static int ocfs2_write_end(struct file *
}
const struct address_space_operations ocfs2_aops = {
- .readpage = ocfs2_readpage,
- .readpages = ocfs2_readpages,
- .writepage = ocfs2_writepage,
- .write_begin = ocfs2_write_begin,
- .write_end = ocfs2_write_end,
- .bmap = ocfs2_bmap,
- .sync_page = block_sync_page,
- .direct_IO = ocfs2_direct_IO,
- .invalidatepage = ocfs2_invalidatepage,
- .releasepage = ocfs2_releasepage,
- .migratepage = buffer_migrate_page,
+ .readpage = ocfs2_readpage,
+ .readpages = ocfs2_readpages,
+ .writepage = ocfs2_writepage,
+ .write_begin = ocfs2_write_begin,
+ .write_end = ocfs2_write_end,
+ .bmap = ocfs2_bmap,
+ .sync_page = block_sync_page,
+ .direct_IO = ocfs2_direct_IO,
+ .invalidatepage = ocfs2_invalidatepage,
+ .releasepage = ocfs2_releasepage,
+ .migratepage = buffer_migrate_page,
+ .is_partially_uptodate = block_is_partially_uptodate,
};
Hisashi Hifumi
2009-Mar-17 04:12 UTC
[Ocfs2-devel] [PATCH] OCFS2: Pagecache usage optimization on OCFS2
Hi Mark.
I introduced "is_partially_uptodate" aops for OCFS2.
A page can have multiple buffers and even if a page is not uptodate, some
buffers
can be uptodate on pagesize != blocksize environment.
This aops checks that all buffers which correspond to a part of a file
that we want to read are uptodate. If so, we do not have to issue actual
read IO to HDD even if a page is not uptodate because the portion we
want to read are uptodate.
"block_is_partially_uptodate" function is already used by ext2/3/4.
With the following patch random read/write mixed workloads or random read after
random write workloads can be optimized and we can get performance improvement.
I did a performance test using the sysbench.
#sysbench --num-threads=16 --max-requests=150000 --test=fileio --file-num=1
--file-block-size=8K --file-total-size=1G --file-test-mode=rndrw
--file-fsync-freq=0 --file-rw-ratio=0.5 run
-2.6.29-rc7
Test execution summary:
total time: 82.3114s
total number of events: 150000
total time taken by event execution: 989.1867
per-request statistics:
min: 0.0000s
avg: 0.0066s
max: 0.4400s
approx. 95 percentile: 0.0417s
Threads fairness:
events (avg/stddev): 9375.0000/257.65
execution time (avg/stddev): 61.8242/0.01
-2.6.29-rc7-patched
Test execution summary:
total time: 71.9416s
total number of events: 150000
total time taken by event execution: 845.7592
per-request statistics:
min: 0.0000s
avg: 0.0056s
max: 0.5892s
approx. 95 percentile: 0.0374s
Threads fairness:
events (avg/stddev): 9375.0000/263.46
execution time (avg/stddev): 52.8600/0.01
arch: ia64
pagesize: 16k
blocksize: 4k
Please merge following patch.
Thanks.
Signed-off-by: Hisashi Hifumi <hifumi.hisashi at oss.ntt.co.jp>
diff -Nrup linux-2.6.29-rc8.org/fs/ocfs2/aops.c
linux-2.6.29-rc8.ocfs2/fs/ocfs2/aops.c
--- linux-2.6.29-rc8.org/fs/ocfs2/aops.c 2009-03-16 15:52:58.000000000 +0900
+++ linux-2.6.29-rc8.ocfs2/fs/ocfs2/aops.c 2009-03-16 16:15:42.000000000 +0900
@@ -1953,15 +1953,16 @@ static int ocfs2_write_end(struct file *
}
const struct address_space_operations ocfs2_aops = {
- .readpage = ocfs2_readpage,
- .readpages = ocfs2_readpages,
- .writepage = ocfs2_writepage,
- .write_begin = ocfs2_write_begin,
- .write_end = ocfs2_write_end,
- .bmap = ocfs2_bmap,
- .sync_page = block_sync_page,
- .direct_IO = ocfs2_direct_IO,
- .invalidatepage = ocfs2_invalidatepage,
- .releasepage = ocfs2_releasepage,
- .migratepage = buffer_migrate_page,
+ .readpage = ocfs2_readpage,
+ .readpages = ocfs2_readpages,
+ .writepage = ocfs2_writepage,
+ .write_begin = ocfs2_write_begin,
+ .write_end = ocfs2_write_end,
+ .bmap = ocfs2_bmap,
+ .sync_page = block_sync_page,
+ .direct_IO = ocfs2_direct_IO,
+ .invalidatepage = ocfs2_invalidatepage,
+ .releasepage = ocfs2_releasepage,
+ .migratepage = buffer_migrate_page,
+ .is_partially_uptodate = block_is_partially_uptodate,
};