summaryrefslogtreecommitdiff
path: root/mm/filemap.c
diff options
context:
space:
mode:
authorWu Fengguang <fengguang.wu@intel.com>2010-07-22 06:11:53 +0400
committerWu Fengguang <fengguang.wu@intel.com>2011-06-08 04:25:21 +0400
commit424b351fe1901fc909fd0ca4f21dab58f24c1aac (patch)
tree51fc3aa51991cf21756adcb5158cd95c89209b20 /mm/filemap.c
parentba9aa8399fda48510d80c2fed1afb8fedbe1bb41 (diff)
downloadlinux-424b351fe1901fc909fd0ca4f21dab58f24c1aac.tar.xz
writeback: refill b_io iff empty
There is no point to carry different refill policies between for_kupdate and other type of works. Use a consistent "refill b_io iff empty" policy which can guarantee fairness in an easy to understand way. A b_io refill will setup a _fixed_ work set with all currently eligible inodes and start a new round of walk through b_io. The "fixed" work set means no new inodes will be added to the work set during the walk. Only when a complete walk over b_io is done, new inodes that are eligible at the time will be enqueued and the walk be started over. This procedure provides fairness among the inodes because it guarantees each inode to be synced once and only once at each round. So all inodes will be free from starvations. This change relies on wb_writeback() to keep retrying as long as we made some progress on cleaning some pages and/or inodes. Without that ability, the old logic on background works relies on aggressively queuing all eligible inodes into b_io at every time. But that's not a guarantee. The below test script completes a slightly faster now: 2.6.39-rc3 2.6.39-rc3-dyn-expire+ ------------------------------------------------ all elapsed 256.043 252.367 stddev 24.381 12.530 tar elapsed 30.097 28.808 dd elapsed 13.214 11.782 #!/bin/zsh cp /c/linux-2.6.38.3.tar.bz2 /dev/shm/ umount /dev/sda7 mkfs.xfs -f /dev/sda7 mount /dev/sda7 /fs echo 3 > /proc/sys/vm/drop_caches tic=$(cat /proc/uptime|cut -d' ' -f2) cd /fs time tar jxf /dev/shm/linux-2.6.38.3.tar.bz2 & time dd if=/dev/zero of=/fs/zero bs=1M count=1000 & wait sync tac=$(cat /proc/uptime|cut -d' ' -f2) echo elapsed: $((tac - tic)) It maintains roughly the same small vs. large file writeout shares, and offers large files better chances to be written in nice 4M chunks. Analyzes from Dave Chinner in great details: Let's say we have lots of inodes with 100 dirty pages being created, and one large writeback going on. We expire 8 new inodes for every 1024 pages we write back. With the old code, we do: b_more_io (large inode) -> b_io (1l) 8 newly expired inodes -> b_io (1l, 8s) writeback large inode 1024 pages -> b_more_io b_more_io (large inode) -> b_io (8s, 1l) 8 newly expired inodes -> b_io (8s, 1l, 8s) writeback 8 small inodes 800 pages 1 large inode 224 pages -> b_more_io b_more_io (large inode) -> b_io (8s, 1l) 8 newly expired inodes -> b_io (8s, 1l, 8s) ..... Your new code: b_more_io (large inode) -> b_io (1l) 8 newly expired inodes -> b_io (1l, 8s) writeback large inode 1024 pages -> b_more_io (b_io == 8s) writeback 8 small inodes 800 pages b_io empty: (1800 pages written) b_more_io (large inode) -> b_io (1l) 14 newly expired inodes -> b_io (1l, 14s) writeback large inode 1024 pages -> b_more_io (b_io == 14s) writeback 10 small inodes 1000 pages 1 small inode 24 pages -> b_more_io (1l, 1s(24)) writeback 5 small inodes 500 pages b_io empty: (2548 pages written) b_more_io (large inode) -> b_io (1l, 1s(24)) 20 newly expired inodes -> b_io (1l, 1s(24), 20s) ...... Rough progression of pages written at b_io refill: Old code: total large file % of writeback 1024 224 21.9% (fixed) New code: total large file % of writeback 1800 1024 ~55% 2550 1024 ~40% 3050 1024 ~33% 3500 1024 ~29% 3950 1024 ~26% 4250 1024 ~24% 4500 1024 ~22.7% 4700 1024 ~21.7% 4800 1024 ~21.3% 4800 1024 ~21.3% (pretty much steady state from here) Ok, so the steady state is reached with a similar percentage of writeback to the large file as the existing code. Ok, that's good, but providing some evidence that is doesn't change the shared of writeback to the large should be in the commit message ;) The other advantage to this is that we always write 1024 page chunks to the large file, rather than smaller "whatever remains" chunks. CC: Jan Kara <jack@suse.cz> Acked-by: Mel Gorman <mel@csn.ul.ie> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
Diffstat (limited to 'mm/filemap.c')
0 files changed, 0 insertions, 0 deletions