[RFC 0/2] iomap: Add support for subpage dirty state tracking to improve write performance

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hello,

Please find the RFC patchset which adds support for iomap subpage dirty state 
tracking which improves write performance and should reduce the write amplification 
on platforms with smaller filesystem blocksize compared to pagesize.
E.g. On Power with 64k default pagesize and with 4k XFS filesystem blocksize.

I have done some minimal fsstress and fstests testing using this patchset 
and haven't noticed any issues as such. Posting this RFC to get some 
initial comments/thoughts on the patch. 
I will run full fstests with XFS if this RFC looks good. 

>From review perspective, it will be helpful if one can also review the error 
handling path. I wasn't sure on whether we need to clear the dirty state bitmap 
of blocks within a folio or not in iomap_writepage_map(). I don't clear that,
since AFAIU, the error in that function is due to failed ->map_blocks() function
which has nothing to do with tracking subpage dirty state of a block within
folio. But please let me know your thoughts on this or other error handling path.


Performance results
======================
1. Performance testing of below fio workload reveals ~16x performance
improvement on nvme with XFS (4k blocksize) on Power (64K pagesize)
FIO reported write bw scores, improved from ~28 MBps to ~452 MBps.

<test_randwrite.fio>
[global]
	ioengine=psync
	rw=randwrite
	overwrite=1
	pre_read=1
	direct=0
	bs=4k
	size=1G
	dir=./
	numjobs=8
	fdatasync=1
	runtime=60
	iodepth=64
	group_reporting=1

[fio-run]

2. Also our internal performance team reported that this patch improves there
   database workload performance by around ~83% (with XFS on Power)


Note: I did come across an older RFC around the same logic to track subpage
dirty tracking here [1]. But it seems no one pursued it after iomap received
folio changes update. 

[1]: https://lore.kernel.org/linux-xfs/20200821123306.1658495-1-yukuai3@xxxxxxxxxx/#t

Ritesh Harjani (IBM) (2):
  iomap: Change uptodate variable name to state
  iomap: Support subpage size dirty tracking to improve write performance

 fs/iomap/buffered-io.c | 79 ++++++++++++++++++++++++++++++++++--------
 1 file changed, 64 insertions(+), 15 deletions(-)

-- 
2.37.3




[Index of Archives]     [XFS Filesystem Development (older mail)]     [Linux Filesystem Development]     [Linux Audio Users]     [Yosemite Trails]     [Linux Kernel]     [Linux RAID]     [Linux SCSI]


  Powered by Linux