Re: regression in page writeback

[Wu Fengguang <fengguang.wu@intel.com>]

On Tue, Sep 22, 2009 at 07:50:15PM +0800, Li, Shaohua wrote:
> On Tue, Sep 22, 2009 at 06:49:15PM +0800, Wu, Fengguang wrote:
> > Shaohua,
> > 
> > On Tue, Sep 22, 2009 at 01:49:13PM +0800, Li, Shaohua wrote:
> > > Hi,
> > > Commit d7831a0bdf06b9f722b947bb0c205ff7d77cebd8 causes disk io regression
> > > in my test.
> > > My system has 12 disks, each disk has two partitions. System runs fio sequence
> > > write on all partitions, each partion has 8 jobs.
> > > 2.6.31-rc1, fio gives 460m/s disk io
> > > 2.6.31-rc2, fio gives about 400m/s disk io. Revert the patch, speed back to
> > > 460m/s
> > > 
> > > Under latest git: fio gives 450m/s disk io; If reverting the patch, the speed
> > > is 484m/s.
> > > 
> > > With the patch, fio reports less io merge and more interrupts. My naive
> > > analysis is the patch makes balance_dirty_pages_ratelimited_nr() limits
> > > write chunk to 8 pages and then soon go to sleep in balance_dirty_pages(),
> > > because most time the bdi_nr_reclaimable < bdi_thresh, and so when write
> > > the pages out, the chunk is 8 pages long instead of 4M long. Without the patch,
> > > thread can write 8 pages and then move some pages to writeback, and then
> > > continue doing write. The patch seems to break this.
> > 
> > Do you have trace/numbers for above descriptions?
> No. Just guess, because there is less io merge. And watch each bdi's states,
> bdi_nr_reclaimable < bdi_thresh seems always true.

Ah OK.

> > > Unfortunatelly I can't figure out a fix for this issue, hopefully
> > > you have more ideas.
> > 
> > Attached is a very verbose writeback debug patch, hope it helps and
> > won't disturb the workload a lot :)
> Hmm, the log buf will get overflowed soon, there is > 400m/s io. I tried
> to produce this issue in a system with two disks, but fail. Anyway, I'll try
> it out tomorrow.

Thank you~  I'd recommend to use netconsole or serial line, and stop
local klogd because the write of log messages could add noises. 

Thanks,
Fengguang