2007-10-11 17:19:45

by Richard Kennedy

Subject: [RFC] mm - background_writeout exits when pages_skipped ?

When background_writeout() (mm/page-writeback.c) finds any pages_skipped
in writeback_inodes() and it didn't meet any congestion, it exits even
when it hasn't written enough pages yet.

Performing 2 ( or more) concurrent copies of a large file, often creates
lots of skipped pages (1000+) making background_writeout exit and so
pages don't get written out until we reach dirty_ratio.

I added some instrumentation to fs/buffer.c in
__block_write_full_page(..) and all the skipped pages come from here :-

done:
if (nr_underway == 0) {
/*
* The page was marked dirty, but the buffers were
* clean. Someone wrote them back by hand with
* ll_rw_block/submit_bh. A rare case.
*/
end_page_writeback(page);

/*
* The page and buffer_heads can be released at any time from
* here on.
*/
wbc->pages_skipped++; /* We didn't write this page */

maybe not such a rare case! :)

I've been testing 2.6.23 on an AMD64x2.

Here's a quick patch for background_writeout to ignore pages_skipped. It
helps keep nr_dirty between dirty_background_ratio & dirty_ratio, and
once the copies have finish nr_dirty quickly drops back to
dirty_background_ratio.

Without the patch during the copy nr_dirty stays around dirty_ratio and
takes a long time to drop after it finishes.

It seems that this patch tackles the problem, but is there a better way
to fix it?
And is there a good reason to abandon this writeout loop if a page gets
skipped for any other reason?

thanks
richard



diff --git a/mm/page-writeback.c b/mm/page-writeback.c
index 4472036..5a6747b 100644
--- a/mm/page-writeback.c
+++ b/mm/page-writeback.c
@@ -371,6 +371,7 @@ static void background_writeout(unsigned long _min_pages)
.nr_to_write = 0,
.nonblocking = 1,
.range_cyclic = 1,
+ .encountered_congestion = 0,
};

for ( ; ; ) {
@@ -382,17 +383,16 @@ static void background_writeout(unsigned long _min_pages)
global_page_state(NR_UNSTABLE_NFS) < background_thresh
&& min_pages <= 0)
break;
- wbc.encountered_congestion = 0;
+ if (wbc.encountered_congestion) {
+ congestion_wait(WRITE, HZ/10);
+ wbc.encountered_congestion = 0;
+ }
wbc.nr_to_write = MAX_WRITEBACK_PAGES;
wbc.pages_skipped = 0;
writeback_inodes(&wbc);
min_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write;
- if (wbc.nr_to_write > 0 || wbc.pages_skipped > 0) {
- /* Wrote less than expected */
- congestion_wait(WRITE, HZ/10);
- if (!wbc.encountered_congestion)
- break;
- }
+ if (wbc.nr_to_write > 0 && !wbc.encountered_congestion)
+ break;
}
}

2007-10-14 11:09:49

by Wu Fengguang

Subject: Re: [RFC] mm - background_writeout exits when pages_skipped ?

On Thu, Oct 11, 2007 at 06:19:34PM +0100, richard kennedy wrote:
> When background_writeout() (mm/page-writeback.c) finds any pages_skipped
> in writeback_inodes() and it didn't meet any congestion, it exits even
> when it hasn't written enough pages yet.
>
> Performing 2 ( or more) concurrent copies of a large file, often creates
> lots of skipped pages (1000+) making background_writeout exit and so
> pages don't get written out until we reach dirty_ratio.
>
> I added some instrumentation to fs/buffer.c in
> __block_write_full_page(..) and all the skipped pages come from here :-
>
> done:
> if (nr_underway == 0) {
> /*
> * The page was marked dirty, but the buffers were
> * clean. Someone wrote them back by hand with
> * ll_rw_block/submit_bh. A rare case.
> */
> end_page_writeback(page);
>
> /*
> * The page and buffer_heads can be released at any time from
> * here on.
> */

> wbc->pages_skipped++; /* We didn't write this page */

FYI: The above line has just been removed in 2.6.23-mm1, which fixed the bug.

Thank you,
Fengguang