[lkml]   [2003]   [Aug]   [6]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [patch] real-time enhanced page allocator and throttling
Robert Love <> wrote:
> On Tue, 2003-08-05 at 17:45, Andrew Morton wrote:
> > It's testing time.
> Just via some instrumenting, I can see that a real-time task never
> begins throttling and this translates to a ~1ms reduction in worst case
> allocation on a fast machine latency under extreme page dirtying and
> writeback (basically, I cannot reproduce any variation in page
> allocation, now, for a real-time test app). So it works.
> But I do not have any real world test to confirm a benefit, which is
> what matters. Have you poked and prodded?

It's pretty easy to demonstrate the benefit of the balance_dirty_pages()
change. Just do:

while true
dd if=/dev/zero of=foo bs=1M count=512 conv=notrunc

and also:

rm 1 ; sleep 3; time dd if=/dev/zero of=1 bs=16M count=1

The 16M dd normally takes 1.5 seconds (I'm pretty please with that btw.
Very repeatable and fair). If you run the 16M dd with SCHED_FIFO it takes
a repeatable 0.12 seconds.

But it's pretty easy to make the machine do bad things with no dirty memory
throttling. Probably we should just treat realtime tasks in the same way

It's a bit harder to demonstrate benefits from the page allocator change.
Allocate 4 megs while there is a huge amount of swapout happening is much
quicker and repeatable. But generally things work pretty well with just
SCHED_OTHER. Needs some more careful testing.

Often you get great big stalls because your SCHED_RR task is stuck in
ext3's journal_start, waiting for kjournald to finish a commit, due to an
atime update. So noatime is needed there. And then it gets stuck in a
disk read.

So running a program off disk isn't a very good test.

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
More majordomo info at
Please read the FAQ at

 \ /
  Last update: 2005-03-22 13:47    [W:0.043 / U:3.992 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site