On Sat 23-07-11 15:43:45, Wu Fengguang wrote:

On Fri, Jul 15, 2011 at 05:34:09AM +0800, Jan Kara wrote:

quoted

quoted

- tasks dirtying close to 25% pages probably cannot be called light
  dirtier and there is no need to protect such tasks

  The idea is interesting. The only problem is that we don't want to set
dirty_exceeded too late so that heavy dirtiers won't push light dirtiers
over their limits so easily due to ratelimiting. It did some computations:
We normally ratelimit after 4 MB. Take a low end desktop these days. Say
1 GB of ram, 4 CPUs. So dirty limit will be ~200 MB and the area for task
differentiation ~25 MB. We enter balance_dirty_pages() after dirtying
num_cpu * ratelimit / 2 pages on average which gives 8 MB. So we should
set dirty_exceeded at latest at bdi_dirty / TASK_LIMIT_FRACTION / 2 or
task differentiation would have no effect because of ratelimiting.

So we could change the limit to something like:
bdi_dirty - min(bdi_dirty / TASK_LIMIT_FRACTION, ratelimit_pages *
num_online_cpus / 2 + bdi_dirty / TASK_LIMIT_FRACTION / 16)

Good analyze!

quoted

But I'm not sure setups where this would make difference are common...

I think I'd prefer the original simple patch given that the common
1-dirtier is not impacted.

  OK, thanks. So will you merge the patch please?

								Honza
-- 
Jan Kara [off-list ref]
SUSE Labs, CR

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>