Message-ID: <28c262361001280009u509f169dnc558d013150ca00b@mail.gmail.com>
Date:	Thu, 28 Jan 2010 17:09:43 +0900
From:	Minchan Kim <minchan.kim@...il.com>
To:	Steve VanDeBogart <vandebo-lkml@...dbox.net>
Cc:	Wu Fengguang <fengguang.wu@...el.com>,
	Chris Frost <frost@...ucla.edu>,
	Andrew Morton <akpm@...ux-foundation.org>,
	Steve Dickson <steved@...hat.com>,
	David Howells <dhowells@...hat.com>,
	Xu Chenfeng <xcf@...c.edu.cn>,
	"linux-mm@...ck.org" <linux-mm@...ck.org>,
	"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] mm/readahead.c: update the LRU positions of in-core 
	pages, too

On Thu, Jan 28, 2010 at 4:16 PM, Steve VanDeBogart
<vandebo-lkml@...dbox.net> wrote:
> On Wed, 27 Jan 2010, Minchan Kim wrote:
>
>> This patch effect happens when inactive file list is small, I think.
>> It means it's high memory pressure. so if we move ra pages into
>
> This patch does the same thing regardless of memory pressure - it
> doesn't just apply in high memory pressure situations.  Is your concern
> that in high memory pressure situations this patch with make things worse?

Yes.

>
>> head of inactive list, other application which require free page urgently
>> suffer from latency or are killed.
>
> I don't think this patch will affect the number of pages reclaimed, only
> which pages are reclaimed.  In extreme cases it could increase the time

Most likely.
But it can affect the number of pages reclaimed at sometime.

For example,

scanning window size for reclaim = 5.
P : hard reclaimable page
R : readaheaded page(easily reclaimable page)

without this patch
HEAD-P - P - P - P ................ - P - R -R -R -R -R- TAIL

reclaimed pages : 5

with this patch
HEAD-R-R-R-R-R .................... - P -P -P -P -P -P -TAIL

reclaimed pages : 0 => might be OOMed.

Yes. It's very extreme example.
it is just for explanation. :)

> needed to reclaim that many pages, but the inactive list would have to be
> very short.

I think short inactive list means now we are suffering from shortage of
free memory. So I think it would be better to discard ra pages rather than
OOMed.

>
>> If VM don't have this patch, of course ra pages are discarded and
>> then I/O performance would be bad. but as I mentioned, it's time
>> high memory pressure. so I/O performance low makes system
>> natural throttling. It can help out of  system memory pressure.
>
> Even in low memory situations, improving I/O performance can help the
> overall system performance.  For example if most of the inactive list is
> dirty, needlessly discarding pages, just to refetch them will clog
> I/O and increase the time needed to write out the dirty pages.

Yes. That's what I said.
But my point is that it makes system I/O throttling by clogging I/O naturally.
It can prevent fast consumption of memory.
Actually I think mm don't have to consider I/O throttling as far as possible.
It's role of I/O subsystem. but it's not real.
There are some codes for I/O throttling in mm.

>
>> In summary I think it's good about viewpoint of I/O but I am not sure
>> it's good about viewpoint of system.
>
> In this case, I think what's good for I/O is good for the system.
> Please help me understand if I am missing something.  Thanks

You didn't missed anything. :)
I don't know how this patch affect in low memory situation.
What we just need is experiment which is valuable.

Wu have a catch in my concern and are making new version.
I am looking forward to that.

>
> --
> Steve
>



-- 
Kind regards,
Minchan Kim
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists