Re: [RFT][PATCH] mm: drop behind

From: Chris Snook
Date: Thu Jul 12 2007 - 15:20:50 EST


Peter Zijlstra wrote:
Hi Tim,

On Wed, 2007-07-11 at 15:37 -0700, Tim Pepper wrote:
On 7/9/07, Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
Use the read-ahead code to provide hints to page reclaim.

This patch has the potential to solve the streaming-IO trashes my
desktop problem.

It tries to aggressively reclaim pages that were loaded in a strong
sequential pattern and have been consumed. Thereby limiting the damage
to the current resident set.
Interesting...

Would it make sense to tie this into (finally) making
POSIX_FADV_NOREUSE something more than a noop?

We talked about that, but the thing is, if we make the functionality
conditional, nobody will ever use it :-/

So, yes, in a perfect world that would indeed make sense. However since
nobody ever uses these [fm]advise calls,..

So the big question is, does this functionally hurt any workload? If it
turns out it does (which I still doubt) then we might hide it behind
knobs, otherwise I'd like to keep it always on.

Then do what we do for FADV_SEQUENTIAL. With that advice, we double the readahead window. We're already doing readahead, but we do a lot more when we have the advice. NOREUSE should put much greater pressure on the vm to drop these pages quickly, or perhaps simply eliminate the heuristic evaluation of the access pattern and short-circuit straight to dropping the pages.

We should be encouraging application writers to actually use things like fadvise when they can tune things more intelligently than kernel heuristics can.

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