[lkml]   [2012]   [May]   [16]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [patch 0/5] refault distance-based file cache sizing

On 2012/05/16 14:51, Johannes Weiner wrote:
> Hi Nai,
> On Wed, May 16, 2012 at 01:25:34PM +0800, nai.xia wrote:
>> Hi Johannes,
>> Just out of curiosity(since I didn't study deep into the
>> reclaiming algorithms), I can recall from here that around 2005,
>> there was an(or some?) implementation of the "Clock-pro" algorithm
>> which also have the idea of "reuse distance", but it seems that algo
>> did not work well enough to get merged? Does this patch series finally
>> solve the problem(s) with "Clock-pro" or totally doesn't have to worry
>> about the similar problems?
> As far as I understood, clock-pro set out to solve more problems than
> my patch set and it failed to satisfy everybody.
> The main error case was that it could not partially cache data of a
> set that was bigger than memory. Instead, looping over the file
> repeatedly always has to read every single page because the most
> recent page allocations would push out the pages needed in the nearest
> future. I never promised to solve this problem in the first place.
> But giving more memory to the big looping load is not useful in our
> current situation, and at least my code protects smaller sets of
> active cache from these loops. So it's not optimal, but it sucks only
> half as much :)

Yep, I see ;)

> There may have been improvements from clock-pro, but it's hard to get
> code merged that does not behave as expected in theory with nobody
> understanding what's going on.
> My code is fairly simple, works for the tests I've done and the
> behaviour observed so far is understood (at least by me).

OK, I assume that you do aware that the system you constructed with
this simple and understandable idea looks like a so called "feedback
system"? Or in other words, I think theoretically the refault-distance
of a page before and after your algorithm is applied is not the same.
And this changed refault-distance pattern is then feed as input into
your algorithm. A feedback system may be hard(and may be simple) to
analyze but may also work well magically.

Well, again I confess I've not done enough course in this area. Just hope
that my words can help you think more comprehensively. :)



 \ /
  Last update: 2012-05-16 15:41    [W:0.249 / U:2.300 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site