lkml.org 
[lkml]   [2013]   [Aug]   [12]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [patch 8/9] mm: thrash detection-based file cache sizing
Hello Vlastimil!

On Sun, Aug 11, 2013 at 11:57:37PM +0200, Vlastimil Babka wrote:
> On 08/07/2013 12:44 AM, Johannes Weiner wrote:
> >To accomplish this, a per-zone counter is increased every time a page
> >is evicted and a snapshot of that counter is stored as shadow entry in
> >the page's now empty page cache radix tree slot. Upon refault of that
> >page, the difference between the current value of that counter and the
> >shadow entry value is called the refault distance. It tells how many
> >pages have been evicted from the zone since that page's eviction,
> This explanation of refault distance seems correct...
> >which is how many page slots are missing from the zone's inactive list
> >for this page to get accessed twice while in memory.
> But this part seems slightly incorrect. IMHO the correct formulation
> would be "...how many page slots are AT MOST missing...". See below.

Yes, I think this would be better phrasing.

> >If the number of
> >missing slots is less than or equal to the number of active pages,
> >increasing the inactive list at the cost of the active list would give
> >this thrashing set a chance to establish itself:
> >
> >eviction counter = 4
> > evicted inactive active
> > Page cache data: [ a b c d ] [ e f g h i j k ] [ l m n ]
> > Shadow entries: 0 1 2 3
> >Refault distance: 4 3 2 1
> Consider here that if 'd' was now accessed before 'c', I think 'e'
> would be evicted and eviction counter would be incremented to 5. So
> for 'c' you would now say that three slots would prevent the
> refault, but in fact two would still be sufficient. This potential
> imprecision could make the algorithm challenge more active pages
> than it should, but I am not sure how bad the consequences could
> be... so just pointing it out.

Yes, pages are not accessed in strict order all the time and there is
very much the occasional outlier that gets accessed before its peers.

In fact, the code treats it as an AT MOST already. Every page fault
is evaluated based on its refault distance, but a refault distance of
N does not mean that N pages are deactivated. Only one page is.

An outlier is thus harmless but we catch if a whole group of pages is
thrashing.

The changelog and documentation should be updated, thanks for pointing
this out.


\
 
 \ /
  Last update: 2013-08-12 19:21    [W:0.145 / U:0.280 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site