How kernel applies LRU to swap out physical pages?

How kernel applies LRU to swap out physical pages?

[Pan ruochen]

Hi All,
I'm reading the linux kernel 2.6.14. But can't understand how LRU algorithm
is applied to page swapping. what is the 'USED' of LRU referred to? The
reference counter which the page is mapped by processes, or the frequency that
CPU reads/writes in the page. In common sense, the LRU algorithm will be more
optimal for the second case. But it depends on CPU capability.
I will be very appreciated if somebody can explain that in details.

-------------
Best Regards,
PRC
Nov 11, 2008

Re: How kernel applies LRU to swap out physical pages?

[Peter Zijlstra]

On Tue, 2008-11-11 at 16:22 +0800, Pan ruochen wrote:
> Hi All,
> I'm reading the linux kernel 2.6.14. But can't understand how LRU algorithm
> is applied to page swapping. what is the 'USED' of LRU referred to? The
> reference counter which the page is mapped by processes, or the frequency that
> CPU reads/writes in the page. In common sense, the LRU algorithm will be more
> optimal for the second case. But it depends on CPU capability.
> I will be very appreciated if somebody can explain that in details.

We don't actually use LRU but something like CLOCK.

Re: How kernel applies LRU to swap out physical pages?

[KOSAKI Motohiro]

Hi

> Hi All,
> I'm reading the linux kernel 2.6.14. 

it seems too old.
I recommend to read newest kernel.

> But can't understand how LRU algorithm
> is applied to page swapping. what is the 'USED' of LRU referred to? The
> reference counter which the page is mapped by processes, or the frequency that
> CPU reads/writes in the page. In common sense, the LRU algorithm will be more
> optimal for the second case. But it depends on CPU capability.
> I will be very appreciated if somebody can explain that in details.

I recommend to read page_referenced() and its related function :)

Re: How kernel applies LRU to swap out physical pages?

[KOSAKI Motohiro]

> On Tue, 2008-11-11 at 16:22 +0800, Pan ruochen wrote:
> > Hi All,
> > I'm reading the linux kernel 2.6.14. But can't understand how LRU algorithm
> > is applied to page swapping. what is the 'USED' of LRU referred to? The
> > reference counter which the page is mapped by processes, or the frequency that
> > CPU reads/writes in the page. In common sense, the LRU algorithm will be more
> > optimal for the second case. But it depends on CPU capability.
> > I will be very appreciated if somebody can explain that in details.
> 
> We don't actually use LRU but something like CLOCK.

Yup.

However, VM replace algorithm folks commonly call it to "LRU"
because MACH call so.

IOW, it is historical reason word, shouldn't imazine strict 
"Least Recently Used" algorithm.

Re: How kernel applies LRU to swap out physical pages?

[Peter Zijlstra]

On Tue, 2008-11-11 at 18:56 +0900, KOSAKI Motohiro wrote:
> > On Tue, 2008-11-11 at 16:22 +0800, Pan ruochen wrote:
> > > Hi All,
> > > I'm reading the linux kernel 2.6.14. But can't understand how LRU algorithm
> > > is applied to page swapping. what is the 'USED' of LRU referred to? The
> > > reference counter which the page is mapped by processes, or the frequency that
> > > CPU reads/writes in the page. In common sense, the LRU algorithm will be more
> > > optimal for the second case. But it depends on CPU capability.
> > > I will be very appreciated if somebody can explain that in details.
> > 
> > We don't actually use LRU but something like CLOCK.

FWIW in case you hadn't found it yet:

 http://en.wikipedia.org/wiki/Page_replacement_algorithm

is a fairly decent page on the subject.

Re: How kernel applies LRU to swap out physical pages?

[Pan ruochen]

Thanks to all, I've have a study.

2008/11/11 Peter Zijlstra <peterz@infradead.org>:
> On Tue, 2008-11-11 at 18:56 +0900, KOSAKI Motohiro wrote:
>> > On Tue, 2008-11-11 at 16:22 +0800, Pan ruochen wrote:
>> > > Hi All,
>> > > I'm reading the linux kernel 2.6.14. But can't understand how LRU algorithm
>> > > is applied to page swapping. what is the 'USED' of LRU referred to? The
>> > > reference counter which the page is mapped by processes, or the frequency that
>> > > CPU reads/writes in the page. In common sense, the LRU algorithm will be more
>> > > optimal for the second case. But it depends on CPU capability.
>> > > I will be very appreciated if somebody can explain that in details.
>> >
>> > We don't actually use LRU but something like CLOCK.
>
> FWIW in case you hadn't found it yet:
>
>  http://en.wikipedia.org/wiki/Page_replacement_algorithm
>
> is a fairly decent page on the subject.
>