some silly things

some silly things

[jordi polo]

 - it seems to me that you don't do nr_inactive_clear_pages   when you get a new inactive_clear page
 
 - I don't know why you have to test if a page is dirty in reclaim_page(), there isn't the place, it is supposed that when the page is written, in other place must be allocated in inactive_dirty. Here we can expect inactive_clean pages are really inactive clean pages.
 
 - what do you think of 4 lists, active_clean active_dirty inactive_clean inactive_dirty. When a write operation occurs in a page this will go to active_dirty, from active_dirty to inactive_dirty , from active_clean to inactive_clean directly without need to test if it's dirty or not.
 In fact is the same but with 4 list you can make a state machine and you can track exactly a page trought the states, but your 3 list approach seems all right for me too.
 
> - And the improvement I was trying to explain you the other day, something this way:
> 
> if ((PageActiveClear(page)) && (page->age < MINIM )){
> deactivate_page(page); //this page will go to inactive_clear
> }
> else if ((PageActiveDirty(page)) && (!page->age )){
> deactivate_page(page); // this will go to inactive_dirty 
> }
 
> Here MINIM will be something like 1 (or 2) (maybe need to change
PAGE_AGE_START and PAGE_AGE_ADV), this will do that the clear pages have it
easier to go to inactive_clear and then easier to become free pages and I think
this is a good thing because it's harder to get a free page from inactive_dirty
than from inactive_clear. > I know in your code first begin to free the
inactive_clear and when you are running off it you use inactive_dirty, that's
allright, but if you do it my way there will be more pages in inactive_clear
that will make that you need to free less pages from inactive_dirty  

- age_page_down() : page->age /=2;   > you make a silly improvement with  
page->age >>=2; 
- you can do in the zone structure a field named available_pages=
free_pages inactive_free so you mustn't calculate it everytime.   


 Just hoping these things will be useful to you.
 Jordi Polo (trusmis)
 mumismo@wanadoo.es






_________________________________________________________
http://www.latinmail.com.  Gratuito, latino y en espanol.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux.eu.org/Linux-MM/

Re: some silly things

[Rik van Riel]

On Thu, 17 Aug 2000, root wrote:

>  - it seems to me that you don't do nr_inactive_clear_pages++
> when you get a new inactive_clear page

That's because there is no global counter for the number
of inactive_clean pages. We only count these per zone.

>  - I don't know why you have to test if a page is dirty in
> reclaim_page(), there isn't the place, it is supposed that when
> the page is written, in other place must be allocated in
> inactive_dirty. Here we can expect inactive_clean pages are
> really inactive clean pages.

Unless somebody does something with the page _after_ it is
added to the inactive_clean list. Granted, __find_page_nolock
should have moved the page to the active list by then, but
that is only valid in the _current_ VM and I would like to
have the code resistant against some future VM modifications
too.

Defensive programming is important if you want to avoid (or
detect) bugs. Both bugs in your own code and in other code.

>  - what do you think of 4 lists, active_clean active_dirty
> inactive_clean inactive_dirty. When a write operation occurs in
> a page this will go to active_dirty, from active_dirty to
> inactive_dirty , from active_clean to inactive_clean directly
> without need to test if it's dirty or not.

Doesn't make any sense. Checking for dirty state will have to
be done somewhere, so why not do it just before we actually
plan on writing the page to disk?

>  In fact is the same but with 4 list you can make a state
> machine and you can track exactly a page trought the states,

Wooohoooo, fun!  ;)

> > if ((PageActiveClear(page)) && (page->age < MINIM )){
> > deactivate_page(page); //this page will go to inactive_clear
> > }
> > else if ((PageActiveDirty(page)) && (!page->age )){
> > deactivate_page(page); // this will go to inactive_dirty 
> > }

NOOOOOOO!  The whole idea is to have -page aging- and
-page flushing- SEPARATE, so that we always make the
right decision in flushing pages.

Whether a page is dirty or clean should not make any
difference in how "important" we think it is to keep
the page in memory.

regards,

Rik
--
"What you're running that piece of shit Gnome?!?!"
       -- Miguel de Icaza, UKUUG 2000

http://www.conectiva.com/		http://www.surriel.com/

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux.eu.org/Linux-MM/