Re: [ckrm-tech] [RFC][PATCH 5/5] RSS accounting at the page level

Re: [ckrm-tech] [RFC][PATCH 5/5] RSS accounting at the page level

[Patrick.Le-Dot]

> ...
> This would limit the numbers to groups to the word size on the machine.

yes, this should be the bigger disadvantage of this implementation...
But may be acceptable for a prototype, at least to explain the concept ?


> It would be interesting if we can support shared pages without any
> changes to struct page.

I suppose that means you are on a system without kswapd...

Is everybody OK with that ?
This is a question for the linux-mm list...


> Any particular reason for not implementing migration in this patch.

Nothing special, only incremental code, step by step.
So first try to have a sane shared pages accounting...

> Do you have any test results with this patch? Showing the effect of
> tracking shared pages

Only the RSS counter after reboot (same hw/software config) :

with your patch :
# mount -t container none /dev/container
# cat /dev/container/memctlr.stats
RSS Pages 10571

and with my shared pages accounting patch :
# mount -t container none /dev/container
# cat /dev/container/memctlr.stats
RSS Pages 7329


Patrick

+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+    Patrick Le Dot
 mailto: Patrick.Le-Dot@bull.net         Centre UNIX de BULL SAS
 Phone : +33 4 76 29 73 20               1, Rue de Provence     BP 208
 Fax   : +33 4 76 29 76 00               38130 ECHIROLLES Cedex FRANCE
 Bull, Architect of an Open World TM
 www.bull.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-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [ckrm-tech] [RFC][PATCH 5/5] RSS accounting at the page level

[Balbir Singh]

Patrick.Le-Dot wrote:
>> ...
>> This would limit the numbers to groups to the word size on the machine.
> 
> yes, this should be the bigger disadvantage of this implementation...
> But may be acceptable for a prototype, at least to explain the concept ?
> 

I think we need to find a more efficient mechanism to track shared pages

> 
>> It would be interesting if we can support shared pages without any
>> changes to struct page.
> 
> I suppose that means you are on a system without kswapd...
> 
> Is everybody OK with that ?
> This is a question for the linux-mm list...
> 


No, I have kswapd, like I said earlier, I have a patch that uses rmap
information for detecting and accounting shared pages. I hope to
post a patch soon.

> 
>> Any particular reason for not implementing migration in this patch.
> 
> Nothing special, only incremental code, step by step.
> So first try to have a sane shared pages accounting...

Aah, ok

> 
>> Do you have any test results with this patch? Showing the effect of
>> tracking shared pages
> 
> Only the RSS counter after reboot (same hw/software config) :
> 
> with your patch :
> # mount -t container none /dev/container
> # cat /dev/container/memctlr.stats
> RSS Pages 10571
> 
> and with my shared pages accounting patch :
> # mount -t container none /dev/container
> # cat /dev/container/memctlr.stats
> RSS Pages 7329
> 
> 

Is there any way to print out the shared pages, I think it should
easy to track shared pages per container as an accountable parameter.

-- 

	Balbir Singh,
	Linux Technology Center,
	IBM Software Labs

--
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-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [ckrm-tech] [RFC][PATCH 5/5] RSS accounting at the page level

[Patrick.Le-Dot]

Balbir Singh wrote:
>>> ...
>>> This would limit the numbers to groups to the word size on the machine.
>> 
>> yes, this should be the bigger disadvantage of this implementation...
>> But may be acceptable for a prototype, at least to explain the concept ?
>> 
> 
> I think we need to find a more efficient mechanism to track shared pages

To clarify, bitmap is just an idea to avoid the rmap walk when the number
of groups is not too large and then kswapd can use a very fast check for
each page...


> ...
> Is there any way to print out the shared pages, I think it should
> easy to track shared pages per container as an accountable parameter.

May be "private pages per container" is more representative ?
I have to think about that...

Patrick

--
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-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>