Re: [Lhms-devel] [PATCH] Page eviction support in vmscan.c

[KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>]

Christoph Lameter wrote:
> On Thu, 13 Oct 2005, Joel Schopp wrote:
> 
> 
>>I'm curious what use motivated you to write it.  I think for migration it
>>would usually make more sense to let the swapper free up LRU memory and then
>>do a memory to memory migration.  But I'm not really a migration expert
> 
> 
> The motiviation was the complexity and the problems with the existing hot 
> plug implementation.
> 
> I just tried to simplify page migration as much as possible to come with 
> something that is easy to verify and that may be easily acceptable. We can 
> build on that later and incorporate more elements from the hotplug patch.
> 
Forcing pages swapped-out itself looks useful in some case.
But I think using swap in memory-hotplug is not good because of its performance.

So, this patch will not simplify memory_migrate() ;)
I think that valid direction is simplify memory_migrate() on memory.

-- Kame

> 
>>>However, swapout_pages may not be able to evict all pages for a variety of
>>>reasons.
>>
>>Have you thought about using this in combination with the fragmentation
>>avoidance patches Mel has been posting?  __GFP_USER flag that adds would go a
>>long way toward determining what can and can't be swapped out.  We use that
>>for migration with great success.  I'd assume the criteria for swapout and
>>migration are pretty similar.
> 
> 
> The patch does not determine what can and cannot be swapped out. That is 
> up to the user of the functions defined here. See my other patch that I 
> posted today for one example of a user of this patch.
> 

--
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>