Re: [RFC] [PATCH 3/4] Reclaim from groups over their soft limit under memory pressure

[Balbir Singh <balbir@linux.vnet.ibm.com>]

KAMEZAWA Hiroyuki wrote:
> On Wed, 13 Feb 2008 20:42:42 +0530
> Balbir Singh <balbir@linux.vnet.ibm.com> wrote:
> 
>> +	read_lock_irqsave(&mem_cgroup_sl_list_lock, flags);
>> +	while (!list_empty(&mem_cgroup_sl_exceeded_list)) {
>> +		mem = list_first_entry(&mem_cgroup_sl_exceeded_list,
>> +				struct mem_cgroup, sl_exceeded_list);
>> +		list_move(&mem->sl_exceeded_list, &reclaimed_groups);
>> +		read_unlock_irqrestore(&mem_cgroup_sl_list_lock, flags);
>> +
>> +		nr_bytes_over_sl = res_counter_sl_excess(&mem->res);
>> +		if (nr_bytes_over_sl <= 0)
>> +			goto next;
>> +		nr_pages = (nr_bytes_over_sl >> PAGE_SHIFT);
>> +		ret += try_to_free_mem_cgroup_pages(mem, gfp_mask, nr_pages,
>> +							zones);
>> +next:
>> +		read_lock_irqsave(&mem_cgroup_sl_list_lock, flags);
> 
> Hmm... 
> This is triggered by page allocation failure (fast path) in alloc_pages()
> after try_to_free_pages(). 

We trigger it prior to try_to_free_pages() in __alloc_pages()

Then, what pages should be reclaimed is
> depends on zones[]. Because nr-bytes_over_sl is counted globally, cgroup's
> pages may not be included in zones[].
> 

True, that is quite possible.

> And I think it's big workload to relclaim all excessed pages at once.
> 
> How about just reclaiming small # of pages ? like
> ==
> if (nr_bytes_over_sl <= 0)
> 	goto next;
> nr_pages = SWAP_CLUSTER_MAX;

I thought about this, but wanted to push back all groups over their soft limit
back to their soft limit quickly. I'll experiment with your suggestion and see
how the system behaves when we push back pages slowly. Thanks for the suggestion.

> ==
> 
> Regards,
> -Kame


-- 
	Warm Regards,
	Balbir Singh
	Linux Technology Center
	IBM, ISTL

--
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>