Re: [PATCH -mm 0/5] towards I/O aware memory cgroup v3.

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

* KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> [2010-08-02 19:11:13]:

> 
> This is v3. removed terrble garbages from v2 and tested.(no big changes)
> 
> Now, it's merge-window and I'll have to maintain this in my box for a while.
> I'll continue to update this. Maybe we can make new progress after LinuxCon.
> (And I'll be busy for a while.)
>


I was catching up with my inbox, did not realize you had moved onto v3
and hence reviewed v1 first.
 
> This set has 2+1 purposes.
>  1. re-desgin struct page_cgroup and makes room for blocckio-cgroup ID.
>  2. implement quick updating method for memcg's file stat.
>  3. optionally? use spin_lock instead of bit_spinlock.
> 
> Plans after this.
> 
>  1. check influence of Mel's new writeback method.
>     I think we'll see OOM easier. IIUC, memory cgroup needs a thread like kswapd
>     to do background writeback or low-high watermark.
>     (By this, we can control priority of background writeout thread priority
>      by CFS. This is very good.)

Agreed, background watermark based reclaim is something we should look
at.

> 
>  2. implementing dirty_ratio.
>     Now, Greg Thelen is working on. One of biggest problems of previous trial was
>     update cost of status. I think this patch set can reduce it.

That is good news

> 
>  3. record blockio cgroup's ID.
>     Ikeda posted one. IIUC, it requires some consideration on (swapin)readahead
>     for assigning IDs. But it seemed to be good in general.
> 
> Importance is in this order in my mind. But all aboves can be done in parallel.
> 
> Beyond that, some guys has problem with file-cache-control. If it need to use
> account migration, we have to take care of races.
> 

-- 
	Three Cheers,
	Balbir

--
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>