Re: [PATCH 5/6 v4] cfq-iosched: CFQ group hierarchical scheduling and use_hierarchy interface

[Gui Jianfeng <guijianfeng@cn.fujitsu.com>]

Vivek Goyal wrote:
> On Wed, Feb 16, 2011 at 09:44:39AM +0800, Gui Jianfeng wrote:
>> Vivek Goyal wrote:
>>> On Tue, Feb 15, 2011 at 10:38:32AM +0800, Gui Jianfeng wrote:
>>>> Vivek Goyal wrote:
>>>>> On Sat, Feb 12, 2011 at 10:21:47AM +0800, Gui Jianfeng wrote:
>>>>> [..]
>>>>>>>> +static struct cfq_group *
>>>>>>>> +cfq_find_alloc_cfqg(struct cfq_data *cfqd, struct cgroup *cgroup, int create)
>>>>>>>> +{
>>>>>>>> +	struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup);
>>>>>>>> +	struct cfq_group *cfqg = NULL;
>>>>>>>> +	void *key = cfqd;
>>>>>>>> +	struct backing_dev_info *bdi = &cfqd->queue->backing_dev_info;
>>>>>>>> +	unsigned int major, minor;
>>>>>>>> +
>>>>>>>> +	cfqg = cfqg_of_blkg(blkiocg_lookup_group(blkcg, key));
>>>>>>>> +	if (cfqg && !cfqg->blkg.dev && bdi->dev && dev_name(bdi->dev)) {
>>>>>>>> +		sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor);
>>>>>>>> +		cfqg->blkg.dev = MKDEV(major, minor);
>>>>>>>> +		goto done;
>>>>>>>> +	}
>>>>>>> Should we make this updation of this info hierarhical?
>>>>>> IMHO, it's fine to defer the updation when we really get the cfqg.
>>>>> But if cfqg is alrady present, we will never hit the allocation path 
>>>>> again. So if somebody creates 2-3 level deep hierarchy and does IO
>>>>> only in the children cgroup, parent cgroups will potentially not get
>>>>> device info updated hence no visible stats?
>>>> Ahh, I see your concern. But do we really need to show the stats even if
>>>> a cgroup doesn't issue any IO on a given device? 
>>> I am assuming that once use_hierarchy=1, you are aggregating the stats
>>> in parent cgroups? So if a child services 5 IOs, these are accounted
>>> to parent group also when user_hier=1?
>>>
>>> What happens in case of  memoy cgroup controller?
>> Hmm, it seems memcg aggregating stats in parent group.
>> But do we really need to do that in kernel? I think it's easier to do it in
>> userland, and it makes kernel much simpler.
> 
> I think at some point of time hierarchical aggregated stats will also be
> required. I am also looking at "memory.stat" file of meomory controller
> and they seem to be reporting both aggregated as well as individual group
> stats.
> 
> So we can probably skip implementing hierarhical stats in this patchset
> and implement it on a need basis in future.

Ok, I agree.

Thanks,
Gui

> 
> Thanks
> Vivek
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
>