Re: [RFC, PATCH] CLONE_NEWIPC and exit_group()

[Doug Ledford]

On 07/12/2012 02:54 PM, Serge Hallyn wrote:
> Quoting Kirill A. Shutemov (kirill.shutemov@xxxxxxxxxxxxxxx):
>> On Wed, Jul 11, 2012 at 03:24:22PM -0700, Andrew Morton wrote:
>>> Am I reading that right?  1000 forks take 33 seconds, with basically
>>> all of it just sitting there asleep?  This look quite terrible - what
>>> causes this?
>>
>> It seems free_nsproxy() + synchronize_rcu() are too heavy to be in
>> exit_group() path. Patch below helps: 8s -> ~0.5s for me.
> 
> And sys time goes down by that much too, or only user time?
> 
> Given that, with user namespaces, it'll soon be possible for users who
> are unprivileged toward the host to be able to create and destroy
> namespaces, if the patch ends up making it easy for a user to consume a
> bunch of system time and not have it accounted at all to himself, then
> I think we should keep it as is.

Indeed.


-- 
Doug Ledford <dledford@xxxxxxxxxx>
              GPG KeyID: 0E572FDD
	      http://people.redhat.com/dledford

Attachment: signature.asc
Description: OpenPGP digital signature