Re: [ckrm-tech] [RFC 3/5] sched: Add CPU rate hard caps

[Peter Williams <pwil3058@bigpond.net.au>]

Mike Galbraith wrote:
> On Fri, 2006-06-02 at 23:18 +1000, Peter Williams wrote:
>> Mike Galbraith wrote:
>>> On Fri, 2006-06-02 at 15:55 +1000, Peter Williams wrote:
>>>> Chandra Seetharaman wrote:
>>>>> On Thu, 2006-06-01 at 14:04 +0530, Balbir Singh wrote:
>>>>>> Hi, Kirill,
>>>>>>
>>>>>> Kirill Korotaev wrote:
>>>>>>>> Do you have any documented requirements for container resource 
>>>>>>>> management?
>>>>>>>> Is there a minimum list of features and nice to have features for 
>>>>>>>> containers
>>>>>>>> as far as resource management is concerned?
>>>>>>> Sure! You can check OpenVZ project (http://openvz.org) for example of 
>>>>>>> required resource management. BTW, I must agree with other people here 
>>>>>>> who noticed that per-process resource management is really useless and 
>>>>>>> hard to use :(
>>>>> I totally agree.
>>>> "nice" seems to be doing quite nicely :-)
>>>>
>>>> To me this capping functionality is a similar functionality to that 
>>>> provided by "nice" and all that's needed to make it useful is a command 
>>>> (similar to "nice") that runs tasks with caps applied.
>>> Similar in that they are both inherited.  Very dissimilar in that the
>>> effect of nice is not altered by fork whereas the effect of a cap is.
>>>
>>> Consider make.  A cap on make itself isn't meaningful, and _any_ per
>>> task cap you put on it with the intent of managing the aggregate, is
>>> defeated by the argument -j.  Per task caps require omniscience to be
>>> effective in managing processes.  That's a pretty severe limitation.
>> These caps aren't trying to control aggregates but with suitable 
>> software they can be used to control aggregates.
> 
> How?  How would you deal with the make example with per task caps.

I'd build a resource management tool that uses task statistics, nice and 
caps to manage CPU resource allocation.  This could be a plug in kernel 
module or a user space daemon.  It doesn't need to be in the scheduler.

Peter
-- 
Peter Williams                                   pwil3058@bigpond.net.au

"Learning, n. The kind of ignorance distinguishing the studious."
  -- Ambrose Bierce