Performance guarantees

Performance guarantees

[Sukanto Ghosh]

Do you think, this area is worth exploring in KVM, or is it against the
design goals ? Shouldn't it be nice and useful to have such guarantees (to
a certain extent) ?

Also, I do want to be a good kernel/KVM developer. How can I approach this
goal ? Currently I am looking at KVM along with my other course-works.

And lastly, thanks to you all for answering my questions (which might be
stupid sometimes) after taking out time from your busy schedules.



Regards,
Sukanto Ghosh

Re: Performance guarantees

[Sukanto Ghosh]

> Do you think, this area is worth exploring in KVM, or is it against the
> design goals ? Shouldn't it be nice and useful to have such guarantees (to
> a certain extent) ?

By Performance guarantees I mean,
i) certain %age of CPU time
ii) certain %age of memory dedicated to each guest
iii) anything else ?


Regards,
Sukanto Ghosh

Re: Performance guarantees

[Andi Kleen]

"Sukanto Ghosh" <sukanto@cse.iitb.ac.in> writes:
>
> By Performance guarantees I mean,
> i) certain %age of CPU time
> ii) certain %age of memory dedicated to each guest
> iii) anything else ?

The beauty of KVM is that there's no need to actually implement
this for KVM. It can just use all the infrastructure of standard 
Linux that is available for this.

The hierarchical group scheduler and "memory cgroup" work that is being 
worked on in recent kernels should be able to do i) and ii). Partly
it could be also done using older technology like cpusets or
numa emulation or nice levels.

-Andi

Re: Performance guarantees

[Sukanto Ghosh]

> The hierarchical group scheduler and "memory cgroup" work that is being
> worked on in recent kernels should be able to do i) and ii).
>

Are these features already part of 2.6.26 or will be merged in future
releases.
Can you give me some documentation/article on them ?

By 'hierarchical group scheduler', do you mean 'Fair CPU Scheduler for
Linux' (http://fairsched.sourceforge.net/doc/fairsched.txt)




Thanks and Regards,
Sukanto Ghosh

Re: Performance guarantees

[Andi Kleen]

Sukanto Ghosh wrote:
>> The hierarchical group scheduler and "memory cgroup" work that is being
>> worked on in recent kernels should be able to do i) and ii).
>>
> 
> Are these features already part of 2.6.26 

They are, at least in a preliminary form.

> By 'hierarchical group scheduler', do you mean 'Fair CPU Scheduler for
> Linux' (http://fairsched.sourceforge.net/doc/fairsched.txt)

No.


-Andi