Re: [Xen-devel] [Patch 2/2]xen/sched_credit2.c : Runqueue per core

[Uma Sharma <uma.sharma523@xxxxxxxxx>]

Thank you :-)

I will work on the things you mentioned and resend the patch.
It's great to work on patches. I was trying to figure out how to
change the code so it looks neat and now I have the answer. Thank you.
:-)

I will summarize the performance in cover patch.

Regards,
Uma Sharma

On Mon, Mar 9, 2015 at 6:33 PM, Dario Faggioli
<dario.faggioli@xxxxxxxxxx> wrote:
> On Mon, 2015-03-09 at 12:18 +0000, George Dunlap wrote:
>> On Mon, Mar 9, 2015 at 8:55 AM, Uma Sharma <uma.sharma523@xxxxxxxxx> wrote:
>
>> > --- a/xen/common/sched_credit2.c
>> > +++ b/xen/common/sched_credit2.c
>
>> > @@ -1935,15 +1938,36 @@ static void init_pcpu(const struct scheduler *ops, 
>> > int cpu)
>> >          return;
>> >      }
>> >
>> > +    /*Figure out which type of runqueue are to be created */
>> > +    if (!strcmp(opt_credit2_runquque, "socket")) {
>> > +        rq = 's';
>> > +    } else if (!strcmp(opt_credit2_runquque, "core")) {
>> > +        rq = 'c';
>> > +    } else {
>> > +        rq = 's';
>> > +    }
>>
>
>> It would be more typical, rather than have this be a char resolving to
>> 's' and 'c', to have it be an int, and have the values be #defines;
>> for example, "CREDIT2_OPT_RUNQUEUE_CORE" and
>> "CREDIT2_OPT_RUNQUEUE_SOCKET".
>>
> I was about to suggest the same.
>
>> Also, given that your experiments show 'core' to work quite a bit
>> better than 'socket', I'd suggest making it default to core rather
>> than socket. :-)
>>
> +1.
>
> Of course, as I said already, you should explain and provide the numbers
> about this performance improvement in the cover letter of the series
> and, IMO, reference that in the changelog of this patch too (not putting
> the full results, but a quick summary of them would be good).
>
> Regards,
> Dario



-- 
Regards,
Uma Sharma
http://about.me/umasharma

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel