[Codel] Fwd: count and rec_inv_sqrt initialization

Kathleen Nichols nichols at pollere.com
Fri Jul 27 16:09:40 EDT 2012


Probably best to say the control law is where we are still working.

On Jul 26, 2012, at 10:12 PM, Eric Dumazet <eric.dumazet at gmail.com> wrote:

> On Thu, 2012-07-26 at 16:47 -0700, Anton Mich wrote:
>> Hi Eric,
>> 
>> Thanks for looking into this! 
>> May I also ask what is the intuition behind using this delta (which
>> is, in my understanding, equal to the number of drops between the
>> lastcount and the current count) as the new count value? I tried to
>> track down in the codel list when this changed from what's in the
>> paper, pseudocode and NS-2 to what is now in Linux but I didn't find
>> anything.
>> 
> 
> Kathleen & Van are still discussing of what should be exactly done here.
> 
> The delta idea came from them, at a moment no public discussion was
> happening (before publication, we were privately implementing codel)
> 
> Before the delta, we had following codes :
> 
> Version v1 -> v6 : (05/03/2012 -> 05/05/2012)
> 
> +                       if (now.tv64 - q->drop_next.tv64 < 
> +                               16 * q->interval.tv64) {
> +                               int c = q->count - 1;
> +                               q->count = c < 1 ? 1 : c;
> +                       } else {
> +                               q->count = 1;
> +                       }
> 
> But count was growing and was not going back to lower values
> 
> I suggested to use instead :
> 
> if (now - drop_next<  16.*interval) {
>        unsigned int c = q->count >> 1;
> 
>        count = c < 1 ? 1 : c;
> else {
>        count = 1;
> }
> 
> Then we also tried 
> 
> c = min(q->count - 1, q->count - (q->count>>4));
> 
> (Dave even had a  module param :)
> 
> static u32 count_rescale(struct codel_sched_data *q, codel_time_t now) {
>        s32 c = 1;
> 
>        if (codel_time_after(now - q->drop_next, 16 * q->interval)) {
>                switch(decrease_method) {
>                        case 3:
>                                break;
>                        case 2: /* Dumazet 2 */
>                                c = q->count >> 1;
>                                break;
>                        case 1: /* Dumazet 1 */
>                                c = min(q->count - 1,
>                                        q->count - (q->count >> 4));
>                                break;
>                        case 0: /* Codel Paper Default */
>                        default:
>                                c = q->count - 1;
>                }
>                c = max(1U, c);
>        }
>        return (u32) c;
> }
> 
> in v13 ( 05/10/2012) we had the current form
> 
> But codel is not finalized ;)
> 
> 
> 
> _______________________________________________
> Codel mailing list
> Codel at lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/codel



More information about the Codel mailing list