[Codel] [PATCH 2/2] codel: The missing fourth state

[Eric Dumazet]

On Mon, 2012-05-07 at 02:16 -0700, Dave Täht wrote:
> I'm not saying this is correct. But it would account for a quantum miss.
> 
> As for good queue vs bad queue... don't know.
> 
> In testing with 1 or 50 streams, at 100Mbit, it seems to do the job.
> 
> I look forward to reading the actual paper.
> ---
>  net/sched/sch_codel.c |   10 ++++++----
>  1 file changed, 6 insertions(+), 4 deletions(-)
> 
> diff --git a/net/sched/sch_codel.c b/net/sched/sch_codel.c
> index a9e6383..55c75de 100644
> --- a/net/sched/sch_codel.c
> +++ b/net/sched/sch_codel.c
> @@ -195,9 +195,11 @@ static u32 count_rescale(struct codel_sched_data *q, codel_time_t now) {
>  			default:
>  				c = q->count - 1;
>  		}
> -		c = max(1U, c);
> -	} 
> -	return (u32) c;
> +	} else {
> +		c = q->count >> 4;
> +		} 
> +	  c = max(1U, c);
> +	  return (u32) c;
>  }
>  
>  static struct sk_buff *codel_dequeue(struct Qdisc *sch)
> @@ -247,7 +249,7 @@ static struct sk_buff *codel_dequeue(struct Qdisc *sch)
>  		}
>  	} else if (drop &&
>  		   (codel_time_before(now - q->drop_next,
> -				      16 * q->interval) ||
> +				     16 * q->interval) ||
>  		    codel_time_after_eq(now - q->first_above_time,
>  					2 * q->interval))) {
>  		codel_drop(sch, skb);


I am refactoring the whole code to be able to include codel in other
qdiscs. Splitting data into 3 separate structures (params, vars, stats)

SFQ for example will share params,stats, but have separate 'vars' perf
flow.

by the way the last compare was wrong : it should be :

	if (codel_time_before(now - q->vars.drop_next,
				16 * q->params.interval)) {
			u32 c = q->count - 1;
	} else {
		q->count = 1;
	}