[Codel] [PATCH v5] pkt_sched: codel: Controlled Delay AQM

[dave taht]

On 05/05/2012 02:20 PM, Eric Dumazet wrote:
> On Sat, 2012-05-05 at 14:12 -0700, dave taht wrote:
>> On 05/05/2012 02:11 PM, Eric Dumazet wrote:
>>> On Sat, 2012-05-05 at 22:36 +0200, Eric Dumazet wrote:
>>>> On Sat, 2012-05-05 at 22:20 +0200, Eric Dumazet wrote:
>>>>
>>>>> I believe we should allow the last packet to be sent even if
>>>>> sch->qstats.backlog>= q->minbytes
>>>>>
>>>>> Hmm... this means we should do the
>>>>> sch->qstats.backlog -= qdisc_pkt_len(skb);
>>>>> right after the calls to __skb_dequeue(&sch->q);
>>>>>
>>>>> (and not in the codel_drop() or at end of codel_dequeue())
>>>>>
>>>>>
>>>> I am also adding a dump_stats capability, so I'll resend a v6
>>>>
>>>> (So that we can check runtime param like 'q->count' and other
>>>> interesting stuff)
>>>>
>>>>
>>> I believe we can remove the cache of 64 values, since q->count is way
>>> bigger most if the time, according to my results.
>>>
>>>
>>>
>> I went the other way and made it be 1000. How much bigger?
>>
> We can compute the thing in less time than a cache line miss.

Maybe on your arch, but highly doubtful on a 680Mhz mips that isn't even 
superscalar.

I'd prefer to leave it in and be able to compile it out, and actually 
measure the difference.

>
> So I removed the cache, it makes code simpler.

Well, it was fun while it lasted.
>
>