[Cerowrt-devel] htb has internal pfifo queue

[Dave Taht]

OH, thank you!

One of the things that irked me on my htb+netem delay/loss setup was
that the need to also feed arp and ND through it resulted in major
startup transients that wouldn't exist in the real world.

So now I understand how to feed those protocols through a separate
unshaped htb queue.

thankyou, thankyou,thankyou

On Sun, Jun 16, 2013 at 8:54 PM, Eric Dumazet <edumazet at google.com> wrote:
> This 'direct' queue is only used if one of your filter explicitly
> targets the '0' classid of htb qdisc as in :
>
> (Note the 'direct_packets_stat 1145408' in tc qdisc output)
>
>
>
> lpq83:~# cat fq_codel0.sh
> DEV=eth4
> CEIL="ceil 10000Mbit"
> EST="est 1sec 4sec"
>
> tc qdisc del dev $DEV root 2>/dev/null
>
> tc qdisc add dev $DEV root handle 1: est 1sec 8sec htb default 1
> tc class add dev $DEV parent 1: classid 1:1 $EST htb rate 1Gbit $CEIL
> mtu 1500 quantum 1514
> tc class add dev $DEV parent 1: classid 1:2 $EST htb rate 1Gbit $CEIL
> mtu 1500 quantum 1514
>
> # access to direct queue : following trafic doesnt need overhead of an
> htb_class,
> # as we do not want it to be shaped at all.
> tc filter add dev $DEV parent 1: protocol ip u32 \
> match ip dst 7.7.7.84 flowid 1:0
>
> lpq83:~# ./netperf -H 7.7.7.84;tc -s qd sh dev eth4
> MIGRATED TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to
> 7.7.7.84 () port 0 AF_INET
> Recv   Send    Send
> Socket Socket  Message  Elapsed
> Size   Size    Size     Time     Throughput
> bytes  bytes   bytes    secs.    10^6bits/sec
>
>  87380  16384  16384    10.00    18390.46
> qdisc htb 1: root refcnt 257 r2q 10 default 1 direct_packets_stat 1145408
>  Sent 67913929786 bytes 46865570 pkt (dropped 0, overlimits 0 requeues 11)
>  rate 13754Mbit 1186190pps backlog 0b 0p requeues 11
>
> On Sun, Jun 16, 2013 at 2:32 PM, Dave Taht <dave.taht at gmail.com> wrote:
>> In looking over the htb mess, I noticed that htb has an internal pfifo
>> queue, which is controllable via new stuff in net-next.
>>
>> http://snapon.lab.bufferbloat.net/~cero2/htb_mess/11256-htb-add-HTB_DIRECT_QLEN-attribute.patch
>>
>> I have no idea what, if anything, it affects. I was under the
>> impression, until now, that htb buffered up one, and only one packet
>> no matter how many classes you have. Still am retaining that
>> impression...
>>
>> Ah well... fiddling with setting txqueuelen to 2 on  the device on
>> 3.8.13-7 before running the aqm script might have an interesting
>> result on simple.qos, if anyone wants to try it.
>>
>> such a backlog of stuff to sort through!
>>
>> --
>> Dave Täht
>>
>> Fixing bufferbloat with cerowrt: http://www.teklibre.com/cerowrt/subscribe.html



-- 
Dave Täht

Fixing bufferbloat with cerowrt: http://www.teklibre.com/cerowrt/subscribe.html