[Cake] CS1 behavior in diffserv4

[Cake] CS1 behavior in diffserv4

[Dave Taht]

I went and fixed transmission (though not cleanly) to emit CS1. It
took a while to find the right entry point, and sure enough, it was
not calling the right thing on udp.

Dumb quick patch:

http://snapon.lab.bufferbloat.net/~cero3/fixtransmission.patch

So, great, I could test what I wanted.

I have to admit I thought diffserv4 put CS1 in class 3? (haven't
looked) (seems more likely we are not matching tos properly in the
first place. Because diffserv8 is mostly class 0 also.)

root@ranger:~# tc -s qdisc show dev eth2
qdisc cake 8011: root refcnt 9 bandwidth 10Mbit diffserv4 flows raw
 Sent 316759388 bytes 245396 pkt (dropped 10653, overlimits 456732 requeues 0)
 backlog 39088b 27p requeues 0
           Class 0     Class 1     Class 2     Class 3
  rate        10Mbit    9375Kbit    7500Kbit    2500Kbit
  target       5.0ms       5.0ms       5.0ms       7.3ms
interval     100.0ms     100.0ms     100.0ms     102.3ms
Pk delay      86.8ms      20.3ms       192us       291us
Av delay      18.8ms       6.5ms         3us         4us
Sp delay       4.4ms       1.1ms         3us         4us
  pkts        219623       36390           2          61
way inds       25724         427           0           0
way miss        2719        2126           2          27
way cols           0           0           0           0
  bytes    289564164    41102703         180        7131
  drops        10577          76           0           0
  marks            0           0           0           0
qdisc ingress ffff: parent ffff:fff1 ----------------
 Sent 23422815 bytes 204407 pkt (dropped 0, overlimits 0 requeues 0)
 backlog 0b 0p requeues 0


-- 
Dave Täht
worldwide bufferbloat report:
http://www.dslreports.com/speedtest/results/bufferbloat
And:
What will it take to vastly improve wifi for everyone?
https://plus.google.com/u/0/explore/makewififast

Re: [Cake] CS1 behavior in diffserv4

[Jonathan Morton]

> On 20 Jun, 2015, at 05:19, Dave Taht <dave.taht@gmail.com> wrote:
> 
> I have to admit I thought diffserv4 put CS1 in class 3?

No - class 0 is background, 1 is best-effort, 2 is “video” and 3 is “voice”.  So it’s working just fine.

Try making some best-effort traffic compete with it, and see how the relative throughputs go.

 - Jonathan Morton

Re: [Cake] CS1 behavior in diffserv4

[Dave Taht]

yes, that clobbered the transmission stuff. I wlll argue that "class"
is a lousy name, and misleading to see the rate so....

With the correctly marked CS1 stuff, now, torrent backed way off.

root@ranger:~# watch tc -s qdisc show dev eth2
root@ranger:~# tc -s qdisc show dev eth2
qdisc cake 8014: root refcnt 9 bandwidth 10Mbit diffserv4 flows raw
 Sent 165054263 bytes 534623 pkt (dropped 7727, overlimits 1035673 requeues 0)
 backlog 49462b 26p requeues 0
           Class 0     Class 1     Class 2     Class 3
  rate        10Mbit    9375Kbit    7500Kbit    2500Kbit
  target       5.0ms       5.0ms       5.0ms       7.3ms
interval     100.0ms     100.0ms     100.0ms     102.3ms
Pk delay     215.9ms      28.0ms       445us        11us
Av delay      39.9ms       3.8ms         8us         0us
Sp delay       2.9ms       578us         8us         0us
  pkts         43184      499206           3           1
way inds         323         137           0           0
way miss        1788        1338           3           1
way cols           0           0           0           0
  bytes     51178018   122599036         270          46
  drops         6398        1329           0           0
  marks            0           0           0           0
qdisc ingress ffff: parent ffff:fff1 ----------------
 Sent 1229901568 bytes 919416 pkt (dropped 0, overlimits 0 requeues 0)
 backlog 0b 0p requeues 0

On Fri, Jun 19, 2015 at 7:40 PM, Jonathan Morton <chromatix99@gmail.com> wrote:
>
>> On 20 Jun, 2015, at 05:19, Dave Taht <dave.taht@gmail.com> wrote:
>>
>> I have to admit I thought diffserv4 put CS1 in class 3?
>
> No - class 0 is background, 1 is best-effort, 2 is “video” and 3 is “voice”.  So it’s working just fine.
>
> Try making some best-effort traffic compete with it, and see how the relative throughputs go.
>
>  - Jonathan Morton



-- 
Dave Täht
worldwide bufferbloat report:
http://www.dslreports.com/speedtest/results/bufferbloat
And:
What will it take to vastly improve wifi for everyone?
https://plus.google.com/u/0/explore/makewififast

Re: [Cake] CS1 behavior in diffserv4

[Dave Taht]

Encouraging. I guess I have to figure out how much of that was ack
traffic on the upload.

http://snapon.lab.bufferbloat.net/~d/withtorrent/36peersclassifiedcorrectly.png
(46, actually)

'course, discouraging that transmission's been getting it wrong all
this time.....

On Fri, Jun 19, 2015 at 7:50 PM, Dave Taht <dave.taht@gmail.com> wrote:
> yes, that clobbered the transmission stuff. I wlll argue that "class"
> is a lousy name, and misleading to see the rate so....
>
> With the correctly marked CS1 stuff, now, torrent backed way off.
>
> root@ranger:~# watch tc -s qdisc show dev eth2
> root@ranger:~# tc -s qdisc show dev eth2
> qdisc cake 8014: root refcnt 9 bandwidth 10Mbit diffserv4 flows raw
>  Sent 165054263 bytes 534623 pkt (dropped 7727, overlimits 1035673 requeues 0)
>  backlog 49462b 26p requeues 0
>            Class 0     Class 1     Class 2     Class 3
>   rate        10Mbit    9375Kbit    7500Kbit    2500Kbit
>   target       5.0ms       5.0ms       5.0ms       7.3ms
> interval     100.0ms     100.0ms     100.0ms     102.3ms
> Pk delay     215.9ms      28.0ms       445us        11us
> Av delay      39.9ms       3.8ms         8us         0us
> Sp delay       2.9ms       578us         8us         0us
>   pkts         43184      499206           3           1
> way inds         323         137           0           0
> way miss        1788        1338           3           1
> way cols           0           0           0           0
>   bytes     51178018   122599036         270          46
>   drops         6398        1329           0           0
>   marks            0           0           0           0
> qdisc ingress ffff: parent ffff:fff1 ----------------
>  Sent 1229901568 bytes 919416 pkt (dropped 0, overlimits 0 requeues 0)
>  backlog 0b 0p requeues 0
>
> On Fri, Jun 19, 2015 at 7:40 PM, Jonathan Morton <chromatix99@gmail.com> wrote:
>>
>>> On 20 Jun, 2015, at 05:19, Dave Taht <dave.taht@gmail.com> wrote:
>>>
>>> I have to admit I thought diffserv4 put CS1 in class 3?
>>
>> No - class 0 is background, 1 is best-effort, 2 is “video” and 3 is “voice”.  So it’s working just fine.
>>
>> Try making some best-effort traffic compete with it, and see how the relative throughputs go.
>>
>>  - Jonathan Morton
>
>
>
> --
> Dave Täht
> worldwide bufferbloat report:
> http://www.dslreports.com/speedtest/results/bufferbloat
> And:
> What will it take to vastly improve wifi for everyone?
> https://plus.google.com/u/0/explore/makewififast



-- 
Dave Täht
worldwide bufferbloat report:
http://www.dslreports.com/speedtest/results/bufferbloat
And:
What will it take to vastly improve wifi for everyone?
https://plus.google.com/u/0/explore/makewififast