[Cake] ECN marking and latest cake

[Cake] ECN marking and latest cake

[Antoine Deschênes]

[-- Attachment #1: Type: text/plain, Size: 786 bytes --]

Hello everyone,

I built the latest cake yesterday night and noticed I don't see any changes
between enabling and disabling ECN on Windows with this version anymore.

(DSLreports speedtest Re-xmit avg becomes 5x lower with ECN working on the
client)

It's been a while since I didn't do a factory reset instead of using
sysupgrade on my build.

I did so after finding out that the new tc cake options didn't come up on
my router,
sounds like something didn't work right even though I never use opkg for
upgrades.

I can't pinpoint which commit broke ECN, it looks like I could have been a
few versions late on cake and iproute2 without knowing in the past weeks.

Anyhow, fq_codel still works as expected on the C7 v2, so it doesn't sound
like lost kernel options or anything.

Antoine

[-- Attachment #2: Type: text/html, Size: 1022 bytes --]

Re: [Cake] ECN marking and latest cake

[Jonathan Morton]

> On 17 Oct, 2015, at 18:45, Antoine Deschênes <antdes45@gmail.com> wrote:
> 
> I built the latest cake yesterday night and noticed I don't see any changes between enabling and disabling ECN on Windows with this version anymore. 
> 
> (DSLreports speedtest Re-xmit avg becomes 5x lower with ECN working on the client)

ECN marking still seems to work at this end.

- Jonathan Morton

Re: [Cake] ECN marking and latest cake

[Kevin Darbyshire-Bryant]

[-- Attachment #1: Type: text/plain, Size: 1427 bytes --]



On 18/10/15 08:32, Jonathan Morton wrote:
>> On 17 Oct, 2015, at 18:45, Antoine Deschênes <antdes45@gmail.com> wrote:
>>
>> I built the latest cake yesterday night and noticed I don't see any changes between enabling and disabling ECN on Windows with this version anymore. 
>>
>> (DSLreports speedtest Re-xmit avg becomes 5x lower with ECN working on the client)
> ECN marking still seems to work at this end.
>
> - Jonathan Morton

It works here too.  I see far more ECN marks & drops on my egress than
ingress; gigabit to 10mbit is quite a transition that needs to be
managed.  40mbit to gigabit ingress is less of a problem *and* my ISP
(and BT's infrastructure over which it runs) has a rate limiter to help
prevent overfilling the link.  As an experiment I artificially set the
ingress bandwidth to 1/4 actual and ran a few speed & other tests.  ECN
marks worked a treat.

Also, in terms of Openwrt/cerowrt, the versions of cake/tc-adv have been
behind those in the source repositories of sch_cake & tc-adv.  As of
approx 10:00 GMT Oct 18th 2015 (and compile/run test on Archer C7 v2) I
updated ceropackages-3.10 to point to the latest commits to those source
repositories.  Those commits are
0251bc5fa32a2caff32b108e55239bf667e851c2 &
256fbf86642f7e180d507be2c0824c03536ac140 in Dave's sch_cake & tc-adv
repos respectively.

I hope that explanation in some way helps :-)

Kevin





[-- Attachment #2: S/MIME Cryptographic Signature --]
[-- Type: application/pkcs7-signature, Size: 4816 bytes --]

Re: [Cake] ECN marking and latest cake

[Antoine Deschênes]

[-- Attachment #1: Type: text/plain, Size: 876 bytes --]

Hi Kevin,

Cake works beautifully 1 commit behind master. I didn't recompile it since
then.

In my case, broken buffers meant at least 5% drops, no SQM at all meant ~5%
and working cake about 1% drops.

Here are my last three reports:
Oct 29: http://speedtest.dslreports.com/speedtest/1795514 Working
Nov 2: http://speedtest.dslreports.com/speedtest/1832957 Lots of drops
Nov 7: http://speedtest.dslreports.com/speedtest/1880515 Working again

Antoine

2015-11-11 4:52 GMT-05:00 Kevin Darbyshire-Bryant <
kevin@darbyshire-bryant.me.uk>:
>
> Hi Antoine,
>
> Out of interest, have the recent changes to cake's buffer calculations
> made any difference to your experience?  I've noticed a decrease in the
> number of 'drops' and an increase in the number of 'marks' as a result of
> those recent changes - it's a subtle change for my case but is there.
>
> Cheers,
>
> Kevin
>
>

[-- Attachment #2: Type: text/html, Size: 1937 bytes --]

Re: [Cake] ECN marking and latest cake

[Kevin Darbyshire-Bryant]

[-- Attachment #1: Type: text/plain, Size: 775 bytes --]



On 12/11/15 01:32, Antoine Deschênes wrote:
> Hi Kevin,
>
> Cake works beautifully 1 commit behind master. I didn't recompile it
> since then.
>
> In my case, broken buffers meant at least 5% drops, no SQM at all
> meant ~5% and working cake about 1% drops. 
>
> Here are my last three reports:
> Oct 29: http://speedtest.dslreports.com/speedtest/1795514 Working 
> Nov 2: http://speedtest.dslreports.com/speedtest/1832957 Lots of drops
> Nov 7: http://speedtest.dslreports.com/speedtest/1880515 Working again
>
> Antoine
>
That's really good to hear.  I was looking back over the mail archive
and wondered if your report of ECN 'distress' was really early evidence
of the recently solved buffer limit calculation bug.  Looks like it was. 

Kevin


[-- Attachment #2: S/MIME Cryptographic Signature --]
[-- Type: application/pkcs7-signature, Size: 4816 bytes --]