Support for CoDeL (new QoS algorithm)?

General discussion about Gargoyle, OpenWrt or anything else even remotely related to the project

Moderator: Moderators

Post Reply
raz
Posts: 41
Joined: Sun Feb 21, 2010 11:12 am

Support for CoDeL (new QoS algorithm)?

Post by raz »

Hey,

It looks like CoDel ("a new no-knobs Active Queue Management algorithm") has made it into OpenWRT trunks and in linux kernel 3.5.

User reports are positive.

Is the Gargoyle development team considering including CoDeL support in the next release?

Grey
Posts: 8
Joined: Wed Dec 12, 2012 7:16 pm

Re: Support for CoDeL (new QoS algorithm)?

Post by Grey »

I've no idea what the devs for Gargoyle have planned.

I downloaded and compiled a version of Gargoyle 1.5.8 to tinker with my own router. I've been following the bufferbloat with the CeroWRT builds for quite awhile.

Did a rebuild of the qos_gargoyle and put in fq_codel. Honestly the devs have done a good job of tweaking sfq settings. The rtt's are slightly better with fq_codel. It's fun to watch the ping times when you place huge load on the wan. Ping times will start to. climb and then rapidly drops back down to almost a no-load ping.

BQL is also in the latest builds ( linux 3.3 kernel and newer I think) and it's also helping mitigate delays.

For fun I baked in HTB with some new commits from linux 3.8 that implements a massive accuracy improvement plus using hr-timers instead of tics. Totally non-empirical but my wife commented how much better Netflix has been working.

Now I just need to figure out if I want to do some rewritting on the ACC module. It has no support for HTB and it has a hard time dealing with a wireless internet connection.

pbix
Developer
Posts: 1371
Joined: Fri Aug 21, 2009 5:09 pm

Re: Support for CoDeL (new QoS algorithm)?

Post by pbix »

Gargoyle with ACC long ago controlled bufferbloat issues on the WAN link so there will be minimum benefit to using BQL you reference there. As you noted the SFQ settings are already pretty well optimized.

ACC requires the use of HSFC, it cannot work with HTB. It would not be possible to modify it.

If you wish to further discuss issues related to QoS I suggest you post in the relevant sub-forum created for that purpose.
Linksys WRT1900ACv2
Netgear WNDR3700v2
TP Link 1043ND v3
TP-Link TL-WDR3600 v1
Buffalo WZR-HP-G300NH2
WRT54G-TM

raz
Posts: 41
Joined: Sun Feb 21, 2010 11:12 am

Re: Support for CoDeL (new QoS algorithm)?

Post by raz »

Grey,

How's the torrenting on your custom build? Stable?

Moreover, with your build, are you able to define min/max BW using Gargoyle's QoS service classes?

jorgemiguel4
Posts: 85
Joined: Thu Dec 29, 2011 2:17 pm

Re: Support for CoDeL (new QoS algorithm)?

Post by jorgemiguel4 »

Interesting, how to configure it in LuCI?

Grey
Posts: 8
Joined: Wed Dec 12, 2012 7:16 pm

Re: Support for CoDeL (new QoS algorithm)?

Post by Grey »

raz wrote:Grey,

How's the torrenting on your custom build? Stable?

Moreover, with your build, are you able to define min/max BW using Gargoyle's QoS service classes?


I haven't used torrent in ages. I had to reboot the router yesterday but that was my fault. Locked up a script in memory somehow doing the umpteenth restart.

I've been doing everything from scripts. I've went back to HFSC. Think I was biased from past work on much slower routers wrt54g verse wndr3800

Grey
Posts: 8
Joined: Wed Dec 12, 2012 7:16 pm

Re: Support for CoDeL (new QoS algorithm)?

Post by Grey »

jorgemiguel4 wrote:Interesting, how to configure it in LuCI?


Gargoyle uses it's own web pages. I've no idea how in Luci on basic OpenWrt. Web Pages tends to give me a headache to get working right.

lioux
Posts: 5
Joined: Sun May 26, 2013 3:42 pm

Re: Support for CoDeL (new QoS algorithm)?

Post by lioux »

Could you post pointers on how you added fq_codel support? I would like to try that too. :)

dtaht
Posts: 8
Joined: Sat Mar 30, 2013 5:55 pm

Re: Support for CoDeL (new QoS algorithm)?

Post by dtaht »

Incidentally, I have been doing some work with bittorrent and sfq and fq_codel and pie and a few others. The big win is to reduce the number of peers from the defaults of like (30? 60? 150?) to something a bit saner like 10-12...

Post Reply