Bufferbloat issue

[lioux]

I'm using latest Gargoyle on a TP-Link TL-WDR4300 with QoS upload/download and ACC.

I replaced a Tomato Linksys E4200 yesterday with it and my network was saved. Gargoyle has improved EVERYTHING.

I also tested on another network which previously had a Tomato WNDR3500Lv2. Huge improvements.

I've had good results on both Linux workstations and Android running fq_codel. I would like to extend the test bed to the router.

Has there been any change on CODEL situation? I would like to try both fq_codel and sfq_codel to see if it can improve things a little further.

I'm not Linux kernel savvy but I'm not afraid to dangle my feed on source/git. Just let me know and I'm your guinea pig.

[dtaht]

netanylzer actually doesn't detect the presence of fq_codel. It sends a single stream to load up the link, and not 2. codel takes it's time to react to that flow (longer than the netanalyzer test runs), and any new flow will just bypass the congested queue, which is where we win.

We've bugged the developers of the product for a fix for 2 years now.. (it doesn't detect SFQ or SQF (shortest queue first, common on DSL), but they lack funding.

A better test of fq_codel is some sort of high rate upload or download along with a ping, such as what the rrul test suite uses.

You really don't care - if you have 1024 queues, if one is filled up for a while, so long as the other queues get equal service.

[dtaht]

Grump. Thought I replied to this. netanalyzer doesn't correctly detect fq_codel and doesn't run for long enough for the codel portion to kick in.