Dave Taht’s make-wifi-fast effort is seeing some amazing progress. Are you tracking it? Would we be likely to see those changes in the Omnia anytime soon?
Put another way, how closely do you track the upstream kernel?
more to the point would probably: which upstream?
recent contributions from dave’s camp, especially the cake qdisc (which is a HUGE improvement over htb+fq_codel) have only landed in lede afaik.
We’re getting there, in lede and in linux mainline. The last major bugs have been solved, but we still have a few issues left to resolve and a few features left to add. It has long been my hope that the improvements to wifi on the ath9k and ath10k (which I think are the chipsets used in the turris) will land there quickly - but I have no problem with turris aiming for reliability first and performance enhancements second. Right now the ath9k is the most tested, and we have to go back and start over again on the ath10k to some extent.
I note that cake could use a bit more love (performance tuning on this cpu, better analysis of the aqm component), that the BQL stuff landing for this cpu is nice also,
And that I’m overall rooting for the turris omnia project to get all this stuff soon after it’s first release, if not sooner, and I can’t wait to get one of my own to play with.
Some references:
BQL for the mvneta: https://lkml.org/lkml/2016/9/13/55
and the bug that we just lost 4 months to finally beating: http://blog.cerowrt.org/post/crypto_fq_bug/
which links back to the results we got on the ath10k and ath9k.
3 Likes
It’s quite reassuring to know you’re also getting one of these “toys”, Dave. 
I’m really looking forward to seeing the Omnia as a first-class citizen within the Linux kernel, as far as upstream support goes (device tree, drivers, etc.), since I ordered two of them (I’ve also been entertaining the idea of getting a GPON SFP module and replacing my ISP’s ONT, but I’m not sure if that’s even feasible).
I was terrified (and still am a bit, in spite of recent developments) about ath10k and its closed firmware blob, since I watched Felix Fietkau’s “Freedom Considered Harmful?” presentation. I really hope the bufferbloat issue can be solved on these chips, otherwise I’ll have to wait for an MT76xx-based solution. Do you have any idea if there are simultaneous dual band 802.11ac cards available on the market? It would be great to use a single PCIe slot for Wi-Fi, instead of two… 
Very excited to hear that you’re getting a Turris, I’m excited to test out the changes and to thank you for all the work you’ve put into making things better for wifi (and networking in general) over the last few years!
Regrettably I never got around to getting a turris. Hopefully the make-wifi-fast stuff worked out for y’all?
I can’t really tell, to be honest, @Dave_Taht. I never felt any big latency spikes on my 130/8 Mb/s connection, but I have simple.qos with the fq_codel qdisc enabled, just in case. Additionally, all my machines have ECN and TCP BBR congestion control enabled (advantages of using real operating systems ). I’ve been thinking about limiting my WAN bandwidth in order to do some more extensive testing, but haven’t done so yet. Still, you should get one of these babies (I have two 2 GiB units, actually), they’re really great. 
I did try to stress that the bulk of the problem is solved by having good queue management on the isp link. The wifi fixes primarily apply to when multiple stations are contending for full bandwidth, or there is a slow station on the link, contending.
Certainly I notice the make-wifi-fast work when doing videoconferencing on a wifi link that is doing other stuff, but most of the time, it’s just a pretty subtle, relaxing effect.
(I REALLY notice when I’m on a non-fq-codeled isp link). Anyway, the paper’s out now:
https://arxiv.org/abs/1703.00064
I don’t plan on getting a turris anymore - I have something like 30+ spare routers at the moment.
Maybe I’m just lucky and my ISPs (NOS at 130/8 Mb/s and Vodafone at 200/20 Mb/s, in Portugal) are clued enough about bufferbloat.
Ah, indeed. I usually don’t do “serious” work (or even much testing, to be honest) over Wi-Fi because I know what a train wreck it can be on anything but ath9k (thanks to you guys), though I see ath10k making some progress as well. By the way, any news on Minstrel Blues? 
[quote](I REALLY notice when I’m on a non-fq-codeled isp link). Anyway, the paper’s out now:
https://arxiv.org/abs/1703.00064[/quote]
Juicy! Will dig in as soon as I have some free time.
Yeah, I imagine you have a lot on your plate, all the time. Anyway, thanks a lot for your continuing work on bufferbloat and making Wi-Fi “great again”! 