CAKE w/ Adaptive Bandwidth [October 2021 to September 2022]

On vpns I feel compelled to lecture. One thing that is not well understood is that tcp's reaction to congestion is quadratic relative to the delay. it reduces its rate by 1/2 (reno) or 1/3 (cubic) once per round trip. VPN's inflate the round trip and unless they are actually at the bottleneck and site-tosite cake or fq_codel link where the fq can kick in, end up in a single queue across the whole network, which takes more time for codel portion of the algorithm to kick in at an effective drop or mark rate.

So I would very much avoid tunnelling measurement packets. Userspace vpns have a tendency to bottleneck in userspace at high rates, also.

Lastly:

I have longed for someone to experiment with the RFC3168 response to a ecn mark, instead measuring RTT inflation in the last 100ms and dropping the paced rate to that. This is way less a change to tcp than L4S or SCE are, and yet I feel will work well in a fq+AQM'd environment. More broadly, the autorate code is probably sensitive to the characteristics of the predominant tcp on the link and you might want to worry about what happens if it's BBR rather than cubic. I have also been trying to drag some starlink folk into this testing circle.

That document presents a pretty grim picture about Starlink's feasibility for latency sensitive applications does it not? Or is there still hope?

I am feeling a little smug about my 4G LTE now, which manages at least 25 Mbit/s on download and 30 Mbit/s on upload without significant latency jumps (RTT around 50ms). Anything above this is subject to bufferbloat depending on the available additional variable capacity. But 25 Mbit/s seems plenty for day to day work and personal use.

I think most ISPs have pretty grim latency situation without some queue management on a customer's router. Yes, they've made latency sensitivity into a concern on DOCSIS 3.1 but for the most part most ISPs seem to care about one stat only: download bandwidth, and that's because the public only understands that one stat.

2 Likes

Yes true - I have a friend who has spent a lot of money upgrading his fibre connection to higher bandwidth packages and he doesn't get why his Zoom sessions keep freezing up whilst his kids are downloading files and gaming. He thinks increasing bandwidth should solve the problem, but it only means that he pays more.

But my query was more about whether Starlink can be made to work well from a latency perspective at a reasonable bandwidth or whether there is an inherent limitation that may preclude getting a healthy bandwidth with low latency.

Seems to show that by choosing the minimum observed available bandwidth throughout the day and setting cake to that speed it works fine... but doesn't benefit from the variable speeds we're enabling in this thread. So I don't think it's an inherently impossible problem, just similar to your LTE or a WISP provider.

Thank you .. i have open thread Help me figure out why I have gaming latency

@dtaht we ended up choosing Mozilla Public License Version 2.0 (MPLv2) as a good 'weak copyleft' license.

We've done a big re-organisation and re-write of the introduction & documentation. Any errors left are mine, after the team removed the worst of my hype and nonsense :rofl:

1 Like

Just wanted to chime in to say thank you and that I am really impressed that this project exists. Something I wanted for years when I was stuck on a 4G connection. With 5G and starlink replacing fixed lines in many areas, this is more important than ever. Will look forward to testing!

2 Likes

I also want to thank you for this script. It has been running for two weeks now and it has really made my LTE connection much better! I am just wondering why this topic became so quiet suddenly. It has been the most active topic in this forum, but suddenly, a week ago, it went quiet. What happened?

Real life happened to several of the prolific contributors to the lua threads script. But there's some goodies bubbling up and should arrive 'soon'.

I'm working on a rate control plugin that automatically tunes the maximum delay threshold.

To make this happen, I have to build the plugin infrastructure to support it, which is taking more time than building the plugin. I'm hoping that, when it arrives, smarter people than me (or with more domain experience) can build much more interesting plugins that can use or even replace the current rate control mechanisms, without having to worry about all the 'engineering'.

The plugin that I built and am 'dogfooding' is very, very simplistic with less than 250 lines of code (excluding license boilerplate), see

PS. some of the comments are wrong or misleading :rofl: -- hopefully now fixed

2 Likes

While I appreciate all your enthusiasm and progress on this project, I wanted to point out that if you intend "world domination", a few things might help. One is, funding. NLNET has a grant process closing feb 1 here: https://nlnet.nl/news/2021/20211201-call.html

In addition to paying yourselves something, acquiring and getting more hardware tested (notably any openwrt device that has a native LTE interface available would be good), a logo, t-shirts, and travel to various conferences could be covered in this way. Please try to plan for and worry a little about this work becoming a "success disaster" in the coming months.

https://innovationfund.comcast.com/ has also been a good source for me. They use openwrt (and have LTE fallback) in multiple devices.

I hope there are many more funders in the eu?

Officially I'm on https://nlnet.nl/project/CeroWRT-II/ and merely kibitzing here, (yay! I get to do some technical work on wtbb!), but I've also been heavily involved in trying to steer more funding and support to open source engineering in the USA as part of NTIA's broadband grants programs, connecting geeks to governments. In addition to helping write this official bitag report on latency:

(which I hope gains a lot more policymaker readers)

I'm working on a NTIA filing which - believe me - is a lot less fun that being in the cheering section here.

1 Like

UPDATE:
T-Mobile 5G NR signal is non-existent inside my apartment and without good 5G signal, T-Mobile Home Internet (TMHI) is not good. Also the TMHI provided Gateways (Nokia 5G or the newer Arcadyan KVD21) seem to have their own set of problems (overheating, MTU issues, issue with work VPNs etc.).

Therefore, I decided to stay with Comcast Xfinity itself. I also got a good Xfinity deal ($20 / month for 50 Mbps download, 10 Mbps upload, 1 year contract) compared to T-Mobile Home Internet ($50 / month).

Even though my current Xfinity plan speed (50 / 20 Mbps) is lower compared to my previous Xfinity plan (600 / 20 Mbps for $50 / month), it is ok for me since I am the only user, and it is consistent speed compared to TMHI which varies a lot.

lol lucky me why?

want to use but cant install any help would be appreciated...

@_FailSafe do you have any idea what this issue is with the name of OpenWrt?

@Reeves0724 are you running on an official OpenWrt build or a vendor modified version?

My friend makes snapshots for us... I guess i have to wait for him to make a new image... Is there anyway around it if I may humbly ask?Tried changing things but stays OpenWRT

I installed as a service but want to try new version lol

My guess is if you change DISTRIB_ID='OpenWrt' it might just work.

1 Like

Tried that before i bothered you guys bro... TDS was supposed to install "gig" ftp about a week ago unfortunately I live in Wisconsin where it stay -30 degrees until the end of Feb they said the conduit is frozen... Docsis is just not consistent but with autorate-sqm it really helps in my opinion so until I get fiber need this to work!!!!

So the script is checking the file /etc/os-release for the field NAME so look at that file.

1 Like

ok changing here and will try again... tks bro

thanks DL!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

Time for fresh install and testing!!!!!!

1 Like