FlexQoS FlexQoS 1.2.5 - Flexible QoS Enhancement Script for Adaptive QoS

[Kingp1n]

Not really. Seems rather unintelligible to me. He may know what he’s talking about, but I don’t understand it.

I’d like to see someone (besides me) adapt this script for Merlin:

routerperf/SimpleHFSCgamerscript.sh at master · dlakelan/routerperf

OpenWRT router performance data collection and analysis script - dlakelan/routerperf

github.com

I apologize @dave14305 but I don't speak/understand Greek. I wish I could be of any assistance

Maybe someone can help further

 

[raion969]

@dave14305 do you still recommend all default on the device priority tap or does it make any differenc if i say, lets make the gaming device the highest and the rest to the lowest priority?

 

[dave14305]

@dave14305 do you still recommend all default on the device priority tap or does it make any differenc if i say, lets make the gaming device the highest and the rest to the lowest priority?

I don’t really recommend any setting for device priority. I think it is a feature that has no impact on the Adaptive QoS rules. If it has any impact, I never found it. That’s why I was willing to destroy the default htb classes for devices and replace that additional set of htb qdiscs with fq_codel.

You can experiment if you like by switching from fq_codel to ASUS in the FlexQoS options and seeing if you notice any change in behavior.

 

[raion969]

so with fq_codel its deactivated in the menue or does it still changes something ^^

 

[dave14305]

so with fq_codel its deactivated in the menue or does it still changes something ^^

Changing the device priorities on Bandwidth Monitor tab would at least require you to switch the FlexQoS mode back to ASUS. I don’t know if this is the exact question you’re asking, but it’s the one I feel like answering.

 

[raion969]

jea sry for my bad english you know in austria we dont speak very often english ^^ but jea this was my question.

so if i am on the flexqos mode on fq_codel it doestn matter what priority i set in the bandwidth Monitor because it dont work with flexqos.

 

[Bensam123]

Seems to be some sort of weird bug with QoS and Steam. For some reason, even when functioning well under cap it causes all sorts of latency issues and throttling among other programs while downloading. It doesn't do this with any other downloads or uploads. I have to manually set steam to 50%, even then it causes issues among other things happening on the network. Anyone have any ideas? Origin works fine, speed tests work absolutely fine, running OBS normally works perfectly fine, but when a steam download starts it wreaks havoc on the network, even under cap. If I turn off QoS and run manually with the built in Steam throttle the network functions normally.

I want to point out I found another scenario in which this happens. It seems the google built in speed test causes issues as well, for some weird reason. Other speed tests don't seem to have a issue, but the google one causes latency issues and hiccups with QoS. This is in addition to Steam. This only happens with the download, not the upload.

 

[BikeHelmet]

I want to point out I found another scenario in which this happens. It seems the google built in speed test causes issues as well, for some weird reason. Other speed tests don't seem to have a issue, but the google one causes latency issues and hiccups with QoS. This is in addition to Steam. This only happens with the download, not the upload.

Just tossing this out there as an idea:

I noticed on my RT-AC3200, certain traffic would spike the CPU on both cores to 100%. At my new place I upgraded to an RT-AX56U, which is handling weird loads much better - but the AC3200 would have individual connections slow down, even though I could mostly still utilize my whole connection.

AC3200 - BCM4709 @ 1 GHz dual core
AX56U - BCM6755 @ 1.5 GHz, quad core
AC86U - BCM4906 @ 1.8 GHz dual core (newest arch)

In the back of my mind, I was pondering over the TrendMicro filtering engine. It's a binary blob. It has weird things like classifying the same traffic several times to different categories. We have no idea what the source code is like, but what if what started out as some engineer's lean and mean dream project, has morphed into some monstrous codebase that is impossible to maintain, and new coders are just slowly strapping things onto it as they figure out how? What if these odd lag inducing moments are parts of the code that are incredibly inefficient, and are hogging one or two cores from the router, for an undetermined amount of time? Is there a way to log how much CPU time that part of the router is taking up?

Back in the WinXP days, it was pretty common for some program to hog 100% of a core for no reason, back when programmers were new to multi-threading. Even if they have the skills, programmers frequently make mistakes, especially when they don't have a clear picture of the whole codebase or all the target devices.

If we add in that Origin might have a defined IP range and ports for downloads, and we know that Steam mixes it up a lot (the whole IP range pretty much can be for game traffic, bulk downloads, community features and communications, etc. - and individual games may even use HTTP or HTTPS for game data, rather than using their own proprietary UDP protocols, so you can't even filter downloads vs game data by watching ports and counting traffic.), then it just seems like Steam might possibly require more intense code to properly filter the connections and traffic into the desired categories.

https://www.reddit.com/r/origin/comments/497qzj

Just a theory, with no evidence to back it up, but some coding experience. What do you think? Is there a way to test/monitor/log resource usage at the sub-second level to find out what is going on? I am not the biggest linux guru, but I know there's a pile of people here with much greater experience.

This really shows the benefit of projects like Cake - lots of open source code to look at. More eyes often spots more bugs.

 

[BoostOver]

Speaking of gaming.
I'm a bit skeptical.
Are we sure that the accelerated nat doesn't lose packets?

 

[Bensam123]

I guess I should clarify where this is happening a bit more. I get this issue while streaming (utilizing my upload to Twitch). When I either have a steam download running or I use the google speed test (only when it's testing the download, the upload has no effect) the upload on my OBS stream will start acting very sporadic. Other speedtests, including speedtest.net, and other distribution platforms like Origin do not cause issues. I can even torrent and it doesn't effect the stream at all.

I have a 600/400 connection as well.

 

[dave14305]

Speaking of gaming.
I'm a bit skeptical.
Are we sure that the accelerated nat doesn't lose packets?

No one can be sure of anything related to the closed source components. Why do you ask?

 

[BoostOver]

No one can be sure of anything related to the closed source components. Why do you ask?

I don't mean much, but reading for various forums it reads this:
Hardware acceleration can be bad for gaming.

 

[Paliv]

I don't mean much, but reading for various forums it reads this:
Hardware acceleration can be bad for gaming.

I’m no pro gamer, and we don’t have a very busy network, but it’s never caused me issues.

 

[BigTim]

I've been using FlexQoS for a while and works great. Just switched to an AI Mesh setup with two AX58U units for better home coverage. Just noticed using Merlin's latest Beta code that I don't have the FlexQoS tab in the Main router. Can I install it on both mesh nodes (SSH and command line to each) or is it not compatible AI Mesh?

 

[dave14305]

I've been using FlexQoS for a while and works great. Just switched to an AI Mesh setup with two AX58U units for better home coverage. Just noticed using Merlin's latest Beta code that I don't have the FlexQoS tab in the Main router. Can I install it on both mesh nodes (SSH and command line to each) or is it not compatible AI Mesh?

It would only need to run on the node that is the router.

 

[BigTim]

It would only need to run on the node that is the router.

Thanks Dave, much appreciated.

 

[BikeHelmet]

I don't mean much, but reading for various forums it reads this:
Hardware acceleration can be bad for gaming.

Hard to say if that is the cause or the symptom. Ex: People with higher speeds like CTF / Hardware acceleration. Higher speeds reduce the need for QOS, though they typically don't fully eliminate it.

If you enable hardware acceleration, you don't enable the same QOS. Stuff like Cake can't keep up with massively fast connections. Tomato QOS has never been accelerated. Only FlexQOS is available as a nice hybrid on these routers. In the past even 100mbit+QOS was too heavy for most router hardware, yet was still within reach of having buffers burst-filled suddenly by large spurts of data from YouTube, Netflix, torrents, Steam downloads, etc..

So you got a mix of people that wanted full internet speed (200mbit comcast, etc.) and didn't use QOS because it slowed their connection down - but then could suffer from BufferBloat and ping spikes in games (when other services start bursty downloads - or services like dropbox hog all the upstream) - and people that do use the (older, less effective) QOS types, get pretty good results, but have their overall maximum speed fall due to lack of acceleration. Especially when there's tons of connections, and the router CPU can't keep up. That explains the stories and conflicting viewpoints. Hardware acceleration is bad, QOS is bad, QOS isn't needed over a certain speed, QOS is still needed over a certain speed, etc.

That situation is pretty logical though when you look at the internet speeds, router hardware, services used and types of complaints.

Now-a-days you just throw more hardware at it, turn on acceleration, turn on FlexQOS. Done. As long as you set your speed %'s low enough to allow UDP traffic some breathing room (many UDP protocols lack retransmit or throttle mechanisms and may not do well with latency volatility) then it should function as expected.

 

[dave14305]

FlexQoS - FlexQoS 1.3.2 - Flexible QoS Enhancement Script for Adaptive QoS

FlexQoS Version 1.3.2 - Released 16-Sep-2022 This is just a minor release with an iptables optimization. Nothing exciting or outwardly visible has changed. CHANGED: Split iptables rules into separate upload and download chains to avoid unnecessary rule traversal CHANGED: Tweaked 'debug'...

www.snbforums.com

 

[BoostOver]

Hard to say….

Great explanation, thank you.

A question:
I have a DSL-AC68U and i can’t install cake for software limitation.
Not having a connection that exceeds 100Megabit, could I turn off CTF and sleep quiet dreams?
Thanks

 

[BikeHelmet]

@BoostOver Apparently that router has a dual-core Broadcom BCM4708 @ 800 MHz - my RT-AC3200 has a dual-core BCM4709 @ 1000mhz. Comparable architectures, so my router is 25% faster.

I personally found that with high connection counts (torrent traffic, lots of downloads and streaming, Steam, games, web traffic, etc.), CTF being off did not allow full utilisation of my connection. I have "150/15" with Shaw Cable in BC, Canada, which is provisioned as somewhere around 180/16. I was getting a lot of dips into the 70-140 range depending on CPU load. A lot of the time my torrents were only going at 9MB/sec despite having 15MB/sec limits in qBittorrent, plus other things were not getting full speed either. (Those torrents could easily do 50MB/sec, so that's not the issue.) That's what ultimately enticed me to try FlexQOS rather than remaining on other types of QOS. That said, YMMV "Your mileage may vary" - we likely have different firmware versions and QOS types and QOS versions (since my testing was a while ago), making identical testing difficult even if you ignore that our networks and devices and use will also be different.

You'd probably just have to try it, and report back on your result - although to be useful to other people, they need to understand your network setup and use to incorporate anything into their own plans... so try to be detailed.

By the way, Dave is working on integrating Cake into the next FlexQOS... you may be able to try it out sooner than you think, if he's successful!