[Solved] Getting bad bufferbloat, SQM only makes it WORSE - reverted to stock firmware

I do not have the liberty to use my own modem. The ISP will not give the credentials.

uhm.. i could not live with that ISP. I dont know what you could do else...
In my country they have to give you the credentials and you can always use your own hardware but they won't give you any support for your own hardware, which is understandable.

I feel kinda sorry for you...

Sigh...yeah I don't have that luxury. I think I'm probably just going to end up reinstalling the factory firmware that doesn't have firewalls.

Imho it should be your right to use your own hardware but i think that differs from country to country (or from ISP to ISP).
And the factory firmware worked better for you in terms of bufferbloat/QoS ? Somehow I cant believe it...

If you ever change your ISP make sure they will give away your internet credentials and allow you to use your own modem. :wink:

Why are you thinking that it's a requirement to put the ISP router in bridge
mode? you always have to deal with an upstream router that you don't control and
that can have bufferbloat problems, it doesn't matter if it's the one in your
house or the one at the central office on the other end of the wire.

It just means that you may need to sacrafice a little more bandwidth to get good
latency.

The poster has had settings that greatly improved his latency under load (the
bufferbloat). Is that improvement not sufficient? or what is the remaining
problem that needs to be solved?

David Lang

Actually, that is not quite accurate.

See the Charter Spectrum Authorized Devices policy...

This was your best test result and the config you used...

Roll with that for a while and see how it goes.

Okay, so my trawl though the dslreports forum did not reveal a smoking gun that would convince e that the speedtest is consistently flawed; though there is enough to show that the test should also not be blindly trusted. Personally I am still of the opinion, tat this test is still the best "standard-type" speedtest that I will recommend for people starting to look into bufferbloat issues, with the caveat that measurement should be repeated before being accepted. Thanks for the pointer.

Please, post links to the detailed results page (you get there by clicking the "Results + Share" button)
Also please repeat the wireless test while running "top -d 1" on the router and monitor the numbers in the header lines:
Example
Mem: 42284K used, 17760K free, 896K shrd, 3792K buff, 7240K cached
CPU: 0% usr 0% sys 0% nic 97% idle 0% io 0% irq 1% sirq
Load average: 0.01 0.01 0.00 1/69 24934
PID PPID USER STAT VSZ %VSZ %CPU COMMAND
Especially, look at idle and sirq, if idle goes to 0 your router runs out of CPU cycles which is always bad, and in many of these cases sirq will go "through the roof" showing that it is the kernel's soft-interrupt? that is working too hard (this is exercised both by traffic shapers and wifi drivers).
The goal is to figure out why on wifi your results are so much worse than in a wired configuration:

in both bandwidth and bufferbloat ratings.

It is worth mentioning again, that once your wifi connection is slower than your internet access you will see bufferbloat internet to your home network and the shaper on the wan interface can never fix this.

You could try to test this hypothesis (wifi is the bottleneck) by setting the shaper to say 20000/2500 (or 18000/1800) and test via wifi, if the bufferbloat is gone you have pin pointed your "pain-point" to your wifi set-up. In that case it makes sense thinking about optimizing that, but first confirm that this is the root cause of your observed undesired link behavior.

So, running a double NAT set-up certainly is not ideal, but the biggest issues with double NAT are affecting the ability to actually reach internal machines from the outside and should be mostly independent of the bufferbloat/induced latency under load issue that caused this thread.
I agree that getting rid of any not strictly unavoidable layer of NAT is a goal to aim for, but to some degree orthogonal to the bufferbloat issue.
So keep working on getting the modem in bridge mode, but do not let this stop you from tryng to improve your home network.

So if running a cable is not an option, maybe you could use a pair of power-line communication devices (https://en.wikipedia.org/wiki/Power-line_communication) to get a wifi access point closer to the computer (hook up the PLC closer to the router with an ethernet cable); heck this might even allow you to run a cable from your computer to the second PLC device.
Now, PLCs often introduce addition RF-noise that can be especially devastating for VDSL links, but I would hope your DOCSIS link should be mostly immune to this.

Possibly because wireless is turned on on both the cable modem and the router...

Now I am confused, why should that be a problem per se? As long as there is no traffic passed that wifi should be mostly silent, modulo a few beacons, no?
Now, maybe you are talking about RF "pollution", and that is also a decent hypothesis;
@Kampfkarren could you try to run a wifi scanner (like free android wifianalyzr app) to see how the channel situation is around your home?

Best Regards

The OP can't fix RF pollution if the cable modem is not accessible.

IMO, the fix is to get a device you can control...

https://www.amazon.com/dp/B00AJHDZSI/?tag=aboutcom02lifewire-20&ascsubtag=4082523|google.com|||66%2C21%2C83%2C49%2C27%2C3%2C33%2C8%2C79|1|

Approved up to 100 Mbps (OP needs to find out what tier he's on) -

??? First, let's see whether this is the root cause or not, maybe switching his router to a different channel can already help, but for that we would need to see information about his RF environment. I guess until @Kampfkarren re-joins the thread I will stay quiet...

If nothing is connecting to the cable modem, the amount of RF pollution is
minimal, "I exist" broadcasts are small.

David Lang

Which it appears, is never the case...

Okay, but that is easy to tests, simply wrap the ISs modem router in aluminum foil to effectively disable its RF output (if need be be ground the foil). That should certainly silence that sucker. But from my reading it is not at all clear, that the other users went around the OP's wrt1900ACS, he simply stated that there was traffic he could not "silence"...

other devices using the Internet != other devices connecting the the cable modem
supplied wifi.

but even if there are people connecting to the cable modem wifi, all you have to
do is set the LEDE to a different (suitable) channel, just like you would to
avoid an AP in the apartment next door.

It still doesn't result in a 'nothing you can do until you replace the
cablemodem' situation.

Don't see where that was stated.