Poor video conferencing performance

[gintek]

I've been up and running with OpnSense for a couple weeks now, but since moving to it, my wife and I have been encountering poor performance with our video conferencing. We both work from home--she uses Google Meet and I use MS Teams.

I did some research, and found tips pointing to potential bufferbloat issues. Sure enough, I ran the DSLReports speed test and consistently got "Cs" for the bufferbloat rating. I went down a rabbit hole of tweaking FQ-CoDel according to the following guides: 1 and 2

Unfortunately, turning on the traffic shaping as instructed, and tweaking the quantum and limit values produce zero effect on my test results, so I'm hoping for help with two questions:

1) Am I barking up the wrong tree with the whole traffic shaping/FQ-CoDel approach for our video conferencing issues?
2) If I am on the right track, is there a likely reason that all of my tweaking is resulting in no changes to the test results?

Thanks a million for any help!

[RamSense]

Dear Gintek,

I did the same as you and used your guide #2.
Did you test afterwards at DSLreports? I went to a+, A (bufferbloat) and A+.

If you have increased results at DSLreports, you did it correcly.
Did you hit [apply] on the tabs in the shaper at opnsense? Maybe a fresh reboot (just in case?)
or otherwise
Maybe it was not at your end and at MS Teams?

[hushcoden]

Same here, no improvement, before and after the Shaper (guide no. 2) bufferbloat always at B

[gintek]

Thank you both for your replies!

To quickly confirm, I did make sure to apply and re-test at DSLReports after each setting change. Unfortunately, I wasn't getting any change in results.

But I do have an update! I'm on a 30mbps up/30mbps down fiber connection. At the time of my original post, I'd tweaked settings over and over using bandwidth settings of 29-30mbps and FQ-CoDel quantum settings ranging from 100-1000. Eventually I figured why not try something drastic, and I set the pipe to 25mbps. That finally made a difference in my DSLReports results.

From there I recommenced fiddling with things until I ended up with 26mbps for bandwidth and an FQ-CoDel quantum of 200 for up and down (everything else set at default values). That got me to a B in bufferbloat, and an A for quality.

Most importantly, my wife patted me on the head and told me I'm a good boy because she can now use Google Meet with no problems.    My Teams meetings today are also working much better.

Hopefully this info will come in handy for others in the future!

Of course, if anyone has additional advice on things I can try tweaking to get bufferbloat to an A, I'd be very grateful to hear it.

[muchacha_grande]

Hi,
I did what is pointed on this thread and it worked fine for me

https://forum.opnsense.org/index.php?topic=7423.0

It's an old guide but I actually get A or A+ in all tests

Cheers

[gintek]

Thanks!

That was the first guide I went through. At this point, since I'm still getting a B, I'm wondering if it may be an issue with my ISP (Ziply Fiber). But I'll keep playing with things to see if I can get some further improvements.

[hushcoden]

Would someone explain what FQ-CoDel Quantum is and how to determine the best value ?

Tia.

[Supermule]

The reason is a misconfigured network device on OPNSense.

Dont use Jumbopackets on the NICS doing traffic and set your buffer on the NICS to medium or small.

Make sure that when you test for bufferbloat, that the hardware can keep up with the traffic as well.

On pfsense it makes a hugge difference in the test results running Suricata Legacy mode vs Inline IPS. The troughput on server grade hardware is a factor 5 in Legacy modes favour.

[gintek]

Hey hushcoden, according to this document, the quantum sets "the maximum amount of bytes to be dequeued from a queue at once." My simplistic interpretation of that boils down to "when the traffic starts to jam, how much traffic you pull off the road to get it flowing again." As for the setting, this page recommends 300 for anything below 100mbps, and then another site (one of the guides in my first post) recommends an additional 300 for every 100mbps increase.

Supermule, thanks for the interesting notes. I'm hoping you might be up for adding some details.

I did some research on jumbo packets. In OpnSense, would we manipulate that by adjusting the MTU size? I have it set at default (1500) on my interfaces now. Are you thinking something other than that would be ideal?

For buffers, is what you're referring to visible by looking at the MBUF usage in the dashboard? I ran a speed test and kept an eye on the MBUF usage, but it never went above 1%, so perhaps my issue isn't here. But if I'm looking at the wrong thing, any info would be appreciated.    (FWIW, my NIC is an i350-based dual port.)

In some of my earlier testing, I tried with Suricata disabled, but that didn't seem to change anything. For thoroughness, though, after seeing your message I looked through Suricata's settings, but didn't find anything that jumped out as being a toggle between Legacy vs Inline, unfortunately. Maybe that's a setting that pfSense offers, but OpnSense doesn't?

Thanks again to everyone for the interesting exchange this is generating!