Jump to content

Bufferbloat issue


Recommended Posts

Running the XR500 with the latest V2.3.2.134 firmware and getting some interesting bufferbloat issues

with the QoS set correctly I'm getting 30+ms added under load and the internal dumaos test says thats an A+

but that is NOT an A+ that is bad, external tests show it will be an issue with low latency gaming and it shows in my games.

Interestingly with QoS disabled its reduced to 1ms under load (which I know won't stay that way without QoS once the wife starts watching youtube etc with no QoS)

testing done at the worst case time of the day for my internet, to set the max speeds and congestion controls below that to allow headroom

testing done with goodput on and off

 

I'm wondering if this is a CPU constraint because:

 

I know my connection can be sorted to have 0-1ms all the time under max load because I used to have it working that way with my home built openwrt running a really old dual core intel atom ITX embedded cpu board, and it can use cake piece of cake QoS to keep my bufferbloat to near nothing.

But I want the great wifi, and awesome Geo-filter options of dumaos

 

Now I know dumaos is based on openwrt, so I was wondering if IRQ Balance has been enabled on dumaos's Openwrt core.

because all openwrt versions had to have it enabled manually, and it was one of the requirements to get bufferbloat to behave itself when using QoS while running multiple core cpu's in openwrt

(instructions on how to enable for openwrt here: https://forum.openwrt.org/t/enabling-irqbalance/98750 )

 

 

 

Internal test with QoS

internal test.png

External test Qos at the best I could get it, typically Download active says 30-40ms unless I drag the slider ALL the way down to under 280, with WRT I could keep it up at 350-400

external.png

 

Internal test QoS disabled

internal qos disabled.png

 

Test with QoS Disabled:

external no qos.png

Link to comment
Share on other sites

  • Administrators

Benchmark uses the Google servers so depending on your connection to them results will differ. Id recommend using this guide https://support.netduma.com/support/solutions/articles/16000074717-how-to-test-your-internet-ping to test. Saturate your connection with downloads, streams etc and see what percentage brings it down as much as possible. That will be the fairest test as it will be done with actual traffic.

Link to comment
Share on other sites

22 minutes ago, Netduma Fraser said:

Benchmark uses the Google servers so depending on your connection to them results will differ. Id recommend using this guide https://support.netduma.com/support/solutions/articles/16000074717-how-to-test-your-internet-ping to test. Saturate your connection with downloads, streams etc and see what percentage brings it down as much as possible. That will be the fairest test as it will be done with actual traffic.

Yea, I check for bad nodes, packetloss etc using tracert and other diag tools whenever there is an issue as a first check, there isnt.
and I can plug back in my old arse openwrt router and get 0-1ms bloat on all the various bloat tests under every stress test of my network I can do.

its the load on the router, it can't handle my networks workload + Qos + geo filter , my networks got alot going on, bit too much for this, I might put the wrt router back in to do all the heavy lifting and just make a seperate subnet for my gaming machine

Link to comment
Share on other sites

7 hours ago, Netduma Liam said:

Can you provide some screenshots of how you've currently got QoS configured so we can check there?

I have had the sliders in every position with every combo of internet speeds set over the last two days

my net is typically capable of doing 980/490 on a standard day, 700ish/420ish when my isp is under full load with all QoS disabled

the screenshots above were limited to 600, 500, 350, etc made no difference bufferbloat and ping change under load was always the same

goodput on and off at each position of the sliders made no difference

reseting bandwidth allocation made no difference

Typically set qos to 350

setting max speed and sliders a bit lower

limited max .png

 

This is the speed my net typically hangs out at using googles servers with QoS Disabled.

NO QOS speed test.png

this is it typically on my isp's server

QOS off speed test typical to my ISP.png

 

This is the XR500 test with sliders both high and max speed set to 1000

Xr500 sliders higher.png

 

this is the XR 500 with sliders set to around 300 down and up, max speed autoset

Same limits Congestion control 350 xr500.png

 

this is it set to 350 350 max speed auto set

and again same limits 350 XR500.png

 

This is the XR500 with its QoS disabled behind the openwrt router which is using QoS SQM at 350/350

Congestion control turned off openwrt doing the work.png

 

Heres the Openwrt's bufferbloat test

Openwrt bloat.png

And its settings

openwrt qos.png

more settings

openwrt que disc.png

 

Either way, I've got my new setup as Openwrt router - all the rest of the network except wifi stuff - Xr500 - gaming machine/wifi crap

the XR500 no longer has to bear the load of the rest of the network, just some basic stuff

 

Link to comment
Share on other sites

  • Administrators
17 hours ago, Raxis said:

Yea, I check for bad nodes, packetloss etc using tracert and other diag tools whenever there is an issue as a first check, there isnt.
and I can plug back in my old arse openwrt router and get 0-1ms bloat on all the various bloat tests under every stress test of my network I can do.

its the load on the router, it can't handle my networks workload + Qos + geo filter , my networks got alot going on, bit too much for this, I might put the wrt router back in to do all the heavy lifting and just make a seperate subnet for my gaming machine

It's not for checking bad nodes or packet loss per se, it's exactly the same as the bufferbloat tests but you'd be using real traffic, you would see how the ping increases then use QoS to bring it down as much as possible, I would recommend trying it.

Disable Traffic Prioritization completely as that can interfere with the tests you're doing.

Link to comment
Share on other sites

Just now, Netduma Fraser said:

It's not for checking bad nodes or packet loss per se, it's exactly the same as the bufferbloat tests but you'd be using real traffic, you would see how the ping increases then use QoS to bring it down as much as possible, I would recommend trying it.

Disable Traffic Prioritization completely as that can interfere with the tests you're doing.

righto, gonna give it a go, wont be this week, got stuff to do but will update when I reorganize the network again

Link to comment
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
 Share

×
×
  • Create New...