Some kind of 2.5gbit bottleneck issue - testing new GTAX11000 on Merlin latest 388.7

[Tech9]

I just meant a straight Lan to Lan connection

This 350Mbps LAN-LAN what is exactly showing? If LAN-LAN an old RT-AC68U can do 940Mbps easily. Switched traffic, max port speed.

 

[Swistheater]

View attachment 60205

Router to pc; Router 2.5g to 10g nic set to 2.5g on pc.

Hw acceleration is disabled for me... Just remembered since I use cake.

View attachment 60206

No real difference interestingly even worse with qos cake off.

yea these don't look promising.

 

[DJones]

Turned off jumbo frames and made sure mtu was 1500.

When running the test one core is pinned at 100% so it could be a matter of bottleneck from cpu. iperf is not multithreaded. Edit: newer versions of iperf server might be multi threaded however I'm unsure what version or if entware is updated enough or if the router supports multithreading.

PC

Router

For reference ubuntu also ships with 3.16

 

[DJones]

iperf3 FAQ — iperf3 3.17.1 documentation

software.es.net

In other words I can't really test it accurately since it doesn't appear to be multithreaded at least on the router. The other method just seem like a pain, and I don't really know how to run multiple instances over ssh.

 

[DJones]

Looks like I have to state parallel otherwise it’s single threaded. Will try again later.

”Support the use of multiple threads for parallel tests. Requires pthread and stdatomic support.”

 

[Tech9]

Are you testing what the router's CPU can process or something else?

 

[DJones]

I’m trying to figure out how much that port can handle in a as close to wan situation as possible. If it’s a 2.5g port it shouldn’t be worse than using a 1g port. However theirs no way to directly test that port if I can’t produce similar traffic.

I would simply pass iperf from pc to pc through the router, but packets are not going to be processed by the router rather just forwarded doing that. That might be enough to see near 1g speeds, but was trying to see if it can be near or exceed 1g speeds or at least be better then using a 1g port. Only having 1 single 2.5g port on the router doesn’t really help.

 

[Tech9]

You can’t do this the way you test. You hit the CPU limit first and it’s about 350Mbps.

 

[DJones]

You can’t do this the way you test. You hit the CPU limit first and it’s about 350Mbps.

Yea, I was hoping multithreaded iperf would have solved that issue. I haven’t tried parallel processes with that yet. I’m doubtful it would even work on a router.

 

[DJones]

Yea, I was hoping multithreaded iperf would have solved that issue. I haven’t tried parallel processes with that yet. I’m doubtful it would even work on a router.

I guess the point I’m trying to make is the port itself at fault or limits to the cpu. In his case he noticed a difference from 1g port to 2.5g port.

 

[DJones]

iperf3 -c 192.x.x.x -p 5201 -P 4

No issues with the 2.5g port with 4 parallel streams I get 1.30 GBits/s without the connection being bottlenecked by a single core. 1500 mtu

 

[lgkahn]

for comparison

What is your opinion of the GT-AXE16000 and GT-BE98_PRO in comparison with the GT-AX11000 Pro? I may pick one of those up.

cant answer directly but for routers i have

gt-axe11000 pro on 1gb/sec backgone using 2.5gb port

gt-axe16000 on 2.5gb/sec backbone router mode using 2.5gb port for internet and 10gb port connected at 2.5gb for intranet

gt-axe11000 on 2.5gb/sec backbone (ap mode using 2.5 port not working well cannot get the rated port speed




View attachment 60230

a differnt gt-axe1100 in ap mode (different house) again on 2.5 backbone using 2.5 port.. almost the same.. cannot reach rated speed

a differnt gt-axe16000 in ap mode, (not main router) using only the 10gb port on a 10gb backbone. again good but cannot achieve rated speed. my pc on same backbone to same iperf server can get 6.5gb/sec

 

[Swistheater]

View attachment 60225

iperf3 -c 192.x.x.x -p 5201 -P 4

No issues with the 2.5g port with 4 parallel streams I get 1.30 GBits/s without the connection being bottlenecked by a single core. 1500 mtu

So that sounds like when I run the speed test, the router is not hitting all the cores. is there away to change this behavior ?

 

[DJones]

So that sounds like when I run the speed test, the router is not hitting all the cores. is there away to change this behavior ?

You’re already getting above 350Mbps with nat acceleration on. Do a speed test and watch the cores see if one spikes or they all do. I think speedtest.net lets you choose a single stream test vs multi stream.

 

[DJones]

Otherwise to get full line speeds you’re going to need hardware capable of those speeds. You could get a 2.5g small business class router non wireless just to handle the higher speeds and not use wan of the gt-ax11000 just plug into the lan of 2.5g or use double nat for wan and use dhcp automatic and that might help. I’m unsure. I haven’t investigated this since I don’t have fast speeds, but lots of people have issue hitting top line speeds on asus routers.

 

[Swistheater]

Otherwise to get full line speeds you’re going to need hardware capable of those speeds. You could get a 2.5g small business class router non wireless just to handle the higher speeds and not use wan of the gt-ax11000 just plug into the lan of 2.5g or use double nat for wan and use dhcp automatic and that might help. I’m unsure. I haven’t investigated this since I don’t have fast speeds, but lots of people have issue hitting top line speeds on asus routers.

So when I switch the GT-AX11000 to AP mode and connect it to a 2.5gbit business/consumer class x86-64 router that has a 2.5gbit wan, and 4- 2.5gbit lan ports, my speed test is roughly the same results with devices connecting to the GT-AX11000 over wired connection. Note the business class router does not provide wireless gateway, so I must use wireless from the GT-AX11000 in this configuration. Better speeds when using the 1gbit port over the 2.5gbit port. When I connect a device to the GT-AX11000 over Wifi, the speed is roughly between 600 and 800mbps with wifi 6 connected device and using the 2.5gbit port as the main cable between the router and the GTAX11000 AP.

 

[DJones]

So when I switch the GT-AX11000 to AP mode and connect it to a 2.5gbit business/consumer class x86-64 router that has a 2.5gbit wan, and 4- 2.5gbit lan ports, my speed test is roughly the same results. Better speeds when using the 1gbit port over the 2.5gbit port. When I connect a device to the GT-AX11000 over Wifi, the speed is roughly between 600 and 800mbps with wifi 6 connected device.

That’s unusual. Maybe try iperf3 just to narrow down it isn’t a hardware issue per say. Configuration wise theirs nothing I know to change. Hardware acceleration is handled by broadcoms engine so that would be on rmerlin / asus’s side. If 1gb ports work better use that. But seems unusual.

 

[Swistheater]

That’s unusual. Maybe try iperf3 just to narrow down it isn’t a hardware issue per say. Configuration wise theirs nothing I know to change. Hardware acceleration is handled by broadcoms engine so that would be on rmerlin / asus’s side. If 1gb ports work better use that. But seems unusual.

Heres an image of the device's speed test connected to the GTAX11000 in wifi 6 mode---> the GTAX11000 is using the 2.5gbit port for its inbound connection from the main router.

Here is an image of a 1gbit devices speed test physically connected to the GTAX11000 device using one of the 1gbit ports.

 

[DJones]

Wifi speeds are also going to be variable due to conditions. Whats the connected speed say on 2.5g port when using the Internet speed function? That will tell you the speed coming in from wan 2.5g rather then what your getting over wireless.

?

 

[Swistheater]

If I move the cable coming from the router to the GTAX11000 to the 1gbit port here is the different results over cable versus wifi6.

wifi6:

ethernet: