Slow speedtest results on router compare to LAN clients

[MadChorus]

Hi There.

After several years with NetGear and DD-WRT, I am so glad to come back to the Asus Routers and Merlin.
I recently got the AX88U connected to a 1gb up/down fiber connection. Of course, as soon as I got the router I immediately
installed the Asuswrt-Merlin build and started tinkering with it.

One of the first things I did is speed test using the spdMerlin script. I immediately noticed that the speed is lower than usual:

Latency: 1.03 ms (0.89 ms jitter)
Download: 389.80 Mbps (data used: 382.9 MB)
Upload: 424.54 Mbps (data used: 396.0 MB)
Packet Loss: 0.0%

I thought it might have to do with QOS settings (probably not, because it is directly on the router), so turn on and off just for the sake of testing
but the same results.

I sshed to my ubuntu machine (connected to the router over ethernet cable) and tested the speed again using speedtest-cli:

Testing download speed................................................................................
Download: 619.04 Mbit/s
Testing upload speed......................................................................................................
Upload: 359.20 Mbit/s

I am going to ignore for now the fact that both are nowhere near the promised 1gb up/down speed (with the dd-wrt I used to get 750mbit/500mbit).

For a start, I was wondering if anyone has any idea why the client yields a much faster result than the router itself? could it be because of the hw specs
of the Linux server compare to the router?

Appreciate it if anyone has any ideas of how/where I should look into to get to the bottom of this issue.

Thanks,
MC

 

[Jack Yaz]

From the first post of spdMerlin:

spdMerlin uses SpeedTest CLI
If the speeds from this utility are significantly lower than those you see from the desktop app/browser test, the limiting factor will be your router's CPU. That being said, there is probably room for improvement/optimisation by the Ookla team in the CLI binary itself. Please direct feedback about speed issues to Ookla via the above link. There is nothing that I can do about them.

Running on the router itself also cannot fully leverage hardware acceleration for the test.

I'm very interested to see what Asus say/do for this with their speedtest in the 386 beta firmware...

 

[amplatfus]

Same issue on my site with RT AC88U. Max speed: D 341; U 369. From my clients on LAN, CAT 5e cable: D 924; U: 824.
I will direct feedback about speed issues to Ookla.

Thank you!

 

[Ronald Schwerer]

I didn't notice this issue until I updated to 1Gb fiber. I'm getting similar slow speed reports on the router and expected (close enough) throughput on the clients. I previously had a 100Mb down and 10Mb up connection and always got the expected speeds (or better) on both router and client tests.
I had used the Ookla binaries in my own script to run the test on the 68U router every hour and post the results in the system log. I stopped testing on the router. It's a waste of CPU cycles generating meaningless statistics.

 

[R1-Limited]

Min is the same, if I turn off QOS it is still slower but far better. QOS on with rates at 999/999 or 1000/1000 setup results in 100/100 in speed test or above

 

[BeLikeWater]

Hi guys, I haven’t been part of this issue until now.
Has there been any change to the situation?

 

[phoenixxko]

I don't think so - router test - 250/250 - wireless client - 400/400

 

[BeLikeWater]

Yeah mine is REALLY odd though. I'm on an Internet link of 1000/50 and it will test as much.
If I put it on with spdMerlin doing its thing, it will only do 4Mbps/50Mbps
I can’t find else in-play to do it. Touch nothing and repeat without spdMerlin and we are back to 985/48Mbps
I'm stumped on this one.

 

[DevilSShadoW]

Stumbled upon this thread while looking for potential reasons for the speed discrepancy between spdMerlin and regular web speedtesting (or even win64 speedtest CLI). Feared it might have something to do with hardware acceleration when I noticed the CPU usage when running spdMerlin.

Using a 500mbps/250mbps connection and I'm getting nowhere close the correct reported speed -web/win64 CLI reports a constant 480~/250, while spdMerlin reports... well... see below.

If I remember correctly, the built-in asus speedtest had similar issues.
Sad to see because the potential of this addon is great. Hopefully asus addresses this issue with one of their upcoming firmware updates.
For now I'll uninstall the addon to avoid wasting cpu cycles while I monitor this thread for further developments.

PS: using an RT-AC66U B1 running 384.19 (ARMv7/1ghz 2x cores/256mb ram)

 

[bennor]

In the recent Merlin beta threads people have reported that their router run speed tests were crippled by the TrendMicro AI Protection settings. Some folks found their speed tests dramatically improved once they forced a withdraw from the TrendMicro AI protection on the Administration > Privacy page. Couldn't hurt to disable AI protection and withdraw from it on the Privacy page as a troubleshooting step.

 

[DevilSShadoW]

Well I'm definitely getting higher speed results after withdrawing from that privacy agreement (wasn't using any AI protection features at all though). Getting 270mbps~ downloads, still nowhere near my 490~mbps speeds but an improvement nonetheless.

 

[L&LD]

Did you reboot, wait for the router to settle afterward, and then retest?

 

[bradbort]

Just confirming that as of 5/2021, this is still an issue. The built in speed test is too low compared to reality for higher cable modem speeds. I get 800, and if I test directly through a wired client using ookla, that is what I see. If I run the speed test built into the GUI, it shows about 200 less. Turning off QOS or AI protection has not meaningful impact on the speed test results. It is just wrong. Hopefully, ASUS will eventually fix it. I imagine it was built before 1 gig connections were becoming common, and there is some code that needs to be adjusted

 

[ColinTaylor]

Just confirming that as of 5/2021, this is still an issue. The built in speed test is too low compared to reality for higher cable modem speeds. I get 800, and if I test directly through a wired client using ookla, that is what I see. If I run the speed test built into the GUI, it shows about 200 less. Turning off QOS or AI protection has not meaningful impact on the speed test results. It is just wrong. Hopefully, ASUS will eventually fix it. I imagine it was built before 1 gig connections were becoming common, and there is some code that needs to be adjusted

I think this is hardware limitation so there's nothing that can be done to "fix" it. Just ignore it and use a LAN speed test instead.

 

[MadChorus]

Just confirming that as of 5/2021, this is still an issue. The built in speed test is too low compared to reality for higher cable modem speeds. I get 800, and if I test directly through a wired client using ookla, that is what I see. If I run the speed test built into the GUI, it shows about 200 less. Turning off QOS or AI protection has not meaningful impact on the speed test results. It is just wrong. Hopefully, ASUS will eventually fix it. I imagine it was built before 1 gig connections were becoming common, and there is some code that needs to be adjusted

I agree with ColinTaylor. I am no longer performing speed tests on my Asus. I just use my Linux server to perform periodic speed tests.

 

[DeltaB]

My experience with my ASUS router is that speedtests utilizing the CPU (no NAT accel) will be limited to what the CPU can handle. (around 250mb on mine) Enabling NAT acceleration in LAN-Switch Control tab, can allow clients to bypass the router CPU and obtaining full 1Gb bandwidth, however it is not applied to the WAN port during testing from the router itself, it passes through the CPU. You will see this same thing from clients on LAN ports (limited to around 250mb) if you disable NAT acceleration. This is a limitation of the speed of the CPU of the router itself. (it's at 100% utilization)

 

[kinglevel]

Same here.
Got me worried that my vpn network with multiple sites was faulty set up.

But after digging in grafana graphs i noticed the tun interfaces were low and cpu on the router was high during tests.
snmp ftw

 

[Tech Junky]

Turn off all the crap on the router. These devices aren't meant to be doing 1/2 the crap that's baked into them.

QOS / WMM is needed for high speed WIFI to operate correctly or your speeds will nose dive.

spdMerlin - not sure how this is built but, it seems like a possibility that it's not scanning available servers for optimal results or there's a limiter being applied

Going from LAN >> WAN for speed testing hits the CPU differently than a direct WAN test as the ASIC's are involved in the switching portion of the device. Any time you have to "route" you hit the CPU with cycles that slow things down. Same goes for Firewall activities if you have a bunch of rules to parse through before exiting the network.

Also, I find using SpeedTest vs Ookla generates more reliable numbers but, I'm not sure you can use it on these devices ST uses the same servers but polls then more accurately for the best GEO match and runs an extended test for accurate results.

The real test though is downloading an actual file vs synthetic testing. I like to use Linux BitTorrent downloads for a more accurate reading on performance because it's ~1GB in size and you can hit line speeds with multiple sources in your DL client.

 

[ColinTaylor]

Turn off all the crap on the router. These devices aren't meant to be doing 1/2 the crap that's baked into them.

QOS / WMM is needed for high speed WIFI to operate correctly or your speeds will nose dive.

spdMerlin - not sure how this is built but, it seems like a possibility that it's not scanning available servers for optimal results or there's a limiter being applied

Going from LAN >> WAN for speed testing hits the CPU differently than a direct WAN test as the ASIC's are involved in the switching portion of the device. Any time you have to "route" you hit the CPU with cycles that slow things down. Same goes for Firewall activities if you have a bunch of rules to parse through before exiting the network.

Also, I find using SpeedTest vs Ookla generates more reliable numbers but, I'm not sure you can use it on these devices ST uses the same servers but polls then more accurately for the best GEO match and runs an extended test for accurate results.

The real test though is downloading an actual file vs synthetic testing. I like to use Linux BitTorrent downloads for a more accurate reading on performance because it's ~1GB in size and you can hit line speeds with multiple sources in your DL client.

@Tech Junky You clearly don't understand how the hardware works in these Asus routers or the issue this thread is describing. That's understandable as you don't have an Asus router. But I suggest you refrain from posting inaccurate information about things don't understand as it may confuse people who read it.

 

[Tech Junky]

have an Asus router.

I've had one in the past and it's sitting in a parts pile in the other room. I tested it out when I switched to 1GE service and it flopped miserably for throughput. The special sauce baked into the system OS hobbles the ability to do basic functions. When I was only using a 50mbps plan the lack of throughput wasn't noticed because speed tests wouldn't reveal just how bad the HW / SW were. It has gig ports on it but, PC <> PC transfers didn't add up upon further testing.

Maybe that's why there's so many threads here?

You shouldn't have to hack your device to get basic functions to actually work. I've tried many different routers over the years and it always feels like you're settling for the least worst option. Even w/o taking into consideration of using a homebrew option.

There are faults to all black plastic boxes from retail outlets but, some do better than others at hiding their imperfections. The software component usually differentiates them even more than the HW since well, no company is reinventing the wheel here when it comes to what's in the black boxes.