What's new

AX86U and AT&T 1 gig upload speed problem

  • SNBForums Code of Conduct

    SNBForums is a community for everyone, no matter what their level of experience.

    Please be tolerant and patient of others, especially newcomers. We are all here to share and learn!

    The rules are simple: Be patient, be nice, be helpful or be gone!

I have a similar upload issue on 1gbit Fios. Like you, straight to the ONT gives the expected results. It seems to be a lan to wan issue as it doesn't show if I use the internal router speed test or when I use iperf lan to lan. Also, interestingly it seems to be related to Windows based systems, because I get better results on Linux. Ookal's Speedtest is the worst. I get better results with the fast.com test and of course Verizon's speed test. I've been waiting 8 months for someone to identify and hopefully solve this problem, but nothing so far. If it wasn't for the fact that I get slightly better download performance with the AX86U, I'd probably switch back to a AC86U.
On windows, I noticed that to get the best iPerf3 scores required the -P switch to increase the stream count, e.g. -P 5 for five parallel streams.
 
I have a similar upload issue on 1gbit Fios. Like you, straight to the ONT gives the expected results. It seems to be a lan to wan issue as it doesn't show if I use the internal router speed test or when I use iperf lan to lan. Also, interestingly it seems to be related to Windows based systems, because I get better results on Linux. Ookal's Speedtest is the worst. I get better results with the fast.com test and of course Verizon's speed test. I've been waiting 8 months for someone to identify and hopefully solve this problem, but nothing so far. If it wasn't for the fact that I get slightly better download performance with the AX86U, I'd probably switch back to a AC86U.

For me, Windows and Linux suffer equally. When either OS is connected directly to the ONT, blazing fast uploads. Behind the AX86U, 50% reduction. Yes, the speed tests are meaningless for troubleshooting this problem. I even tried putting a switch between the ONT and AX86U, no help.

I'm probably going to try the Synology RT6600AX or........ and this will be such a PAIN IN THE BUTT ..... dust off a blade server I have sitting in my basement (some beast of a system..... dual Xeons (24? 32? cores total), 512GB RAM, 4 NICs, few TB of storage)) and repurpose it as firewall with pfSense or OPNSense and let -IT- do the routing and turn my AX86U into a dumb access point. The person in this thread was having the same problem and that was the fix for them.

I would take my ONT out of passthrough mode and use it as the router but when I first got AT&T, my Plex server was constantly having problems because the ONT (even though I configured it to forward the Plex port), was always dropping inbound 32400 TCP traffic and Plex didn't like that. :) Doing some reading and lots of folks had the same complaints about how the ONT was a crappy router with known port forwarding issues.
 
On windows, I noticed that to get the best iPerf3 scores required the -P switch to increase the stream count, e.g. -P 5 for five parallel streams.

Noticed that too. When I use parallel streams, I can reach my upload limit (930-950Mbps). So the AX86U can handle that no problem but single streams, not to much. That makes no sense to me.
 
Noticed that too. When I use parallel streams, I can reach my upload limit (930-950Mbps). So the AX86U can handle that no problem but single streams, not to much. That makes no sense to me.
It might be a Linux vs Windows implementation of iPerf3, where Linux defaults to 5 streams, and Windows to 1, requiring the -P 5.

So, are your speed problems sorted then?
 
It might be a Linux vs Windows implementation of iPerf3, where Linux defaults to 5 streams, and Windows to 1, requiring the -P 5.

So, are your speed problems sorted then?

Nope. I've tried iPerf 2.0, iPerf 3 and sender/receiver on Linux, sender/receiver on Windows and sender on Linux, receiver on Windows and vice versa .... makes no difference, I always see the same, 50% reduction in upload speeds when going through the AX86U. Doing all these tests directly through the ONT, full upload speed. And yes, when using the parallel switch, THEN I get full speed.

I'm convinced this is just a hardware limitation of the AX86U. Doing a factory reset and reset with initialization didn't solve the problem.
 
Nope. I've tried iPerf 2.0, iPerf 3 and sender/receiver on Linux, sender/receiver on Windows and sender on Linux, receiver on Windows and vice versa .... makes no difference, I always see the same, 50% reduction in upload speeds when going through the AX86U. Doing all these tests directly through the ONT, full upload speed. And yes, when using the parallel switch, THEN I get full speed.

I'm convinced this is just a hardware limitation of the AX86U. Doing a factory reset and reset with initialization didn't solve the problem.
Sorry, it's not clear, are you indeed getting full speed w/ iperf3 switch "-P 5", through the ax86u?

I believe many others here are reporting full up/down LAN<->WAN. This is likely an issue with your particular setup, and not just the ax86u.
 
That makes no sense to me.
It reminds me of testing my TB4 enclosure / drives. When dragging everything from one drive to the TB enclosure it tops out at 1.5GB/s but, when dragging individual folders it hits 2.8GB/s which is close to the max of the protocol. There's different methodologies in how data is handled but, multithreading things gets beyond the cap of the OS / protocol.

Going with a DIY setup as you mentioned will surely get you the speed you're paying for though. I did this on my cable connection to get beyond the 1gbps cap of the port to gain the over provisioning up to 1.2-1.5gbps for the same price. There may well be some additional speed ATT isn't disclosing and using a 2.5 or 5GE card will unlock it for use. A cheap / easy way to find out is a USB-C 5GE adapter for $70 before pulling out the old blade server.
 
Sorry, it's not clear, are you indeed getting full speed w/ iperf3 switch "-P 5", through the ax86u?

I believe many others here are reporting full up/down LAN<->WAN. This is likely an issue with your particular setup, and not just the ax86u.

Whoops, sorry for the misunderstanding. Yes, if I use -P 4, I do indeed see what I'd expect on the upload side.... 930Mbps.

Quick update though...... I borrowed a friends AX92U and was FINALLY able to see full upload speeds!!! Tried iPerf a few times (without the -P switch) and some FTP uploads and 920Mbps, 935Mbps..... perfect upload speeds! Making no other changes and putting my AX86U back in-line and ............ 50% reduction in speed. Put the AX92U back in, 930-950Mbps! Using all the same cables.
 
Going with a DIY setup as you mentioned will surely get you the speed you're paying for though. I did this on my cable connection to get beyond the 1gbps cap of the port to gain the over provisioning up to 1.2-1.5gbps for the same price. There may well be some additional speed ATT isn't disclosing and using a 2.5 or 5GE card will unlock it for use. A cheap / easy way to find out is a USB-C 5GE adapter for $70 before pulling out the old blade server.

The server I'm going to use has some 2.5Gbe cards in it. I'm trying to decide between pfSense and OPNSense.
 
The server I'm going to use has some 2.5Gbe cards in it. I'm trying to decide between pfSense and OPNSense.
Good. I just went with Ubuntu and added the necessary networking options to make it a router / FW using iptables. This allowed for more flexibility in other apps I'm using on the box such as hosting Plex. Using Raid for NAS drives. Network monitoring / logging through ntop. DNS through pihole. Prior to switching to an external AP it was using a PCIE card for an AP that covered everything for WIFI and some other functions.

Really it just comes down to the skin and the underlying package management options.
 
Been reading on some other sites about my problems and keep seeing mention of 'NAT acceleration, hardware acceleration, hardware NAT, Cut Through, Flow Acceleration'.... where in the world are these settings? I've combed over all my options and don't see that stuff listed anywhere. My WAN type is 'automatic' if that matters?!
 
Those settings are not managed directly by the user, but enabled/disabled by various features, like Ai Protection for example will disable h/w acceleration which is required for max throughout.

I had posted some useful links here in this thread about how to check if acceleration is disabled.

https://www.snbforums.com/threads/ax86u-and-at-t-1-gig-upload-speed-problem.79904/#post-778379

**If your properly reset the router then acceleration should be enabled. That would indeed be interesting if for whatever reason you had acceleration disabled, but yet could achieve ~900Mbps with a multi-stream iperf3 test.
 
Those settings are not managed directly by the user, but enabled/disabled by various features, like Ai Protection for example will disable h/w acceleration which is required for max throughout.

I had posted some useful links here in this thread about how to check if acceleration is disabled.

https://www.snbforums.com/threads/ax86u-and-at-t-1-gig-upload-speed-problem.79904/#post-778379

**If your properly reset the router then acceleration should be enabled. That would indeed be interesting if for whatever reason you had acceleration disabled, but yet could achieve ~900Mbps with a multi-stream iperf3 test.
AiProtection does not disable anything on the newer routers such as the AX86U, and I would guess the AX6000. But, due to the fact that packet inspection is limited to a single core it will be limited by the CPU. I was able to get gigabit down on my connection with AiProtection on, but with it off my bandwidth is consistently hitting the max over provisioned speed from Comcast (1400Mbps off vs 1100ish Mbps with it on). QoS will kill it, but with symmetric fiber that’s not even a worthwhile feature.
 
AiProtection does not disable anything on the newer routers such as the AX86U, and I would guess the AX6000. But, due to the fact that packet inspection is limited to a single core it will be limited by the CPU. I was able to get gigabit down on my connection with AiProtection on, but with it off my bandwidth is consistently hitting the max over provisioned speed from Comcast (1400Mbps off vs 1100ish Mbps with it on). QoS will kill it, but with symmetric fiber that’s not even a worthwhile feature.
Ok, good to know. I was reading elsewhere that Ai Protection was limiting throughout to ~600Mbps on AX86U, so that’s interesting that you still got full 1Gbps.
 
**If your properly reset the router then acceleration should be enabled. That would indeed be interesting if for whatever reason you had acceleration disabled, but yet could achieve ~900Mbps with a multi-stream iperf3 test.

When I was doing my router reset, I did it both ways (using the tiny button on the back) and from within the GUI, and picking the 'initialization' option. Both times, as SOON as the router came up, I did not restore the backup file and carried out my FTP, scp, iPerf tests and speed was limited to about 500Mbps on the upload. There was no QoS or AIProtection enabled. Speed tests sites showed 940Mbps up and down but in real world testing, only getting about 50% on the upload side.

This morning, I reset (both ways) the AX86U again and put it in access point mode and ran it through my beast of a server running OPNSense and same, exact results. around 500Mbps up on the upload side in real world testing.

Can somebody with this router use iPerf to an external source (and without the -P switch) and post their results? And/or, do an upload to Google drive or Azure's large file upload test site and post a screen shot? I'm going to post on ASUS's forums or email their support and see what they say I guess.
 
I am happy to give it a try since I am on 1 Gig Fiber as well - but I'd need an external source that can handle the speed for testing
 
Ok, good to know. I was reading elsewhere that Ai Protection was limiting throughout to ~600Mbps on AX86U, so that’s interesting that you still got full 1Gbps.
It depends on what else you have the router doing. I only had the Trend Micro stuff enabled and no scripts when running merlin. Adaptive QoS will slow it down more despite not disabling acceleration. I got anywhere from 750-850 down when using AQoS. I have since turned all that off since AiProtection only seemed to give false positives that ended with me dealing with unhappy users. May enable it again sometime.
 
This morning, I reset (both ways) the AX86U again and put it in access point mode and ran it through my beast of a server running OPNSense and same, exact results. around 500Mbps up on the upload side in real world testing.
What do you mean by "put it in access point mode". If you're doing that then you're not touching any of the AX86U's routing functions or things like hardware acceleration, AiProtection, QoS, etc. So assuming you're using Ethernet connections and not WiFi you are effectively just using the LAN ports on the AX86U as a dumb switch.
 

Latest threads

Sign Up For SNBForums Daily Digest

Get an update of what's new every day delivered to your mailbox. Sign up here!
Top