RT-AC68U speeds for hosting (aka what is my bottleneck)?

[bengalih]

I just updated from a 300/30mbps setup to AT&T fiber. I am running NGINX on my AC68U and proxy several web services (including a web file server and Plex) to a backend server which is connected at 1 GBps.
If I run a speed test from this server (using any of the big speed test providers), I normally get in the range of 450-850mbps up/down. Depends on the time of day, but I haven't seen it below 450.

I am not using the AT&T gateway, I am running wpa_supplicant with the private certs directly on my AC68U, so we don't need to worry about any equipment other than this.
I have also run some test between this PC and another PC wired to the AC68U and file transfers (via SMB) range between 750-880 mbps (about 110MB/s).
I believe this shows that the AC68U is capable of routing (at least over my LAN) at GB speed.

I realize that the AC68U is not capable of full GB throughput to the WAN, but my speedtests out to the internet show that I should be able to sustain at least half of this.

However, when any remote clients access my services they can't seem to download at a speed higher than 60-70mbps, with many of them closer to the 30mbps mark.
This is true regardless of their download speeds. In fact my father's house on FIOS was seemingly only measuring about 90mbps down from Ookla, but only measuring about 50-60 down from me.
Alternatively, a friend who was getting 250mbps in his speedtests were only seeing 20-30 down from me.

If I download over http between servers in my LAN I see approximately 70-80MBps (5-600 mbps). However if I open a port forward direct to these services and attempt to download directly over the internet, the remote clients get about 1/10th of this speed.

Is the RC68U really throttling these speeds to this extent? And, if that is the case, why is it able to achieve such high speeds on the speed test?

 

[bengalih]

I just updated from a 300/30mbps setup to AT&T fiber. I am running NGINX on my AC68U and proxy several web services (including a web file server and Plex) to a backend server which is connected at 1 GBps.
If I run a speed test from this server (using any of the big speed test providers), I normally get in the range of 450-850mbps up/down. Depends on the time of day, but I haven't seen it below 450.

I am not using the AT&T gateway, I am running wpa_supplicant with the private certs directly on my AC68U, so we don't need to worry about any equipment other than this.
I have also run some test between this PC and another PC wired to the AC68U and file transfers (via SMB) range between 750-880 mbps (about 110MB/s).
I believe this shows that the AC68U is capable of routing (at least over my LAN) at GB speed.

I realize that the AC68U is not capable of full GB throughput to the WAN, but my speedtests out to the internet show that I should be able to sustain at least half of this.

However, when any remote clients access my services they can't seem to download at a speed higher than 60-70mbps, with many of them closer to the 30mbps mark.
This is true regardless of their download speeds. In fact my father's house on FIOS was seemingly only measuring about 90mbps down from Ookla, but only measuring about 50-60 down from me.
Alternatively, a friend who was getting 250mbps in his speedtests were only seeing 20-30 down from me.

If I download over http between servers in my LAN I see approximately 70-80MBps (5-600 mbps). However if I open a port forward direct to these services and attempt to download directly over the internet, the remote clients get about 1/10th of this speed.

Is the RC68U really throttling these speeds to this extent? And, if that is the case, why is it able to achieve such high speeds on the speed test?

I've done some more testing having people at various location with download test speeds well above 200mbps accessing large files from my web server. Results were all fairly similar:

- 112 mbps doing http straight passthrough via a port forward
- 80 mbps http through NGINX (which is running on the Asus)
- 40 mbps https through the proxy.

Contrast this with the 700-800 mbps I get doing straight http within the LAN. Also, my speedtests (even to remote servers in the locations being tested from) shows upwards of 300mbps.

I am not seeing excessive CPU utilization at these speeds of the router, so why if my speedtests report 300+ mbps, can people only access my data at speeds ~100 mbps ?

 

[W1F1]

I've done some more testing having people at various location with download test speeds well above 200mbps accessing large files from my web server. Results were all fairly similar:

- 112 mbps doing http straight passthrough via a port forward
- 80 mbps http through NGINX (which is running on the Asus)
- 40 mbps https through the proxy.

Contrast this with the 700-800 mbps I get doing straight http within the LAN. Also, my speedtests (even to remote servers in the locations being tested from) shows upwards of 300mbps.

I am not seeing excessive CPU utilization at these speeds of the router, so why if my speedtests report 300+ mbps, can people only access my data at speeds ~100 mbps ?

This may help? https://www.snbforums.com/threads/rt-ac68u-getting-1gbps-wan-lan-throughput.29471/

 

[bengalih]

This may help? https://www.snbforums.com/threads/rt-ac68u-getting-1gbps-wan-lan-throughput.29471/

Thanks, but doesn't seem relevant. I can hit high speeds via my speed tests, so I know the setup is capable of that. My issue is the speeds reported by users actually accessing data from mw are much slower.

 

[OzarkEdge]

Thanks, but doesn't seem relevant. I can hit high speeds via my speed tests, so I know the setup is capable of that. My issue is the speeds reported by users actually accessing data from mw are much slower.

Maybe the bottleneck is beyond your LAN. Maybe your ISP throttles traffic to/from you vs traffic from/to you... unless you pay for a static IP(?). Plus possible slowdowns at the other endpoints. Just wondering about it... I don't really know except big telco like AT&T can be difficult.

OE

 

[W1F1]

Thanks, but doesn't seem relevant. I can hit high speeds via my speed tests, so I know the setup is capable of that. My issue is the speeds reported by users actually accessing data from mw are much slower.

Do you have jumbo frames enabled and/or NAT cut through? Is QoS enabled? Ideally, jumbo frames are disabled, NAT cut through is enabled and QoS disabled then test again. This may be obvious, but the router's LAN side only has to switch packets so that throughput will always be substantially higher than the WAN side since there's no routing involved.

 

[bengalih]

Maybe the bottleneck is beyond your LAN. Maybe your ISP throttles traffic to/from you vs traffic from/to you... unless you pay for a static IP(?). Plus possible slowdowns at the other endpoints. Just wondering about it... I don't really know except big telco like AT&T can be difficult.

OE

That's what I've been thinking, that the peering at the testing points is just too indirect. It's orders of magnitude though - fairly disappointing.