RT-N66U download speed half the upload speed on fiber

[blackholesun]

For the past year I ran Tomato by Shibby on my RT-N66U. This week I got upgraded on my FTTH service from 100 to 600 Mbit/s and while Shibby's FW is very nice it does not offer ctf for N66U.

Today I switched back to the original Asus FW, 3.0.0.4.378_7410.

My speeds when I connect the WAN cable directly to my laptop are around 550 Mbit/s download and 550 Mbit/s upload.
However, when I connect WAN -> RT-N66U -> laptop, I get 225 Mbit/s download and 550 Mbit/s upload.
After flashing I restored factory settings and checked two different cables between the router and the laptop.

Is there some setting on the router that I am missing? Or is that max download speed that the N66 can handle?
LAN/Switch Control shows CTF is enabled.

 

[hggomes]

QoS?

 

[blackholesun]

Did not touch the QoS settings after returning to Asus FW. Smart QoS is OFF.

 

[hggomes]

Have you reset the router using RESET button (pressing it 5-10s) after switching from Tomato to ASUSWRT?

Have you imported any old settings .cfg file? Or you have configure it manually from scratch?

 

[blackholesun]

I configured everything from scratch. Rebooted the router but did not use the reset button.
Meanwhile I contacted Asus support and was told that I had reached the hardware limitation of this router.

edit: ok I pressed the reset button for 10 sec and nothing's changed for dl speed

 

[hggomes]

You can easily do 700-750MBIT on your WAN on RT-N66U.

I tested RT-N66U on a 500MBIT connection back in time and i was able to get the full speed, so theres no doubts about it.

 

[hggomes]

Check this out:

http://www.smallnetbuilder.com/wire...n900-gigabit-router-reviewed?showall=&start=2

Try RMerlin 378.55 or John FW and check if the same happens.

 

[blackholesun]

Did you get such speeds on Asus or Merlin FW?

 

[hggomes]

Merlin FW, but theorically the results should be the same, since they are based on same code.

 

[sfx2000]

You can easily do 700-750MBIT on your WAN on RT-N66U.

I tested RT-N66U on a 500MBIT connection back in time and i was able to get the full speed, so theres no doubts about it.

Agreed...

Any AC1900 class router/AP can hit 700 Mbit/Sec, and many can saturate a Gigabit link...

 

[hggomes]

Indeed, though RT-N66U is a N900 class model.

 

[sfx2000]

Indeed ,though RT-N66U is a N900 class model

Most of the N900 routers can burst at around 700, but they'll settle around 550 or so - with one exception - the Airport Extreme N (the flat ones) in NAT mode were pretty slow, in Bridge mode they were fine - doing some digging, apparently there were issues with NetBSD and NAT performance (which the Airport firmware is based on).

Did not really matter if the core CPU on the Router was ARM or MIPS, as most traffic actually runs on the internal switch or DMA in our test environment that we had at the time.

The AC1900's - I think they're faster because the cores are much faster than what we had in the N900 days...

 

[hggomes]

Not sure if i entirely agree on that, look at RT-N16 (480MHZ) / 300-350MBIT and RT-N66U (600MHZ) / 700-750MBIT WAN performance, they are both the same MIPSr2 core CPU, and completely different results.

 

[sfx2000]

Not sure if i entirely agree on that, look at RT-N16 (480MHZ) / 300-350MBIT and RT-N66U (600MHZ) / 700-750MBIT WAN performance, they are both the same MIPSr2 core CPU, and completely different results.

I said burst, not steady state

and yes, clockspeed matters on the WRT based builds as it's interrupt driven.. that probably what drove Broadcom and Atheros over to ARM, as the licensed MIP cores were maxxed out around 600MHz on a 90/65nm process node.

Cortex-A9's could scale up clockwise (and they had libraries that could scale to smaller process nodes, A9 starts at 45nm@800GHz, and many have taken it down to 28nm on TSMC without any problems at 1GHz, not sure about GloFo, but I expect about same)

 

[hggomes]

 

[sfx2000]

Amen - most folks don't know how hard it actually is to maintain a gigabit stream from client to server... it's freaking hard, and most folks will not see this for the next 12-18 months perhaps, and even then...

On the test bench - we had a specific 16 socket (Xeon's, 192 cores/twice that for threads) blade server on a 40Gb link in the data center (in Denver) across a dedicated Level 3 Communications 10GB link to our Denver corp campus (about a 5 mile run), and a dedicated 1GB link to the devices under test - using iPerf on both ends - the client was a MacBook Pro* running RHEL (thanks to bootcamp and refit) as was the server - and the stacks were seriously tweaked...

*2013 Retina Macbook Pro runs a Broadcom 4360 configured as a 3*3:3 client for 802.11ac - at the time, it was the absolute fastest machine we could get that supported all the check boxes

 

[hggomes]

Quite fun testing in there Mr. Wireless Ninja.

PS: Nice signature btw.

 

[sfx2000]

Quite fun testing in there Mr. Wireless Ninja.

PS: Nice signature btw.

It was a fun project.. learned a thing or two, and confirmed some reasonable expectations..

sfx

 

[sfx2000]

On the test bench - we had a specific 16 socket (Xeon's, 192 cores/twice that for threads) blade server on a 40Gb link in the data center (in Denver) across a dedicated Level 3 Communications 10GB link to our Denver corp campus (about a 5 mile run), and a dedicated 1GB link to the devices under test\

BTW - when one throws that much resources against a SOHO gateway feom the WAN side - things generally tend to crash.. a 40Gb pipe into a 10Gb pipe - consumer GW's will eventually fail if someone gets on the back end...

 

[mstombs]

For the past year I ran Tomato by Shibby on my RT-N66U. This week I got upgraded on my FTTH service from 100 to 600 Mbit/s and while Shibby's FW is very nice it does not offer ctf for N66U.

I have also seen that you need CTF enabled to see above 150Mbps in a speedtest on N66.

However, when I connect WAN -> RT-N66U -> laptop, I get 225 Mbit/s download and 550 Mbit/s upload.

Is wan type simple dhcp of do you use a pppoe connection?

If simple dhcp it looks like a new firmware issue, if the cpu is at the limit every extra iptables check will reduce throughput, could also be how the router in the link gets the tcp/ip rwin pacing to be negotiated between server and client.