There's no way those PPPoE numbers are correct!

I spent a bit of time today setting up a PPPoE server and the results for my WDR4900 (which should be faster than the WDR4300) using OpenWRT were 110Mbps.

However, the server was setup in a virtual machine using virtual bridged NICs. Now that i know how to set the server up i'll do it again sometime over the next couple of days using a live linux image.

Just to get a picture of the PPP overhead i connected my laptop (Core 2 Duo T7100) directly to the machine hosting the virtual PPPoE server (Core i3 3225) and did a couple of quick tests:

Without PPPoE: 724Mbps
With PPPoE: 86Mbps

Pretty shocking results, but i'm convinced the virtual machine is slowing things down a fair bit too.

I'll explore things a bit further in a couple of days.

Simultaneous dual-band is hard to find under £50 new.

One I can think of is the TP-LINK TL-WDR3500 N600:
http://www.amazon.co.uk/gp/product/B009EOQPZC/ref=ox...

Though I wouldn't buy it. Have read some reliability and performance issues.

The TP-Link TL-WR2543ND would be better, though it's dual-band it's not simultaneous (only one radio):
http://www.amazon.co.uk/gp/product/B006B3BM9I/ref=ox...

Instead I would recommend saving up for the Asus RT-N66U.

Are you on commission with ASUS?

The reason I ask is that you registered a couple of hours ago and all you have done is recommend the ASUS RT-N66U, and I don't think you own one.

Edited by deleted (Tue 02-Jul-13 19:54:01)

Just done a bit more testing and it does seem that the virtual machine was adding significant overhead.

The iperf and PPPoE servers are run on an Intel Core i3 3225 with Atheros AR8151 Gigabit NIC using Linux Mint 15 (32 bit) from a live USB stick.

The iperf client is a Thinkpad X61 Intel Core 2 Due T7100 with an Intel 82566MM Gigabit NIC using Linux Mint 15 (32 bit) installed to SSD.

The Router is a TP-Link WDR4900 running OpenWRT trunk (r37018 from 26/06/2013).

I previously benchmarked the WDR4900 in a bog-standard DHCP with NAT configuration at 394Mbps using this exact same setup of machines.

With PPPoE i managed to average 340Mbps.

Even though this exceeds the current BT maximum speeds for FTTP i don't think this router is suited to such fast connections. While achieving these speeds access to the router itself (using both http and ssh) was extremely sluggish. A "top" showed 97% for sirq. And this is without any QOS or anything else running.

Still, 340Mbps is pretty impressive considering others have reported that a high-end router like the RT-N66u tops out at around 200Mbps so, for anything other than the 330Mbps FTTP product, the WDR4900 running OpenWRT should be more than suitable.

For anyone wishing to replicate my test or test their own routers here is the guide i used to setup the PPPoE server (make sure the user you add to chap-secrets is a valid local user on the machine else you'll get an authentication error. Stumped me for ages).

The iperf commands are simple:

iperf -s

on the server, and

iperf -c SERVER_IP_ADDRESS -P 2 -i 10 -f m -t 600

on the client for a 10 minute test.

Wow - I'm dead impressed!

That's invaluable, and really interesting information, rhetherington. Both on the router performance and the PPPoE test setup.

Certainly a vastly more helpful insight than just saying this or that product is better!

Thank you so much for putting in all the time and effort to figure it out, do the testing and post the results.

The TP Link routers are certainly very good value and seem to do a good job, even if they aren't quite as flashy or feature-rich as the Asus.

One little annoyance I've found with the stock firmware is that you can't turn off STP (which shouldn't generally be needed on a home or SOHO network). The most visible result is that on Android devices the little orange receive data arrow flashes constantly when they are connected to the TL-WDR4300's wireless network, which doesn't sound like much of a problem but is incredibly irritating!

Presumably it also prevents the Android radio entering power saving mode and so drains the battery faster, although I haven't measured that.

It means the LEDs on any connected switch flash constantly too, even when there's no 'real' data on the network. With that and the broadcast UDP packets the TL-WDR4300 sends out every 3 seconds to advertise the printer sharing service, the thing generates a whole 6KB of traffic per minute just doing nothing! I don't think that will really be reducing the throughput, though

From what I read briefly when googling about it, the Asus routers apparently do the same thing.

Thanks, very useful info! I might have to try a similar test on my machines, I have an X61 (same CPU !) lying around here, and a MacBook Air, and a Win7 desktop, so probably use the X61 and the desktop booted from memory stick - nice test!

Hi rhetherington,

Interesting tests, thanks for posting them!

In reply to a post by rhetherington:

For anyone wishing to replicate my test or test their own routers here is the guide i used to setup the PPPoE server (make sure the user you add to chap-secrets is a valid local user on the machine else you'll get an authentication error. Stumped me for ages).

Removing the login parameter from /etc/ppp/pppoe-server-options should solve that:

Text
1 23 45 67 89 1011 1213	# PPP options for the PPPoE Server # LOC: GPL#require-pap require-chap#login lcp-echo-interval 10lcp-echo-failure 2 ms-dns 4.2.2.1ms-dns 4.2.2.3 netmask 255.255.255.0defaultroute noipdefaultusepeerdns

See man pppd:

login Use the system password database for authenticating the peer using PAP, and record the user in the system wtmp file. Note that the peer must have an entry in the /etc/ppp/pap-secrets file as well as the system password database to be allowed access. See also the enable-session option.

cheers, a

Haven't had chance to do any tests myself as yet, but just thought I'd quickly post some other information that came my way in the meantime, in case it is of any use to others.

Firstly, there is apparently a free (for personal use) PPPoE server which can be installed on a bog standard WindowsXP client computer for testing purposes. Doesn't work on later versions, and I can't personally see how to download it from the website, since clicking the "agree" button does nothing for me, but I may be doing something wrong: I only visited briefly. If anyone wants to look into it instead of using Linux or MacOS, it's called RASPPPOE, by Robert Schlabbach.

EDIT: There's a working download link for RASPPPOE on a US site run by Bob Carrick. Also some further instructions which may or may not be helpful here, here and here. All of it only useful if intending to use an old version of Windows for testing, though, which may not give the highest possible throughput, depending on the hardware.

Secondly, re. the default PPPoE MTU of 1480 rather than 1492 (or even 1500, since that extension of the standard would be supported by BT) in the stock TP Link TL-WDR4300 firmware, this is why WindowsXP was originally configured to use that value, according to Microsoft Technet:

In reference to technet.microsoft.com/en-us/library/bb456997.aspx#ECAA:

The 20 bytes of overhead [compared to a standard 1500 byte ethernet frame] consist of the PPPoE header (6 bytes), the largest possible outer PPP header (4 bytes), the largest possible Multilink PPP header (4 bytes), the largest possible PPP header for compression and encryption (4 bytes), and the PPP header that identifies the actual packet being sent (2 bytes).

"Largest possible" being the main explanatory phrase there, I think.

Edited by deleted (Thu 04-Jul-13 18:52:39)

In reply to a post by Kper:

Interestingly, the Hardware NAT feature seems to alter the ToS and DS IP header fields.

Belated correction: this is rubbish, based on an incorrect assumption! Apologies.

Commenting out the login parameter does indeed solve the problem.

I finished configuring my WDR4900 on Friday and decided to run the tests again now that all my settings are done.

Exact same endpoint setup as before.

With the WAN type set to DHCP and using NAT i managed to hit 416Mbps. This is 20Mbps faster than with default settings and the only thing i can think of to account for it is that in my setup i'm no longer bridging the wired and wireless LANs. Instead the two wifi radios are bridged together and then the LAN and WLAN are on two different subnets (/25 each) with routing between them. According to the CeroWRT developers routing does increase performance over bridging, but i had assumed it would only be noticeable on the WLAN. Seems not.

Over PPPoE i got 326Mbps, which isn't too bad considering everything is now fully configured.

Also did a couple of tests using Wifi. Consistent 75Mbps using 2.4GHz 802.11n (20MHz channel) and 125Mbps using 5GHz 802.11n (40MHz channel). I only have two antennas in my Thinkpad, so results may be improved slightly with three. As expected, results remained the same using both DHCP and PPPoE on the router.