I have just been reading the review on smallnetbuilder.com
It was interesting to see how badly the routing performance was hit using DD-WRT. I was also interested in the wireless performance test where the "Spectrum Expert card and software" were used, I am surprised there was no comment on just how much potential "interfering" noise was generated by tweaking the o/p power to max in DD-WRT.
Careful... Tim ran a test that simultaneously tested both the wireless and router performance. I can't tell from the results whether the rather drastic drop in performance was due to slothish packet processing and filtering, or due to the wireless. What worries me more is that the router crashes when too many streams are run simultaneously.
The TX power test was interesting but not very useful. At 250mw, the xmitter probably goes out of spec for the FCC spectrum mask. It also generates some non-linearities that are going to have an effect on speed. Worse, it creates an "alligator" where the TX range is much more than the RX range. It does no good to crank up the TX power, if you can't hear the responses from the other end of the link. It would take *TWO* WNR3500L boxes, acting as a bridge (i.e. no routing), where the transmit power is cranked up on both bridge boxes, to do a proper test.
The maker of the $4,000 Spectrum Expert Card, Cognio, was purchased by Cisco and is now "Cisco Spectrum Expert".
Despite my usual pot shots about methodology and conclusions, the tests performed by Tim Higgins on SmallNetBuilder.com are the best (and most creative) that I've seen anywhere and should be considered seriously.
A list has been started on DD-WRT of throughput tests for "N" products .
There was also this on the forum "It really depends on what features you use. If you run anything that puts a high load on the CPU then it will cut into your routing throughput. For instance, if you set up a VPN tunnel that's encrypted then using the tunnel will put a lot of load on the CPU. If you're running default settings then it shouldn't be as drastic as in that article, somewhere around 30% less throughput off the top of my head." "
30% drop in throughput expected.
It was a pity he didn't use the same settings that are used for testing the 802.11 "Spectral Mask" but I would guess the "Spectrum Card" was on the limit and went into auto RBW.
I was more interested in the possible consequences to nearby networks rather than whether it improved a link in a network.
If you filter on Netgear, you'll see the results for the WNR3500L and others.
Thanks, but not very useful. The 2MByte window is cheating as many IP device stacks are limited to the same as ethernet (1500 bytes). The exact Iperf/Jperf command lines should be supplied so that the test are reproducible. Testing at 1 meter range is not exactly indicative of "typical" operations.
How SmallNetBuilder tests:
There should also be a test for: - Ethernet to ethernet to prove the wire speed is possible. - LAN ethernet to wireless bridging (not through router) Up/Down. - WAN ethernet to wireless routing Up/Down. - Wireless to Wireless bridging Up/Down. The last is amazingly lousy on some devices I've tested.
This might be more informative:
WRN3500L (not DD-WRT) results table:
Incidentally, the maximum simultaneous connections test results are kinda amusing:
This has not been a problem as the routers usually limit the number of available IP addresses to 253. However, when used as a bridge, the approx 200 connection limit is fatal as many more MAC addresses are used. Many routers also don't do a proper job of clearing the MAC address table from old addresses to make room for new addresses. Extra credit for those that hang when the limit is reached.
Well, yes... that's true. Terminating the VPN in the router is a great way to overload the CPU. Excessive packet filtering and VLAN's also has a big effect. Multiple SSID's seems to slow the wireless speeds a little. For the WRT54G series of routers, CPU overload is a serious problem. However, the later routers have much faster and more efficient processors. I don't know what the WRN3500L uses, but I would assume is faster than the older Broadcom chips.
I would have settled for a spectrum analyzer photo at various power levels with the spectrum mask overlay, as in the FCC Part 19 tests. Incidentally, some of the newer PA chips are now using cellular PA tricks of pre-distorting the AM linearity in order to get more power at less distortion. I was more interested in the possible consequences to nearby networks
I'll be blunt. I think that 40MHz on 2.4Ghz was a bad idea and should be banned. It works well on 5.7GHz. I also have some nasty things to say about channel bonding between the two bands. The entire effort is only to get more speed, something most users don't need or want. If this additional speed is at the price of additional interference, it's a bad tradeoff.
I'm not too worried about 802.11n 40MHz interference. If the rules/guidelines are followed, where 40MHz is only allowed in a "greenfield" environment on 2.4GHz, any interference will immediately switch everyone to 20MHz mode.
I just blundered across this paper on 802.11n interference: "802.11n Under the Microscope"
Apparently, it doesn't take much to mangle 802.11n without or without channel bonding. We show that the throughput of an 802.11n link can be severely degraded (up to 85%) in presence of an 802.11g link. Our results also indicate that increased amount of interference due to wider channel bandwidths can lead to throughput degradation. In other words, just about any form of interference will trash 802.11n thruput. Most of the N boxes I've played with will revert to the slower 802.11g speeds at the first hint of any co-channel users (interfering or not including adjacent channel spillover).