WiFi transfer rates (Linux->Linux)
Hi, Wifi is not something I consider myself an expert upon. I understand the security problems with WEP, WPS, WPA and WPA2 fairly well, and I've read some docs on fragmentation, RTS thresholds, and channel width. I don't consider myself a newbie here either. Try as I might, I just can't seem to get decent performance out of my home network. It's a 802.11n network, theoretically 300 mbit/s, however testing with iperf between a Linux wireless laptop and a wired linux server results in 60-70 mbits reported in either direction. I tried fiddling with a few options on the router, such as: * disabling a/b/g support. (no effect) * disabling pre-spec-N support. (no effect) * lowering RTS threshold slightly (lower throughput) * lowering fragment threshold (no effect at small changes; perf slightly increases (~5mbit) as fragments decrease for a while, then after a point (around ~2000 bytes) performance just drops as you decrease size.) * WMM no-ack mode (totally wrecked performance) I'm using a Broadcom chipset in the laptop, which seems to have reasonable Linux support. I'm using the opensource brcmsmac driver on a 3.2.0 kernel. The AP is an Asus RT-N15, which also runs Linux, although I'm still on the vendors firmware. So, no matter what I've done, I can't get more than 75 mbits, and on default settings its more like 65 mbits between these two Linux devices. However a friend with an Apple Mac laptop and an Apple airport AP can push about 200 mbit/s according to iperf! So.. I just wondered if anyone has some hints about how to improve wifi performance? I also wondered what sort of bitrates you see? And if someone is getting good speeds, can you report your hardware (incl. chipset) and configuration options? thanks! Toby
Your limit is simply that the RT-N15 is a 2.4GHz only radio The 300Mb/s advertised is one of those wonderful if only no one else used it and you had 5 more channels with no overlap to bond together. this will never be seen on 2.4GHz. 2.4GHz has 3 non overlapping channels and if 802.11N with channel bonding is used 1. The airport on the other hand is an ABGN simultaneous Dual band radio that can bond several 5GHz channels together (less cross talk on these channels (and much less pollution)) to be honest it's kinda amazing that you get 70Mb/s on 2.4GHz :) I run a cisco ap and abgn card in a separate AP and see sustained 250Mb/s link speed over that On Tue, Aug 21, 2012 at 12:17 PM, Toby Corkindale <toby.corkindale@strategicdata.com.au> wrote:
Hi, Wifi is not something I consider myself an expert upon. I understand the security problems with WEP, WPS, WPA and WPA2 fairly well, and I've read some docs on fragmentation, RTS thresholds, and channel width. I don't consider myself a newbie here either.
Try as I might, I just can't seem to get decent performance out of my home network. It's a 802.11n network, theoretically 300 mbit/s, however testing with iperf between a Linux wireless laptop and a wired linux server results in 60-70 mbits reported in either direction.
I tried fiddling with a few options on the router, such as: * disabling a/b/g support. (no effect) * disabling pre-spec-N support. (no effect) * lowering RTS threshold slightly (lower throughput) * lowering fragment threshold (no effect at small changes; perf slightly increases (~5mbit) as fragments decrease for a while, then after a point (around ~2000 bytes) performance just drops as you decrease size.) * WMM no-ack mode (totally wrecked performance)
I'm using a Broadcom chipset in the laptop, which seems to have reasonable Linux support. I'm using the opensource brcmsmac driver on a 3.2.0 kernel. The AP is an Asus RT-N15, which also runs Linux, although I'm still on the vendors firmware.
So, no matter what I've done, I can't get more than 75 mbits, and on default settings its more like 65 mbits between these two Linux devices. However a friend with an Apple Mac laptop and an Apple airport AP can push about 200 mbit/s according to iperf!
So.. I just wondered if anyone has some hints about how to improve wifi performance?
I also wondered what sort of bitrates you see? And if someone is getting good speeds, can you report your hardware (incl. chipset) and configuration options?
thanks! Toby _______________________________________________ luv-main mailing list luv-main@luv.asn.au http://lists.luv.asn.au/listinfo/luv-main
On 21/08/12 12:50, Kevin wrote:
Your limit is simply that the RT-N15 is a 2.4GHz only radio
The 300Mb/s advertised is one of those wonderful if only no one else used it and you had 5 more channels with no overlap to bond together. this will never be seen on 2.4GHz. 2.4GHz has 3 non overlapping channels and if 802.11N with channel bonding is used 1.
I realise there isn't much spectrum available on 2.4 GHz, but scanning in my home only shows a couple of other visible networks, and they're both at around -85 to -90 db, vs my network at -35ish. I hope the significantly more powerful local signal will be rendering the other networks irrelevant for interference. Does it not work that way?
The airport on the other hand is an ABGN simultaneous Dual band radio that can bond several 5GHz channels together (less cross talk on these channels (and much less pollution))
to be honest it's kinda amazing that you get 70Mb/s on 2.4GHz :)
I run a cisco ap and abgn card in a separate AP and see sustained 250Mb/s link speed over that
Ah well, maybe I can improve things by simply throwing money at the problem then. Can you recommend a 5 Ghz AP with reasonable price/performance? (I don't have any issues with simply adding a PCI card to the linux server, but chipset support for AP mode seems rather hit-and-miss) thanks, Toby
Kevin <kevin@fuber.org> wrote:
Your limit is simply that the RT-N15 is a 2.4GHz only radio
The 300Mb/s advertised is one of those wonderful if only no one else used it and you had 5 more channels with no overlap to bond together. this will never be seen on 2.4GHz. 2.4GHz has 3 non overlapping channels and if 802.11N with channel bonding is used 1.
The airport on the other hand is an ABGN simultaneous Dual band radio that can bond several 5GHz channels together (less cross talk on these channels (and much less pollution))
I've heard positive comments regarding the The Netgear WNDR3700v2 wireless router, which has the advantage of open drivers. This is why it was discussed at LWN in relation to efforts to solve the buffer bloat problem. A quick Web search indicates that it includes a 5GHZ-capable radio, but I didn't dig deeply enough to find reviews of its wireless performance. A friend has one with OpenWRT installed and seems satisfied.
On 21/08/12 14:42, Jason White wrote:
I've heard positive comments regarding the The Netgear WNDR3700v2 wireless router, which has the advantage of open drivers. This is why it was discussed at LWN in relation to efforts to solve the buffer bloat problem.
This is likely the article in question about CeroWrt.. http://lwn.net/Articles/508783/ -- Chris Samuel : http://www.csamuel.org/ : Melbourne, VIC
Jason White wrote:
I've heard positive comments regarding the The Netgear WNDR3700v2 wireless router, which has the advantage of open drivers.
+1, I run one of these at home. OpenWRT currently recommends either it or the cheaper TP-Link 1043ND. The WNDR3700 has the advantage of being reflashable via TFTP without having to open the case and fiddle with headers, which is primarily useful if you are building custom (and potentially bricktastic) custom OS images. I operate three 1043NDs at work and they do their job adequately.
A quick Web search indicates that it includes a 5GHZ-capable radio, but I didn't dig deeply enough to find reviews of its wireless performance.
NFI about that, I don't operate anything at home except the 3700 and a DSL modem, and I barely notice even when I'm throttled to 128kbps, so I'm not going to notice any difference between 70mbps and 300mbps :-) * * * I don't think anyone excplicitly mentioned but AIUI you should check what other channels people are operating on, and pick one that no one else is using. Also there is an overlap of one channel to either side of the chosen channel, so that if e.g. someone is using channel "5", then you should not pick 8, so that their {4,5,6} and your {7,8,9} don't overlap. When I was doing this in the office, $coworker had a little graphical doodad for this on their android (or maemo?) phone.
On 21/08/12 15:15, Trent W. Buck wrote:
I don't think anyone excplicitly mentioned but AIUI you should check what other channels people are operating on, and pick one that no one else is using. Also there is an overlap of one channel to either side of the chosen channel, so that if e.g. someone is using channel "5", then you should not pick 8, so that their {4,5,6} and your {7,8,9} don't overlap. When I was doing this in the office, $coworker had a little graphical doodad for this on their android (or maemo?) phone.
It's not just one channel to either side! Each wifi "channel" takes up five channels of space, two to either side. ie. channel 6 will also consume 4,5,7,8. See: http://www.tech-juice.org/wp-content/uploads/2011/04/2.4_GHz_Wi-Fi_channels_... As mentioned earlier in this thread, there are only three possible positions that don't interfere with each other- 1, 6, 11. An unfortunately if anyone is using the 300 mbit N stuff, that takes up two thirds of the band! I have one of the graphical wifi monitoring apps and have indeed used it to try and position mine away from the others.. However since both other networks are 40MHz ones, that consumes the entire band :)
Toby Corkindale wrote:
On 21/08/12 15:15, Trent W. Buck wrote:
I don't think anyone excplicitly mentioned but AIUI you should check what other channels people are operating on, and pick one that no one else is using. Also there is an overlap of one channel to either side of the chosen channel, [...]
It's not just one channel to either side! [Details...]
Ah, sorry; I did bother to fact-check before posting. I defer to my learnèd correspondent :-)
Hi, Wifi is not something I consider myself an expert upon. I understand the security problems with WEP, WPS, WPA and WPA2 fairly well, and I've read some docs on fragmentation, RTS thresholds, and channel width. I don't consider myself a newbie here either.
Try as I might, I just can't seem to get decent performance out of my home network. It's a 802.11n network, theoretically 300 mbit/s, however testing with iperf between a Linux wireless laptop and a wired linux server results in 60-70 mbits reported in either direction.
What is the wireless chipset in the Linux laptop? Is the router connected via gigabit all the way to the server? If it's only 100MBits/second then that's all you get (wouldn't be the first wireless router I've seen advertising 300MBits/second but only having 100MBits/second Ethernet!) What about if you wire the laptop to the router as a test? Can you get gigabit speeds then? James
Is the router connected via gigabit all the way to the server? If it's only 100MBits/second then that's all you get (wouldn't be the first wireless router I've seen advertising 300MBits/second but only having 100MBits/second Ethernet!)
I just re-read your original email then googled and google tells me that the Asus RT-N15 is gigabit capable so unless you have something stuck in 100mbit/second mode this probably won't be your problem. James
On 21/08/12 14:09, James Harper wrote:
Is the router connected via gigabit all the way to the server? If it's only 100MBits/second then that's all you get (wouldn't be the first wireless router I've seen advertising 300MBits/second but only having 100MBits/second Ethernet!)
I just re-read your original email then googled and google tells me that the Asus RT-N15 is gigabit capable so unless you have something stuck in 100mbit/second mode this probably won't be your problem.
I have a gigabit ethernet card in the Linux server with CAT-5E cable, and have previously tested the wired performance of the Linux laptop to server. Over gigabit ethernet (using the RT-N15 as switch) the performance is as expected.
On 21/08/12 13:59, James Harper wrote:
Hi, Wifi is not something I consider myself an expert upon. I understand the security problems with WEP, WPS, WPA and WPA2 fairly well, and I've read some docs on fragmentation, RTS thresholds, and channel width. I don't consider myself a newbie here either.
Try as I might, I just can't seem to get decent performance out of my home network. It's a 802.11n network, theoretically 300 mbit/s, however testing with iperf between a Linux wireless laptop and a wired linux server results in 60-70 mbits reported in either direction.
What is the wireless chipset in the Linux laptop?
It's a Lenovo-branded Broadcom chipset in a Thnkpad using the brcmsmac driver; I don't have the specifics handy but can reply with them later. TC
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 On 21/08/12 12:17, Toby Corkindale wrote:
Try as I might, I just can't seem to get decent performance out of my home network. It's a 802.11n network, theoretically 300 mbit/s, however testing with iperf between a Linux wireless laptop and a wired linux server results in 60-70 mbits reported in either direction.
Are you using 5 GHz? You probably won’t get 300 Mbit/s out of 2.4 GHz. (Wider channel widths mean you are far more vulnerable to interference.) Also keep in mind wireless speeds are measured half-duplex, so you only get at most half of what they say. I’ve never got more than 25 Mbit/s on 802.11g. I have a 5 GHz 802.11n AP (a HP MSM422 if you must know), my MacBook Air that says its TX rate is 270 Mbit/s. Running iperf, I get no more than 150 Mbit/s. (The laptop’s RX is far worse because I set the AP’s multicast TX rate ridiculously high to cope with multicast HDTV.) I live in a country area in the sticks. Using a Ubiquti NanoStation M5 as a spectrum analyser, I see no other 5 GHz activity whatsoever. I suspect you may be suffering from interference, which is possible in both the 2.4 GHz and 5 GHz bands from consumer devices. Also do you have devices connected to the AP that only support 20 MHz channel widths? Just a stab in the dark, but that may be bogging down the 40 MHz transmissions. -----BEGIN PGP SIGNATURE----- Version: GnuPG/MacGPG2 v2.0.17 (Darwin) Comment: GPGTools - http://gpgtools.org Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/ iEYEARECAAYFAlAzSfkACgkQvs6Qqs8TxBoPOwCfYlkHIOAa326quhetFvFAJtix EyEAn3i8mrX1Qr+54YfAz9cW8YB+zuc3 =fZGx -----END PGP SIGNATURE-----
participants (7)
-
Chris Samuel -
James Harper -
Jason White -
Jeremy Visser -
Kevin -
Toby Corkindale -
Trent W. Buck