Jaunty ath5k poor performance

[Chris Bainbridge]

As mentioned in bug #292054 the performance of the ath5k driver is very poor. This was tested under Jaunty. Comparison points are 1) time to associate with an AP 2) bandwidth on large file transfer from test laptop to local server 3) ping time whilst large file transfer is in progress. Results:

1) Comparison of time to associate with an AP, comparing ath5k to zd1211rw on otherwise identical hardware, results show a statistically significant mean time (from starting NetworkManager to binding an IP address) of 21 seconds versus 9 seconds respectively. Even accounting for minor hardware differences (Atheros versus Zydas chipset) the performance difference should not be so large.

2) Send large file. 100MB file transfer from laptop, comparing ath5k to ath_pci. Transfer is to a local unused PC which is directly connected to the wireless AP via ethernet. ath5k mean bandwidth is 800kb/sec, ath_pci bandwidth is 1.9MB/sec.

3) Comparison of ping time whilst transferring a 100MB file, comparing Atheros ath5k to Zydas zd1211rw on otherwise identical hardware. With zd1211rw, ping is from 1ms to 35ms, bandwidth is 2.1MB/s. With ath5k, ping is from 1ms to 3500ms, bandwidth is 573KB/s.

4) Receive large file. 100MB file transfer to laptop. ath_pci bandwidth is 2.2MB/s ath5k bandwidth is 1.5MB/s.

It seems clear that there are performance problems with ath5k, not only comparing against other hardware and drivers, but also some regression from the old ath_pci driver - ath5k is faster to associate than ath_pci, which took around 60s, but bandwidth is slower.

Causes of the problem:

1) Bugs that have been already fixed in ath5k. The mean receive bandwidth of compat-wireless-2009-03-03 is 1.43MB/s, compared to 1.04MB/s for the 2.6.28-8 driver. The difference is highly statistically significant, so this is certainly a problem. The drivers and/or kernel should be updated to fix this.

2) According to http://wireless.kernel.org/en/users/Drivers/ath5k "MadWiFi and ndis drivers perform better on noisy environments than ath5k because we don't support Adaptive Noise Immunity (ANI) yet." This may account for other performance loss.

[xteejx (xteejx)]

I don't see hwy this was marked as the main one, its simply another bug report summarising 4 others, I believe my one bug 331822 should be the main as there is a workaround for the ath5k problem which may be missed if this is main.

[Chris Bainbridge (chris-bainbridge)]

ath5k is the open source driver supplied with the linux kernel. The option of installing the old ath_pci driver is not a great one - it is mostly unsupported upstream and essential stuff like suspend/resume is broken. Installing ndiswrapper and the Window drivers which is mentioned in the ubuntuforums page that you linked to isn't a great option either.

This bug report does not merely summarise the other bugs - this report provides clear information on actual testing that has been carried out, and the information that there are already changes upstream that provide validated improvements.

The dupe bugs may not actually be dupes - they looked similar, but if updating to the latest open source driver doesn't help your situation, then you may have a completely different problem. In that case, feel free to un-dupe the other bug report.

[xteejx (xteejx)]

I already have Jaunty alpha all latest everything, including ath5k, and I'm
experiencing similar problems. When there is high bandwidth usage the
connection goes all crappy, for want of a better word. Also I seem to have
stumbled on another reason causing it...positioning, which could be down to
low power settings, or driver issues - i.e. when the laptop is positioned
facing towards the router, the connection is barely usable, but when tilted
slightly or put side on, it works at near full speed. This problem doesn't
occur in Vista, which makes me think its a power setting and/or driver
problem. This should be enough information to determine if this is a dupe, I
don't think it is. Thanks

[Slobodan Simić (slsimic)]

Mine wireless is cripled too.
I had normal speed with my laptop's wireless card in Intrepid but since I installed Jaunty beta I have poor speed. It jumps, stops then for few secs is normal and the all over again. As I said, no such issue with ath5k in Interpid nor from some others (Live) distros.

[Chris Bainbridge (chris-bainbridge)]

Teej: "MadWiFi and ndis drivers perform better on noisy environments than ath5k because we don't support Adaptive Noise Immunity (ANI) yet." (& tx-power calibration) - possibly this is the cause of your problem, since re-orienting or moving the laptop will affect the noise and requiring tx power.

Slobodan: Installing the latest from compat-wireless http://linuxwireless.org/download/compat-wireless-2.6/ may provide some performance improvement - for me it stopped the stuttering connection you describe.

[Paul Weiss (interweiss)]

The ath5k driver is causing intermittent wireless freezes in wine when playing Warcraft 3. I since switched back to the madwifi driver, and now everything is fine.

[Chris Bainbridge (chris-bainbridge)]

Wine/Warcraft3 is just an application - if you have freezes with it, then you probably also have problems with other applications - it's just less likely that you'll notice them.

[Gregory P Smith (gpshead)]

installing the linux-backports-modules-jaunty package for its more up to date ath5k driver helps significantly.

[flasbang73 (flasbang73)]

Thank you Gregory that helped me I was having problems with this for a while I appreciate it

[xteejx (xteejx)]

Marking Fix Released. This is solvable by installing linux-backports-jaunty as stated above. Thank you.

[Frederik Elwert (frederik-elwert)]

I cannot confirm this. I have installed linux-backports-modules-jaunty, but performance only increased little. Also the vanilla kernel 2.6.30 didn’t solve the issue for me.

In support forums, I have both heard people where the backports fixed the issue, and some where they didn’t. So there seem to be some setups where the contained fixes are not suffcient to restore full performance.

[xteejx (xteejx)]

Perhaps this is dependant on the actual model of wifi card/dongle, and only partially related to the ath5k driver itself. Of course distance, EFI, obstacles are a factor in wifi signals, but I would've expected there to be a significant increase as Gregory described.

Frederick, Gregory and flasbang, can you all provide the model you are using to confirm this please?

Thank you.

[Frederik Elwert (frederik-elwert)]

Yes, sure.

$ lspci | grep -i wireless
02:00.0 Ethernet controller: Atheros Communications Inc. AR242x 802.11abg Wireless PCI Express Adapter (rev 01)

That is that silly little thing that previously required a special version of the madwifi driver (the last release I know of was madwifi-nr-r3366+ar5007). Now ath5k supports these devices out of the box, but with a much worse performance compared to the previous madwifi releases.

$ modinfo ath5k
filename: /lib/modules/2.6.28-13-generic/updates/ath5k.ko
version: 0.6.0 (EXPERIMENTAL)

[xteejx (xteejx)]

Thank you
Can you confirm what model of WiFi card you're using please?
Also if Gregory and flasbang can provide these details it will help.
Also, are you using WEP WPA etc?

[Frederik Elwert (frederik-elwert)]

Oh, sorry, then I misunderstood your question. I thought you needed that lspci information. How do I find out the model of the card?

And I’m using WPA.

[xteejx (xteejx)]

The model number for the WiFi card you're using should be on the box
that you bought it in. If the other two users could also provide the
same information as requested above, it would be very helpful.

[Gregory P Smith (gpshead)]

my device is a Asus Eee 900A. I am using WPA2 AES with a shared key.

lspci & lspci -v report it as:

02:00.0 Ethernet controller: Atheros Communications Inc. AR242x 802.11abg Wireless PCI Express Adapter (rev 01)
Subsystem: Device 1a3b:1026

My most often used AP is a linksys wrt600n running dd-wrt. I am connecting using 2.4ghz (802.11g).

Ever since installing the backported modules I have had zero wireless problems with my device, regardless of location or signal strength.

(fwiw, wireless card's don't normally come in boxes. They've been integrated into every portable device made in the last 5 years. I really wonder what you're asking for there. Sure people can buy mini-pci and mini-pcie wireless adapters but -most- people never need to know that.)

[xteejx (xteejx)]

I think they should be left as fix released, as it is advised to use the proprietary driver not the open source one in this case, since the open source driver does not provide enough stability. If after changing you are still experiencing severe problems, then either report back on here or for a quicker response please go to #ubuntu or #ubuntu--bugs on IRC. Thank you.

[ullix (ullix)]

what actually is the metric for poor performance? I can offer this: computer1 is connected to my local net via fast ethernet. It has a file of 500MB containing pseudo random numbers. I access this from computer2 via WLAN using:

scp 10.0.0.30:/home/test500mb /dev/null

and get these results for transfer rates:
 Easy Peasy 1.1 (based on Ubuntu 8.04):____ 1.1 MB/s
 Ubuntu Jaunty 9.04:_________________________ 2.6-2.9 MB/s
 Ubuntu Karmic alpha: ______________________ 2.2-2.4 MB/s

Although the physical locations of the computers/router were almost the same, I am not sure whether the difference between Jaunty and Karmic is driver related or caused by something else, but the difference to 8.04 is striking.

computer2 is Acer Aspire One 110L, with the WLAN chip reported under Karmic as:

lspci -v -nn -s 03:00
03:00.0 Ethernet controller [0200]: Atheros Communications Inc. AR5001 Wireless Network Adapter [168c:001c] (rev 01)
 Subsystem: Foxconn International, Inc. Device [105b:e008]
 Flags: bus master, fast devsel, latency 0, IRQ 18
 Memory at 35200000 (64-bit, non-prefetchable) [size=64K]
 Capabilities: [40] Power Management version 2
 Capabilities: [50] Message Signalled Interrupts: Mask- 64bit- Queue=0/0 Enable-
 Capabilities: [60] Express Legacy Endpoint, MSI 00
 Capabilities: [90] MSI-X: Enable- Mask- TabSize=1
 Capabilities: [100] Advanced Error Reporting <?>
 Capabilities: [140] Virtual Channel <?>
 Kernel driver in use: ath5k
 Kernel modules: ath5k

[Frederik Elwert (frederik-elwert)]

Am Mittwoch, den 15.07.2009, 15:37 +0000 schrieb Teej:
> I think they should be left as fix released, as it is advised to use the
> proprietary driver not the open source one in this case, since the open
> source driver does not provide enough stability. If after changing you
> are still experiencing severe problems, then either report back on here
> or for a quicker response please go to #ubuntu or #ubuntu--bugs on IRC.

The proprietary driver (when activated from Ubuntu’s hardware driver
dialog) doesn’t work at all for my hardware. After a reboot, the
wireless device doesn’t even show up in ifconfig or Network Manager.

So I do rely on the open source driver, and it does at least work, but
not very well.