b43 reduced performance in Linux-2.6.25

b43 reduced performance in Linux-2.6.25

[Gautam Iyer]

[-- Attachment #1: Type: text/plain, Size: 1042 bytes --]

Hi All,

I saw a few threads about reduced performance in b43 in the 2.6.24 /
2.6.25_rcXX kernels, but couldn't manage to fix the problem myself:

On my system, using ndiswrapper and the windows drivers gives me about
1.5 times the performance of the in kernel b43 drivers.

Here are my system specs:

    Broadcom Corporation BCM4312 802.11a/b/g [14e4:4312] (rev 02)
    Linux 2.6.25 (gentoo-sources-2.6.25-r1).
    b43 wireless driver with firmware 410.2160 (2007-05-26 15:32:10)

I measured the performance by running "ttcp" between my Laptop and my
router. Using b43 on my Laptop gives about 2MB/s reported, and using
ndiswrapper gives me about 3MB/s under exactly the same situations.

Any ideas about what's going on? (I would be happy to provide you with
any further specs on my computer).

Thanks,

GI

-- 
100 THINGS I'D DO IF I EVER BECAME AN EVIL OVERLORD
82. I will not shoot at any of my enemies if they are standing in front
of the crucial support beam to a heavy, dangerous, unbalanced
structure.

[-- Attachment #2: Type: application/pgp-signature, Size: 197 bytes --]

Re: b43 reduced performance in Linux-2.6.25

[Larry Finger]

Gautam Iyer wrote:
> Hi All,
> 
> I saw a few threads about reduced performance in b43 in the 2.6.24 /
> 2.6.25_rcXX kernels, but couldn't manage to fix the problem myself:
> 
> On my system, using ndiswrapper and the windows drivers gives me about
> 1.5 times the performance of the in kernel b43 drivers.
> 
> Here are my system specs:
> 
>     Broadcom Corporation BCM4312 802.11a/b/g [14e4:4312] (rev 02)
>     Linux 2.6.25 (gentoo-sources-2.6.25-r1).
>     b43 wireless driver with firmware 410.2160 (2007-05-26 15:32:10)
> 
> I measured the performance by running "ttcp" between my Laptop and my
> router. Using b43 on my Laptop gives about 2MB/s reported, and using
> ndiswrapper gives me about 3MB/s under exactly the same situations.
> 
> Any ideas about what's going on? (I would be happy to provide you with
> any further specs on my computer).

Yes. Broadcom wrote the Windows drivers, but never gave the specifications to 
anyone else. Obvious those specs from their engineers are better than those 
from the _reverse_ engineers for your card. On my BCM4311/2 I get better 
throughput with the b43 driver than I get when running Windows. I _NEVER_ use 
ndiswrapper. Now that b43 works, why would I want to taint my kernel?

The only loss of performance that we look at is those that happen between 
release n and n+1. Those we call regressions and take them seriously.

The only specs that we are interested in are those that come from reverse 
engineering. If you can provide them, we are very interested.

Larry

Re: b43 reduced performance in Linux-2.6.25

[Gautam Iyer]

[-- Attachment #1: Type: text/plain, Size: 2144 bytes --]

On Tue, Apr 29, 2008 at 05:34:54PM -0500, Larry Finger wrote:

>> I saw a few threads about reduced performance in b43 in the 2.6.24 /
>> 2.6.25_rcXX kernels, but couldn't manage to fix the problem myself:
>> On my system, using ndiswrapper and the windows drivers gives me about
>> 1.5 times the performance of the in kernel b43 drivers.
>> Here are my system specs:
>>     Broadcom Corporation BCM4312 802.11a/b/g [14e4:4312] (rev 02)
>>     Linux 2.6.25 (gentoo-sources-2.6.25-r1).
>>     b43 wireless driver with firmware 410.2160 (2007-05-26 15:32:10)
>> I measured the performance by running "ttcp" between my Laptop and my
>> router. Using b43 on my Laptop gives about 2MB/s reported, and using
>> ndiswrapper gives me about 3MB/s under exactly the same situations.
>> Any ideas about what's going on? (I would be happy to provide you with
>> any further specs on my computer).
>
> Yes. Broadcom wrote the Windows drivers, but never gave the specifications 
> to anyone else. Obvious those specs from their engineers are better than 
> those from the _reverse_ engineers for your card.

Thanks for the explanation! I'm certainly happy that b43 works with good
speeds. Many thanks to you and the other avid hackers that made this
possible...

> On my BCM4311/2 I get better throughput with the b43 driver than I get
> when running Windows. I _NEVER_ use ndiswrapper. Now that b43 works,
> why would I want to taint my kernel?

I don't have Windows so can't compare to it. But even though I use b43
by default, I sometimes have to revert to ndiswrapper on weak signals or
large file transfers. Also ndiswrapper has the power management and
LED's working correctly, which I could not get working with the b43
driver.

> The only specs that we are interested in are those that come from reverse 
> engineering. If you can provide them, we are very interested.

I'll poke around a little with this during Summer. It would certainly
make me a lot happier to use the native b43 drivers than Windows
drivers...

Thanks,

GI

-- 
Will Micro$oft go bankrupt in 1901 because of the Year 2000 Problem?

[-- Attachment #2: Type: application/pgp-signature, Size: 197 bytes --]

Re: b43 reduced performance in Linux-2.6.25

[Larry Finger]

Gautam Iyer wrote:
> 
> I don't have Windows so can't compare to it. But even though I use b43
> by default, I sometimes have to revert to ndiswrapper on weak signals or
> large file transfers. Also ndiswrapper has the power management and
> LED's working correctly, which I could not get working with the b43
> driver.
> 
>> The only specs that we are interested in are those that come from reverse 
>> engineering. If you can provide them, we are very interested.
> 
> I'll poke around a little with this during Summer. It would certainly
> make me a lot happier to use the native b43 drivers than Windows
> drivers...

What LEDS does your interface, and what do they not do? If the only LED you 
have is the one attached to the rfkill switch, then you have not properly 
configured the LED parameters.

Larry

b43 LED status (was Re: b43 reduced performance in Linux-2.6.25)

[Gautam Iyer]

[-- Attachment #1: Type: text/plain, Size: 1165 bytes --]

On Tue, Apr 29, 2008 at 06:54:48PM -0500, Larry Finger wrote:

>> I don't have Windows so can't compare to it. But even though I use b43
>> by default, I sometimes have to revert to ndiswrapper on weak signals or
>> large file transfers. Also ndiswrapper has the power management and
>> LED's working correctly, which I could not get working with the b43
>> driver.
>
> What LEDS does your interface, and what do they not do? If the only
> LED you have is the one attached to the rfkill switch, then you have
> not properly configured the LED parameters.

Yes, the only LED I have is attached to my rfkill switch. It becomes
blue when I load the b43 kernel module, and becomes red when I unload
it. I was hoping I could set it so that it becomes blue when in use
(i.e. associated), and red otherwise. (This happens with ndiswrapper).

I remember messing with the LED power / brightness files in /sys. But
nothing I echoed in there changed the state of the LED. Is there some
parameter I'm missing?

Thanks,

GI

-- 
Twenty Ways To Maintain A Healthy Level of Insanity
7. Finish all your sentences with "In Accordance with the Prophecy."

[-- Attachment #2: Type: application/pgp-signature, Size: 197 bytes --]

Re: b43 LED status (was Re: b43 reduced performance in Linux-2.6.25)

[Larry Finger]

Gautam Iyer wrote:
> On Tue, Apr 29, 2008 at 06:54:48PM -0500, Larry Finger wrote:
> 
>>> I don't have Windows so can't compare to it. But even though I use b43
>>> by default, I sometimes have to revert to ndiswrapper on weak signals or
>>> large file transfers. Also ndiswrapper has the power management and
>>> LED's working correctly, which I could not get working with the b43
>>> driver.
>> What LEDS does your interface, and what do they not do? If the only
>> LED you have is the one attached to the rfkill switch, then you have
>> not properly configured the LED parameters.
> 
> Yes, the only LED I have is attached to my rfkill switch. It becomes
> blue when I load the b43 kernel module, and becomes red when I unload
> it. I was hoping I could set it so that it becomes blue when in use
> (i.e. associated), and red otherwise. (This happens with ndiswrapper).
> 
> I remember messing with the LED power / brightness files in /sys. But
> nothing I echoed in there changed the state of the LED. Is there some
> parameter I'm missing?

You need the following LED- and rfkill-related parameters in your configuration:

CONFIG_MAC80211_LEDS=y
CONFIG_RFKILL=m
CONFIG_RFKILL_INPUT=m
CONFIG_RFKILL_LEDS=y
CONFIG_B43_LEDS=y
CONFIG_B43_RFKILL=y
CONFIG_NEW_LEDS=y
CONFIG_LEDS_CLASS=m
CONFIG_LEDS_TRIGGERS=y

The last one may not be necessary, but the others are. With this setup, the 
LED will be blue when b43 is loaded _AND_ the rfkill switch is on, and red 
when the rfkill switch is off, or b43 is unloaded.

Larry

Re: b43 LED status

[Gautam Iyer]

[-- Attachment #1: Type: text/plain, Size: 2088 bytes --]

On Tue, Apr 29, 2008 at 10:05:34PM -0500, Larry Finger wrote:

>>>> I don't have Windows so can't compare to it. But even though I use b43
>>>> by default, I sometimes have to revert to ndiswrapper on weak signals or
>>>> large file transfers. Also ndiswrapper has the power management and
>>>> LED's working correctly, which I could not get working with the b43
>>>> driver.
>>> What LEDS does your interface, and what do they not do? If the only
>>> LED you have is the one attached to the rfkill switch, then you have
>>> not properly configured the LED parameters.
>> Yes, the only LED I have is attached to my rfkill switch. It becomes
>> blue when I load the b43 kernel module, and becomes red when I unload
>> it. I was hoping I could set it so that it becomes blue when in use
>> (i.e. associated), and red otherwise. (This happens with ndiswrapper).
>> I remember messing with the LED power / brightness files in /sys. But
>> nothing I echoed in there changed the state of the LED. Is there some
>> parameter I'm missing?
>
> You need the following LED- and rfkill-related parameters in your 
> configuration:
>
> CONFIG_MAC80211_LEDS=y
> CONFIG_RFKILL=m
> CONFIG_RFKILL_INPUT=m
> CONFIG_RFKILL_LEDS=y
> CONFIG_B43_LEDS=y
> CONFIG_B43_RFKILL=y
> CONFIG_NEW_LEDS=y
> CONFIG_LEDS_CLASS=m
> CONFIG_LEDS_TRIGGERS=y
>
> The last one may not be necessary, but the others are. With this
> setup, the LED will be blue when b43 is loaded _AND_ the rfkill switch
> is on, and red when the rfkill switch is off, or b43 is unloaded.

Woah! That worked *perfectly*. Super thanks! (It took quite a bit of
searching to figure out B43_RFKILL. There's no kernel prompt for it.
Once I grepped through, enabling INPUT_POLLDEV set this option).

Thanks! It works great now!!

My bandwidth isn't really that much of an issue, since at home my WiFi
is a lot faster than my download speed or the speed of my NAS server.
Good bye ndiswrapper ... :)

GI

-- 
'Psychologist' -- Someone who looks at everyone else when an attractive
woman enters the room.

[-- Attachment #2: Type: application/pgp-signature, Size: 197 bytes --]

Re: b43 reduced performance in Linux-2.6.25

[Gautam Iyer]

[-- Attachment #1: Type: text/plain, Size: 2423 bytes --]

On Tue, Apr 29, 2008 at 05:34:54PM -0500, Larry Finger wrote:

>> I saw a few threads about reduced performance in b43 in the 2.6.24 /
>> 2.6.25_rcXX kernels, but couldn't manage to fix the problem myself:
>> On my system, using ndiswrapper and the windows drivers gives me about
>> 1.5 times the performance of the in kernel b43 drivers.
>> Here are my system specs:
>>     Broadcom Corporation BCM4312 802.11a/b/g [14e4:4312] (rev 02)
>>     Linux 2.6.25 (gentoo-sources-2.6.25-r1).
>>     b43 wireless driver with firmware 410.2160 (2007-05-26 15:32:10)
>> I measured the performance by running "ttcp" between my Laptop and my
>> router. Using b43 on my Laptop gives about 2MB/s reported, and using
>> ndiswrapper gives me about 3MB/s under exactly the same situations.
>> Any ideas about what's going on? (I would be happy to provide you with
>> any further specs on my computer).
>
> Yes. Broadcom wrote the Windows drivers, but never gave the
> specifications to anyone else. Obvious those specs from their
> engineers are better than those from the _reverse_ engineers for your
> card. On my BCM4311/2 I get better throughput with the b43 driver than
> I get when running Windows. I _NEVER_ use ndiswrapper. Now that b43
> works, why would I want to taint my kernel?

I just moved to a different network and compared the performance between
b43 and ndiswrapper. They are now *almost* identical (both 2.5MB/s). I'd
even say that b43 sometimes performs slightly (50KB/s) better on
average! (GO LINUX WIRELESS!)

Looking at my syslog shows that on my home network (where b43 performed
50% worse), I find the following messages just after association:

    wlan0: switched to short barker preamble (BSSID=00:14:a5:0c:17:dc)
    wlan0: WMM queue=2 aci=0 acm=0 aifs=3 cWmin=15 cWmax=1023 burst=0
    wlan0: WMM queue=3 aci=1 acm=0 aifs=7 cWmin=15 cWmax=1023 burst=0
    wlan0: WMM queue=1 aci=2 acm=0 aifs=2 cWmin=7 cWmax=15 burst=30
    wlan0: WMM queue=0 aci=3 acm=0 aifs=2 cWmin=3 cWmax=7 burst=15

I don't find these messages when I connect on campus (and get the high
performance). So maybe the performance loss is WMM related.

My home router is of course running Linux (OpenWRT), so maybe I can
tweak the WMM parameters of the router to match what b43 is happy with!

Any suggestions,

Thanks again for your help!

GI

-- 
'Pessimist' -- Optimist with experience.

[-- Attachment #2: Type: application/pgp-signature, Size: 197 bytes --]