rtl8187 - nice work!

rtl8187 - nice work!

[P.G. Richardson]

Hi,

Installed the compat-wireless-2009-03-10 and the speed has greatly
improved. Thanks very much.

I do have some observations that might be a bug or just configuration.

When powering up, the wireless decided on 54M as its bit rate.
Consequently, tcp was patchy at best - losing packets whilst pinging, s=
sh
failing to connect. Setting the rate back to a fixed 11M solves this an=
d
gives good speed. Interestingly, performance declines as I up the bit r=
ate
from 11 to 18 etc...

Of course, simple specifying the RATE in the ifcfg-wlan0 as 11M sorts t=
his
perfectly.

Anyway, thanks and regards

PGR

--=20
Laws are partly formed for the sake of good men, in order to instruct t=
hem
how they may live on friendly terms with one another, and partly for th=
e
sake of those who refuse to be instructed, whose spirit cannot be subdu=
ed,
or softened, or hindered from plunging into evil. [The Laws, Plato]

You fiend! Never have I encountered such corrupt and foul-minded
perversity=85 Have you ever considered a career in the church?
Bishop of Bath and Wells (Blackadder II)

--
To unsubscribe from this list: send the line "unsubscribe linux-wireles=
s" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: rtl8187 - nice work!

[Larry Finger]

P.G. Richardson wrote:
> Hi,
> 
> Installed the compat-wireless-2009-03-10 and the speed has greatly
> improved. Thanks very much.
> 
> I do have some observations that might be a bug or just configuration.
> 
> When powering up, the wireless decided on 54M as its bit rate.
> Consequently, tcp was patchy at best - losing packets whilst pinging, ssh
> failing to connect. Setting the rate back to a fixed 11M solves this and
> gives good speed. Interestingly, performance declines as I up the bit rate
> from 11 to 18 etc...

It is probably a bug. My Netgear WG111V2 with an RTL8187L chip is perfectly
happy running at 48 or 54 Mb/s with a transmit throughput of 21-23 Mb/s - close
to the maximum for an 802.11g network. There are, however, other reports like
yours where the OFDM rates (6, 9, 18, 24, 36, 48, and 54 Mb/s) don't work very
well. It seems that not all 8187's are the same.

What rate-setting algorithm are you using? Use 'dmesg | grep rate' to see that.
I would have expected it to have settled into 11 Mb/s. Perhaps the auto retry
hasn't really been turned off.

Larry

Re: rtl8187 - nice work!

[P.G. Richardson]

>> Installed the compat-wireless-2009-03-10 and the speed has greatly
>> improved. Thanks very much.
>>
>> I do have some observations that might be a bug or just configuratio=
n.
>>
>> When powering up, the wireless decided on 54M as its bit rate.
>> Consequently, tcp was patchy at best - losing packets whilst pinging=
,
>> ssh
>> failing to connect. Setting the rate back to a fixed 11M solves this=
 and
>> gives good speed. Interestingly, performance declines as I up the bi=
t
>> rate
>> from 11 to 18 etc...
>
> It is probably a bug. My Netgear WG111V2 with an RTL8187L chip is
> perfectly
> happy running at 48 or 54 Mb/s with a transmit throughput of 21-23 Mb=
/s -
> close
> to the maximum for an 802.11g network. There are, however, other repo=
rts
> like
> yours where the OFDM rates (6, 9, 18, 24, 36, 48, and 54 Mb/s) don't =
work
> very
> well. It seems that not all 8187's are the same.
>
> What rate-setting algorithm are you using? Use 'dmesg | grep rate' to=
 see
> that.
> I would have expected it to have settled into 11 Mb/s. Perhaps the au=
to
> retry
> hasn't really been turned off.

dmesg says ->
phy0: Selected rate control algorithm 'pid'

which is a first as it has always been 'mistral'. The only difference w=
ith
compiling last night was I was not running the kernel I compiled agains=
t.
I used 'make KLIB=3D/usr/src/{my new kernel} KLIB_BUILD=3D/usr/src/{my =
new
kernel}' instead.

Just recompiled while running {my new kernel} and mistral correctly
selected as rate control algorithm. Once installed, it does correctly a=
uto
select rate as expected.

iperf shows Mb performance has dropped though but gets improved by agai=
n
fixing the rate at 11Mb.

So cheers for the info.

Regards

PGR

--=20
Laws are partly formed for the sake of good men, in order to instruct t=
hem
how they may live on friendly terms with one another, and partly for th=
e
sake of those who refuse to be instructed, whose spirit cannot be subdu=
ed,
or softened, or hindered from plunging into evil. [The Laws, Plato]

You fiend! Never have I encountered such corrupt and foul-minded
perversity=85 Have you ever considered a career in the church?
Bishop of Bath and Wells (Blackadder II)

--
To unsubscribe from this list: send the line "unsubscribe linux-wireles=
s" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: rtl8187 - nice work!

[Larry Finger]

[-- Attachment #1: Type: text/plain, Size: 952 bytes --]

P.G. Richardson wrote:
> phy0: Selected rate control algorithm 'pid'
> 
> which is a first as it has always been 'mistral'. The only difference with
> compiling last night was I was not running the kernel I compiled against.
> I used 'make KLIB=/usr/src/{my new kernel} KLIB_BUILD=/usr/src/{my new
> kernel}' instead.
> 
> Just recompiled while running {my new kernel} and mistral correctly
> selected as rate control algorithm. Once installed, it does correctly auto
> select rate as expected.
> 
> iperf shows Mb performance has dropped though but gets improved by again
> fixing the rate at 11Mb.
> 
> So cheers for the info.

Minstrel is definitely a better choice than pid.

There is one thing that might make your interface work better. For performance
reasons, I chose not to pause the mac80211 queues at one point in the driver.
For my device, it did not make a difference, but it might for you. Please try
the attached patch.

Thanks,

Larry


[-- Attachment #2: rtl8187_queue_pause --]
[-- Type: text/plain, Size: 736 bytes --]

Index: wireless-testing/drivers/net/wireless/rtl818x/rtl8187_dev.c
===================================================================
--- wireless-testing.orig/drivers/net/wireless/rtl818x/rtl8187_dev.c
+++ wireless-testing/drivers/net/wireless/rtl818x/rtl8187_dev.c
@@ -216,6 +216,7 @@ static void rtl8187_tx_cb(struct urb *ur
 		 * here, thus queue the skb and finish on a work queue. */
 		skb_queue_tail(&priv->b_tx_status.queue, skb);
 		queue_delayed_work(hw->workqueue, &priv->work, 0);
+		ieee80211_stop_queues(hw);
 	}
 }
 
@@ -898,6 +899,7 @@ static void rtl8187_work(struct work_str
 	}
 	retry = tmp;
 	mutex_unlock(&priv->conf_mutex);
+	ieee80211_wake_queues(dev);
 }
 
 static int rtl8187_start(struct ieee80211_hw *dev)

Re: rtl8187 - nice work!

[P.G. Richardson]

> P.G. Richardson wrote:
>> phy0: Selected rate control algorithm 'pid'
>>
>> which is a first as it has always been 'mistral'. The only differenc=
e
>> with
>> compiling last night was I was not running the kernel I compiled
>> against.
>> I used 'make KLIB=3D/usr/src/{my new kernel} KLIB_BUILD=3D/usr/src/{=
my new
>> kernel}' instead.
>>
>> Just recompiled while running {my new kernel} and mistral correctly
>> selected as rate control algorithm. Once installed, it does correctl=
y
>> auto
>> select rate as expected.
>>
>> iperf shows Mb performance has dropped though but gets improved by a=
gain
>> fixing the rate at 11Mb.
>>
>> So cheers for the info.
>
> Minstrel is definitely a better choice than pid.
>
> There is one thing that might make your interface work better. For
> performance
> reasons, I chose not to pause the mac80211 queues at one point in the
> driver.
> For my device, it did not make a difference, but it might for you. Pl=
ease
> try
> the attached patch.
>

Hi Larry,

Applied this patch, this morning. Run with it rest of the day yet seems=
 to
make little or no difference so far.

Cheers

PGR

--=20
Laws are partly formed for the sake of good men, in order to instruct t=
hem
how they may live on friendly terms with one another, and partly for th=
e
sake of those who refuse to be instructed, whose spirit cannot be subdu=
ed,
or softened, or hindered from plunging into evil. [The Laws, Plato]

You fiend! Never have I encountered such corrupt and foul-minded
perversity=85 Have you ever considered a career in the church?
Bishop of Bath and Wells (Blackadder II)

--
To unsubscribe from this list: send the line "unsubscribe linux-wireles=
s" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: rtl8187 - nice work!

[Larry Finger]

P.G. Richardson wrote:
> 
> Hi Larry,
> 
> Applied this patch, this morning. Run with it rest of the day yet seems to
> make little or no difference so far.

That is what I expected, but a test was needed. Thanks for doing that. If I come
up with something else, I'll let you know.

Larry