2011-03-18 05:18:27

by Hin-Tak Leung

[permalink] [raw]
Subject: rtl8187b throughput-related regression in 2.6.35.x fixed in 2.6.37.x

Hi Larry/Herton/John,

There appears to be some kind of throughput related regression in 2.6.35.x (11-85-ish fedora, current long-term maintainance kernel) which is fixed by putting compat-wireless 2.6.37.x (4) on. I just thought I should mention this, in case somebody files a bug about rtl8187(b) being too slow, that putting compat-wireless on might help.

My AP machine was out for almost a year so I have not been using my rtl8187b for a while. When I finally got that fixed, after a while I noticed the throughput between the two machines was low (60KB/s, or 0.5Mb/s, although iwconfig says rate=11Mb/s, etc). Putting compat-wireless on gets me back the throughput I expect from a year ago (400KB/s+ onwards, or a few Mb/s at least). So 2.6.31.x a year ago was okay, 2.6.37.x is okay, but there appear to be something not quite right with 2.6.35.x in the middle. There are too many changes, and it is probably not important since current-ish kernel head is okay.

Regards,
Hin-Tak





2011-03-27 04:01:26

by Hin-Tak Leung

[permalink] [raw]
Subject: Re: rtl8187b throughput-related regression in 2.6.35.x fixed in 2.6.37.x

Stanislaw Gruszka wrote:
> On Fri, Mar 18, 2011 at 05:18:25AM +0000, Hin-Tak Leung wrote:
>> Hi Larry/Herton/John,
>>
>> There appears to be some kind of throughput related regression in 2.6.35.x (11-85-ish fedora, current long-term maintainance kernel) which is fixed by putting compat-wireless 2.6.37.x (4) on. I just thought I should mention this, in case somebody files a bug about rtl8187(b) being too slow, that putting compat-wireless on might help.
>>
>> My AP machine was out for almost a year so I have not been using my rtl8187b for a while. When I finally got that fixed, after a while I noticed the throughput between the two machines was low (60KB/s, or 0.5Mb/s, although iwconfig says rate=11Mb/s, etc). Putting compat-wireless on gets me back the throughput I expect from a year ago (400KB/s+ onwards, or a few Mb/s at least). So 2.6.31.x a year ago was okay, 2.6.37.x is okay, but there appear to be something not quite right with 2.6.35.x in the middle. There are too many changes, and it is probably not important since current-ish kernel head is okay.
>
> Would be nice to have this fixed in stable.
>
> We have already one AP mode bug opened:
> http://bugzilla.redhat.com/show_bug.cgi?id=671514
>
> It was recognized that this patch is needed in .35.11 (.35.9 is ok):
> http://git.kernel.org/linus/09a02fdb919876c01e8f05960750a418b3f7fa48
>
> Not sure if this is the same issue, but perhaps you could try above
> patch. If it does not help, could you check if problem is
> 2.6.35.x -> 2.6.35.11 regression?
>
> Stanislaw
>

Just checked again, downloading and using a bunch of compat tar balls. It
appeared that the throughput problem I reported above was fixed between 2.6.37.4
and 2.6.38-rc7. In any case, it does not appear to be the ones you mentioned.

I might spend a bit more time eventually to see what fixes it. The one thing
interesting
about those two versions were Herton's house-keeping

rtl8187: restore anaparam registers after reset with 8187B
rtl8187: fix wrong register initialization in 8187B

and Thadeu Lima de Souza Cascardo
rtl8187b: do not do per packet TX AGC

And the Signal level did seem to change from about -40 dBm to -30

But these are a bit curious how I did get good throughput with 2.6.31.x-ish .

Hin-Tak


2011-03-18 08:54:43

by Stanislaw Gruszka

[permalink] [raw]
Subject: Re: rtl8187b throughput-related regression in 2.6.35.x fixed in 2.6.37.x

On Fri, Mar 18, 2011 at 05:18:25AM +0000, Hin-Tak Leung wrote:
> Hi Larry/Herton/John,
>
> There appears to be some kind of throughput related regression in 2.6.35.x (11-85-ish fedora, current long-term maintainance kernel) which is fixed by putting compat-wireless 2.6.37.x (4) on. I just thought I should mention this, in case somebody files a bug about rtl8187(b) being too slow, that putting compat-wireless on might help.
>
> My AP machine was out for almost a year so I have not been using my rtl8187b for a while. When I finally got that fixed, after a while I noticed the throughput between the two machines was low (60KB/s, or 0.5Mb/s, although iwconfig says rate=11Mb/s, etc). Putting compat-wireless on gets me back the throughput I expect from a year ago (400KB/s+ onwards, or a few Mb/s at least). So 2.6.31.x a year ago was okay, 2.6.37.x is okay, but there appear to be something not quite right with 2.6.35.x in the middle. There are too many changes, and it is probably not important since current-ish kernel head is okay.

Would be nice to have this fixed in stable.

We have already one AP mode bug opened:
http://bugzilla.redhat.com/show_bug.cgi?id=671514

It was recognized that this patch is needed in .35.11 (.35.9 is ok):
http://git.kernel.org/linus/09a02fdb919876c01e8f05960750a418b3f7fa48

Not sure if this is the same issue, but perhaps you could try above
patch. If it does not help, could you check if problem is
2.6.35.x -> 2.6.35.11 regression?

Stanislaw

2011-04-07 22:04:52

by Hin-Tak Leung

[permalink] [raw]
Subject: Re: rtl8187b throughput-related regression in 2.6.35.x fixed in 2.6.37.x

On Sun, Mar 27, 2011 at 5:01 AM, Hin-Tak Leung
<[email protected]> wrote:
> Stanislaw Gruszka wrote:
>>
>> On Fri, Mar 18, 2011 at 05:18:25AM +0000, Hin-Tak Leung wrote:
>>>
>>> Hi Larry/Herton/John,
>>>
>>> There appears to be some kind of throughput related regression in
>>> 2.6.35.x (11-85-ish fedora, current long-term maintainance kernel) which is
>>> fixed by putting compat-wireless 2.6.37.x (4) on. I just thought I should
>>> mention this, in case somebody files a bug about rtl8187(b) being too slow,
>>> that putting compat-wireless on might help.
>>>
>>> My AP machine was out for almost a year so I have not been using my
>>> rtl8187b for a while. When I finally got that fixed, after a while I noticed
>>> the throughput between the two machines was low (60KB/s, or 0.5Mb/s,
>>> although iwconfig says rate=11Mb/s, etc). Putting compat-wireless on gets me
>>> back the throughput I expect from a year ago (400KB/s+ onwards, or a few
>>> Mb/s at least). So 2.6.31.x a year ago was okay, 2.6.37.x is okay, but there
>>> appear to be something not quite right with 2.6.35.x in the middle. There
>>> are too many changes, and it is probably not important since current-ish
>>> kernel head is okay.
>>
>> Would be nice to have this fixed in stable.
>>
>> We have already one AP mode bug opened:
>> http://bugzilla.redhat.com/show_bug.cgi?id=671514
>>
>> It was recognized that this patch is needed in .35.11 (.35.9 is ok):
>> http://git.kernel.org/linus/09a02fdb919876c01e8f05960750a418b3f7fa48
>>
>> Not sure if this is the same issue, but perhaps you could try above
>> patch. If it does not help, could you check if problem is
>> 2.6.35.x -> 2.6.35.11 regression?
>>
>> Stanislaw
>>
>
> Just checked again, downloading and using a bunch of compat tar balls. It
> appeared that the throughput problem I reported above was fixed between
> 2.6.37.4 and 2.6.38-rc7. In any case, it does not appear to be the ones you
> mentioned.
>
> I might spend a bit more time eventually to see what fixes it. The one thing
> interesting
> about those two versions were Herton's house-keeping
>
> rtl8187: restore anaparam registers after reset with 8187B
> rtl8187: fix wrong register initialization in 8187B
>
> and Thadeu Lima de Souza Cascardo
> rtl8187b: do not do per packet TX AGC
>
> And the Signal level did seem to change from about -40 dBm to -30
>
> But these are a bit curious how I did get good throughput with 2.6.31.x-ish

Down to between:
compat-wireless-2.6.37-4-sn.tar.bz2
compat-wireless-2.6.38-rc3-3.tar.bz2

compat-wireless-2.6.38-rc3-2.tar.bz2 doesn't build.