[ltp] T61: ipw3945 speed limited to 256 kbit/s?

Marius Gedminas linux-thinkpad@linux-thinkpad.org
Thu, 9 Aug 2007 23:52:00 +0300


--VS++wcV0S1rZb1Fb
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

On Thu, Aug 09, 2007 at 09:14:59PM +0100, Richard Neill wrote:
> Marius Gedminas wrote:
>> The worst problem at the moment is wireless speed: it seems to be
>> limited to 32 kilobytes per second.

I brought the laptop home and here it happily downloads stuff at 110 kB/s.
Interesting.  Or it could be the power cycle that I had to do after a
kernel panic from a failed suspend.  Or it could be that I disabled
Network Manager and chose manual configuration (essid any, dhcp) in
/etc/network/interfaces.

32 kB/s is a very interesting number -- 256 Kbit/s.  No matter what I
did at work, GNOME netspeed applet showed numbers around 32 kB/s (+/-
around 0.7 kB/s).

> I haven't seen this specific thing before, but I've seen the same sort of=
=20
> general "it works 3 orders of magnitude too slow" problem in other cases.=
=20

With ipw3945?

> Some troubleshooting suggestions:
>
> 1)Eliminate hardware errors as the cause - try it either in Windows, or=
=20
> with Knoppix. Then, hopefully you know the hardware works. It should also=
=20
> eliminate signal-strength problems as a root cause.

Given the current wifi speed, it's not hardware.

> 2)What's the CPU load doing? Something like the KDE "System Monitor"=20
> taskbar applet is very useful here - it shows usage of RAM/CPU/SWAP, and=
=20
> now it also shows IOWait.

I can't use a computer without a CPU (+IOWait) and RAM usage monitor on
my GNOME panel.

> 3)Packet errors? I had a recent case where we had what turned out to be a=
=20
> faulty network switch.

Other machines in the same wireless LAN did not have this problem.  I
didn't look at error counters in ifconfig.  I checked for strange
messages in dmesg but didn't find any.

> 4) DMA? Is your hard drive causing I/O trouble due to having DMA off? (th=
is=20
> is unlikely, but it can happen).

No, the bottleneck was definitely at network level.  Copying from a USB
disk enclosure ran at 10 Mbyte/s.

> 5) Interrupt storms?

Wouldn't that show up in the CPU applet as system time?

> Look at /proc/interrupts. Except for the system timer,=20
> you shouldn't be getting huge numbers every second. [I had a motherboard=
=20
> recently which needed the graphics aperture to be set to a specific=20
> (non-default) value, or it would cause an interrupt storm.
> For example:    for ((i=3D0; i< 10000; i++)); do echo $i > /dev/null; done
> should take about 1 second. If it's very much slower, something else is=
=20
> wrong!

vmstat shows me about 250 interrupts and about 500 context switches per
second, but then this is now and now when I saw the problem.

> 6) RAM. Do you have > 2GB?

No, just 1 GB (and some part of it is claimed by the Intel video).

> Again, a stupid thing I recently saw is a 64-bit=20
> motherboard, with capacity for 8GB of RAM (4 slots) and a core-2 quad CPU=
=2E=20
> If more than 6 GB is present, the system slows to a crawl (no idea why;=
=20
> again, the for loop above can take nearly a minute to execute).
> Also, you could try memtest.

Thanks for all the excellent pointers for diagnosing hardware problems.

Marius Gedminas
--=20
Truth does not demand belief. Scientists do not join hands every Sunday,
singing, 'Yes, gravity is real! I will have faith! I will be strong! I beli=
eve
in my heart that what goes up, up, up must come down, down, down. Amen!' If
they did, we would think they were pretty insecure about it. - Dan Barker

--VS++wcV0S1rZb1Fb
Content-Type: application/pgp-signature; name="signature.asc"
Content-Description: Digital signature
Content-Disposition: inline

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)

iD8DBQFGu35wkVdEXeem148RApWvAJ932VVIVP64qWxuLdOfIaKx1zAZBACdFSn7
6jGC37Tzk3/bAGGRd3/uh8g=
=GhXK
-----END PGP SIGNATURE-----

--VS++wcV0S1rZb1Fb--