Even more mysterious Tx excessive retries - with solution?

[Jim Carter]

On Sat, 21 Sep 2002, Roger Larsson wrote:
> > [jimc paraphrase: there was a mailing list discussion about slow data
> > rate from Linksys (Intersil) card to various partners under Linux,
> > using the Orinoco driver set...]

> I have run into something similar.
> Did you find a solution on this? I have searched the achieves but found none.
>
> Lets describe my situation:
> [jimc summary:]  Upgraded driver versions and the data rate went through
> the floor.

No, I never got it to improve.  My "solution" was to buy an Agere AP-200
access point.  My son got a Linksys WAP-11.  Both of them work fine with
the Dell Truemobile 1150 (Lucent-Agere firmware).  As well as with the
offending Linksys WPC-11 card, and a Siemens Speedwhatever 1024 PCI
wireless card.

> But when I did
>   iwconfig eth0 rate 2M fixed
> on both computers, it started to behave again.

I saw something similar: about an 8x speed improvement if I forced it to
the lowest speed. In other words, performance went from abysmal to merely
bad: about 21k bit/sec improved to 160k bit/sec. My explanation is that if
it starts at 11 Mb/s it has to retry several times, then downgrade to the
next lower speed... So when starting at the lower speed it has less retries
to time out on.

> My understanding of the problem is:
>   The new driver version detects the "Tx excessive retries" and
>   tries to do something about it - but it only makes it worse...

Unfortunately the driver doesn't adapt to Tx retries; it doesn't even look
at them, just reports them as part of the statistics.

I looked at the wlan-ng sources, comparing them with orinoco.c, but found
nothing useful that they do and orinoco doesn't. I wish someone had the
right NDA terms and conditions that they could look up the relevant area in
the documentation.  Like the wlan-ng people.

Have you tried wlan-ng? There are so many options that it's kind of hard to
get it running properly. I tried it once and it did not improve
performance, but maybe I set it up wrong. If you tried it and the problem
went away, that would be a real breakthrough on the Orinoco side, because
then we could legally steal whatever bug fix they had put in their code.

James F. Carter          Voice 310 825 2897    FAX 310 206 6673
UCLA-Mathnet;  6115 MSA; 405 Hilgard Ave.; Los Angeles, CA, USA  90095-1555
Email: jimc at math.ucla.edu    http://www.math.ucla.edu/~jimc (q.v. for PGP key)