WDS speed issues revisited

[Dave]

> If I understood correctly, you were determining TX rate by using some
> higher level throughput test, not by looking at which rate was used for
> each packet. There are some cases, in which Prism2 cards seem to stop
> sending ACK frames. This would show up as dropped throughput even though
> frames are actually still being sent at 11 Mbps; they are just being
> retransmitted many times.. This could explain some of the issues you
> have seen.
>
> I did some more testing on Repeater mode and noticed few bugs in Host AP
> driver. RX statistics were not being updated in this mode and actually,
> TX errors were not processed at all; so the rate control should not have
> dropped rate at all.. I fixed both bugs. I would appreciate it if you
> could re-test using the latest CVS version and verify the real TX rate
> by looking into both TX and RX statistics in
> /proc/net/hostap/wlan0/<hwaddr>.

I do not have STA statistics in /proc/net/hostap/wlan0/<hwaddr>. Both sides
of the link are in Repeater mode so I would assume that is why I do not have
STA statistics...Should I have those files/statistics even for WDS links
when they do not associate to each other?

>
> On Wed, Apr 09, 2003 at 01:54:33AM -0700, Dave wrote:
>
> > 'iwconfig wlan0 rate 11M fixed'
> >
> > does not fix the TX rate. In fact once I issue that command for one
> > side of
> > a WDS link which is currently at 11Mbps, the rate drops to 2Mbps.
>
> I did not see such behavior. Setting rate to fixed 11M seemed to work
> fine. However, if you set rate after it has first dropped, it was not
> raised when changing configuration. I fixed this so that rate control
> data is updated automatically when changing operational rates. With
> current CVS version, I was able to first let the rate control code to
> drop the rate to 1M and then manually force only 11M to be used. This
> forced the rate to stay at 11M even though I was generating TX error for
> every 10th frame.
>
> > I think I understand Jean's comment that the RX is controlled by the
> > opposite ends TX...but if the opposite ends TX is fixed to 11M then in
> > theory wouldn't this ends RX receive at 11M....minus the overhead,
> > retrys,
> > errors, birds, planes..
>
> Yes. My current theory is that you might have seen the "card stops
> ACKing" problem; i.e., retransmissions are eating most of the
> throughput.

Understood

> > I am at least confident that the displayed speed via iwconfig wlan0 on
> > each
> > side of the link in WDS/Repeater mode is accurate as I perform data
> > transfers each way and it is right at the displayed speed.
>
> STA files in /proc/net/hostap/wlan0 should provide even better
> statistics; both for TX and RX. In the end of those files, there are
> counters showing number of TX/RX frames at each rate.
>
> > mishap...I now have radios seperated by 2.5 miles and initially get an
> > 11Mbps link...which quickly turns into a 1Mbps link as soon as data
> > starts
> > moving...but that is probably related to fade margin or something
> > because I
> > do not have perfect LOS...but still it would be great to have it fixed
> > at
> > 11Mbps and deal with fade and errors and loss as a seperate issue.
>
> Current algorithm used for TX rate controlling is not very good and
> getting stuck into 1 Mbps is probably quite common in many environments.
> Anyway, at least I was able to manually configure ('iwconfig wlan0 rate
> 11M fixed') the rate to be fixed at 11 Mbps even on very poor
> (simulated) link. Please let me know if current CVS version does not
> work for you.

I will give it a try shortly...thanks for the ideas!


-Dave


> --
> Jouni Malinen                                            PGP id EFC895FA
> _______________________________________________
> HostAP mailing list
> HostAP at shmoo.com
> http://lists.shmoo.com/mailman/listinfo/hostap
>