[Madwifi-users] Strange things and problems with WMM

[Aleksej <ale...@go...>]

Hi,

I'm testing WMM operation and have some problems (I use MadWifi
v0.9.4, Gigabyte 802.11abg GNWI01HT cards with Atheros AR5413
chipsets, Ubuntu 7-10). Actually, out of these three problems listed
below, only the first one is a real problem, the other two are just
remarks.

When each station has only one active access category, prioritization
works fine. For example, one station transmits traffic only from AC_BE
and the other one performs transmission only from AC_VI. The traffic
in AC_VI has higher throughput than in AC_BE. That's fine.

Problem 1.

When I perform tests with only one station, the strange behaviour is
observed. It seems that within one station the traffic in AC_BE takes
always precedence over all other access categorises!  For example, if
a station has traffic in two access categories - AC_VI and AC_VO,
prioritization works as expected and traffic in AC_VO has higher
throughput than in AC_VI. If I redirect the traffic from AC_VI to
AC_BE, then traffic in AC_BE results suddenly in higher throughput
than in AC_VO. Moreover, the traffic in AC_VO can be completely
stopped! It seems that the algorithm of internal virtual collision
resolution doesn't work correctly and whenever there is some traffic
in AC_BE it is served first while the other access categories are
served only when AC_BE is empty. Is it some configuration issue? I
checked all WMM parameters, everything is configured by default
according to WMM spec... Has anyone had such a problem?

Problem 2.

Test for maximum achievable throughput (one station has traffic only
in one access category.) The thing is that the throughput of AC_VI is
surprisingly slightly higher than of AC_VO. The difference is not big
at all, but just interesting why… However, when AC_VI and AC_VO
compete together than AC_VO has higher throughput.

Problem 3.

This one is about marking, as was described in
http://thread.gmane.org/gmane.linux.drivers.madwifi.user/6075


I would be happy for any feedback.
Thanks, Aleksej.