-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Ignacy Gawedzki a écrit :
quoted text
> Hi everyone,
>
> During a few tests with madwifi-hal-testing and madwifi-free branches, I've
> noticed strange packet loss at the sender.
>
> I'm using iperf to send small (100 bytes long) UDP packets to a unicast
> destination (in ad-hoc mode, without any protection) and ask
purposefully for
quoted text
> a sending rate of 54 Mbps, just to see how fast the card is able to
transmit
quoted text
> in that very environment.
>
> Sometimes these packets get lost before they even have a chance to get
sent by
quoted text
> the hardware.  Iperf happily shows it is sending at the requested 54 Mbps,
> which is clearly impossible given the mode, the hardware, and given what
> iperf at the server side is reporting (more or less 5 Mbps).  There are
many
quoted text
> gaps in the iperf sequence numbers, but not in the MAC sequence numbers, as
> reported by the monitor trace from a third machine.  In other words,
there may
quoted text
> very well be a big gap of iperf sequence numbers in two packets in two
> successive MAC frames (according to the MAC sequence numbers).
>
> The strangest thing is that if I enable tx debugging (athdebug +xmit), the
> problem disappears at least in a great part and the sender iperf correctly
> reports around 5 Mbps, most of the time.
>
> I'm actually becoming familiar with the code, so I looked at
ath_hardstart and
quoted text
> ath_tx_start, but can't really figure out what the ATH_DEBUG_XMIT setting
> could really change regarding the above problem.
>
> Any idea?
>
Hi,

I already noticed this kind of behavior when using iperf over
Ethernet. Let's say you are trying to send a UDP stream at 200 Mbit/s
over a 100 Mbit/s ethernet interface. If no cable are connected, iperf
will happily reports 200 Mbit/s being sent, which means that the
kernel just drops everything! If an Ethernet cable is connected,
somehow, the kernel (I'm not sure it's the driver itself) is limiting
iperf to say 95 Mbit/s. I think the latter is strange in fact, not the
former. So if there is many gaps in the iperf sequence numbers, it
means that either the linux kernel or the madwifi driver is dropping
packets. For me, it's pretty clear that it's the kernel dropping
packet as expected.

Regards,
Benoit

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iEYEARECAAYFAkpPBdAACgkQOR6EySwP7oKuFQCePtTCKCxnpLes2JGeAnoSGCpu
YKIAnA88bE/vSCIAGiXHqi9QhHML1q0m
=2Yv0
-----END PGP SIGNATURE-----


------------------------------------------------------------------------------
_______________________________________________
Madwifi-devel mailing list
Madwifi-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/madwifi-devel