Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 8 Jul 2002 01:22:52 -0700 (PDT)
From:      John Kozubik <john@kozubik.com>
To:        "M. Warner Losh" <imp@village.org>
Cc:        freebsd-hackers@FreeBSD.ORG
Subject:   Re: multi-link 802.11b through netgraph yields poor performance.
Message-ID:  <Pine.BSF.4.21.0207071629240.40375-100000@www>
In-Reply-To: <20020707.121836.61267901.imp@village.org>

next in thread | previous in thread | raw e-mail | index | archive | help

> Also, before blaming netgraph, which may well be to blame, could it be
> that you have interference from some other source that's making things
> bad?  The exactly every other packet being dropped does seem to be a
> big clue.

I have ruled out interference as a contributor to these results.

My next step was to remove netgraph from the equation - I simply set up
the two channel/SSID pairs across the two laptops, and assigned one pair
addresses 10.10.10.10 and 10.10.10.20 with a /24 netmask and the other
pair 192.168.0.129 and 192.168.0.130 with a /25 netmask.  Each pair could
ping between themselves with 0% packet loss.

Therefore, it seems that there are no real technical difficulties in
running two pair 802.11b networks between two laptops, using two adaptors
in each machine.

So, I switched the master/slave order for the cards participating in the
multi-link on each laptop.  That is, I made wi1 the master and wi0 the
slave on one side, and an1 the master and an0 the slave on the
other.  This was illuminating, as this configuration did not allow any
functionality - not even the 50% packet loss.  Instead, on the system with
two Lucent cards, I got the familiar watchdog timeout errors:

/kernel: wi1: wi_write_data device timeout
/kernel: wi1: xmit failed
/kernel: wi1: watchdog timeout
...
/kernel: wi1: failed to allocate 1594 bytes on NIC
/kernel: wi1: tx buffer allocation failed

Basically the same old watchdog errors that I see quite frequently on
Lucent cards.

So, the point made earlier in this thread that the behavior I produced
looked like the behavior of a ng_one2many setup with 50% of the link not
working seems to be correct.  However, because the first test I mentioned
above (two wireless nets operating simultaneously across two systems
without netgraph) succeeded, I am not simply dealing with a bad card or a
down link.

Instead, I think this may be a firmware issue.  Specifically, different
revisions of the Lucent firmware behave differently as regards promiscuous
mode, etc.  Because the card works in all other tests, and because the
other Lucent card can successfully master the netgraph ng_one2many, I am
tentatively concluding that some Lucent firmwares will not work in a
ng_one2many setting.

OR maybe it is just general `wi` flakiness - this suspect card produces
wi_seek timeouts when given an IBSS to create (`wicontrol -q`) on Toshiba
Libretto laptops, but not on other laptops (even other Toshibas).  

Therefore, it is possible that there is no ng_one2many / firmware problem,
but rather, since Lucent cards (seem to) malfunction based on somewhat
random and arbitrary conditions, perhaps this is just one more of those
random and arbitrary conditions.


Comments ?

In the meantime I will retry this experiment with Cisco cards exclusively.

-----
John Kozubik - john@kozubik.com - http://www.kozubik.com


To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-hackers" in the body of the message




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.BSF.4.21.0207071629240.40375-100000>