Subject: Re: Slow wi0 ftp transfers after kernel upgrade?
To: David Young <firstname.lastname@example.org>
From: Paul Ripke <email@example.com>
Date: 03/02/2003 18:37:00
On Sunday, Mar 2, 2003, at 15:39 Australia/Sydney, David Young wrote:
> On Sun, Mar 02, 2003 at 12:33:39PM +1100, Paul Ripke wrote:
>> I upgraded my kernel (for a variety of reasons) from 20030128 sources
>> up to 20030218, then
>> after discovering the problem, up to 20030301, in case it had been
> Between 1/28 and 2/18, it does not look to me like anything changed
> in wi(4).
Yah. I went looking and couldn't find anything relevant either.
>> BTW: I've also noticed that my wi0 appears to be asymmetric... I get
>> approx 180 KB/s in one
>> direction (NetBSD sending) and 500+ KB/s in the other.
> In Host AP mode, NetBSD does the WEP encryption for transmission.
> Perhaps that explains the asymmetry?
I'm not running WEP, so I don't think that's it. Just shouting
plaintext all over the place.
>> tcpdump shows 0.6 - 1.0 second pauses, which looks like a timeout to
>> me. 192.168.254.130 is
>> the laptop, 192.168.254.129 is the NetBSD box.
> I do not know what could explain the pauses. Does wi0 log anything
> the pauses? You might get to the bottom of this by running in debugging
> mode. Add IEEE80211_DEBUG and WI_DEBUG to your kernel config and use
> ddb to bump up wi_debug and ieee80211_debug.
I've set both, no logging whatsoever. I did get some guff out when
changing rates, so at least I know the debug opts are in there.
BTW, the same symptoms occur with EPSV4, passive mode and active. But
*only* when the NetBSD box runs as the server, and only with an FTP
"put". Sounds wrong to me. I would've expected FTP "get"s initiated
by the NetBSD box to be affected if this were purely interface/tcp
stack related. I'm going to start regressing back through recent
changes to see if I can't find what's done it.
>> 2 Mbit sounds about right. What happened to DS11?
> I don't know why it has adapted to 2Mb/s.
The state I posted was "media ds11 mediaopt hostap". Previously, it
was auto-select, where it appeared to be running DS2 in both directions.
At least now it seems to be doing DS11 Rx. BTW, rate changes don't seem
to affect the above problem.
> BTW, in Host AP mode, the Prism firmware accepts per-packet TX rates,
> wi(4) does not take advantage. It would be nice if somebody would
> a rate algorithm for the IEEE802.11 framework, especially since this
> be necessary for the "dumber" 802.11 cards coming to market. Lucent has
> a technical report on rate adaptation which I can scare up for anybody
> who is interested.
I think that's a little beyond me just now. I seem to be doing much
just finding problems... :(