On Thu, Oct 09, 2003 at 10:24:59PM -0700, Roger Fischer wrote:
> At 9:59 PM -0500 10/9/03, Richard Rauch wrote:
> >Re. http://mail-index.netbsd.org/netbsd-help/2003/10/08/0019.html
> >
> >I have a PCMCIA card that used to do something similar, except I think that
> >mine would (literally) alternate between ne1 and ne2.  on a cold boot, I 
> >think
> >that it came up as ne2.  Warm boot to ne1.  Warmboot again to ne2, ...
> >Power off, power back on, cold boot to ne2...
> >
> >Because this is weird, and the driver worked (sort of) in 1.5 I
> >just hacked the kernel to ignore the last test's results.  Now it
> >works.  (Though I get lots of ring buffer overruns---I'm not sure
> >if I had that before---it is infinitely better than how it was on
> >a stock 1.6 system: Dead.)
> 
> Well, I don't have the problem of it switching.  ne2 is the one that works.

Yes, not exactly the same, but enough echoes that I thought I'd share my
experiences.  You might be able to wire it down to just one interface
with a kernel config, or the kernel may be doing a double-probe.
(I recall that in 1.6, the error messages were confusing as it first
thought that it recognized my card, then said it didn't; I don't know
if it would or could have gone on to probe as something else.)


> But that ring buffer overrun has piqued my interest.  I've had a 
> problem for a while where I'll be transferring and suddenly I'll hear 
> the disk drive go bzzzt and the connection will drop.  For ftp or 
> similar, transfer rate will go to zero for around 20 to 30 seconds, 
> then it'll resume until it happens again.  If it happens while 
> loading a web page, hitting stop then reload usually gets it to 
> resume.  Very annoying, but I haven't been able to track it down. 
> Nothing appears in /var/log/messages and netstat shows nothing 
> unusual.  What, if you don't mind my asking, were the symptoms of 
> your ring buffer overruns?

Symptoms?  I get messages about it in xconsole and also if I'm logged
in as root.

Performance is presumably being shot down, but I don't know by how
much in most cases.  The only solid indicator of performance penalty
is when I was running an X application over my LAN with the display
on the laptop.  The program was a compute+graphics intensive program
that *should* in that situation have saturated some resource.

The ethernet link was very un-active.

The laptop (a 233MHz plain Pentium) was very idle.

The remote system (an 800MHz Athlon) was very idle.

I would have expected the laptop to be maxed out trying to
deal with ssh and OpenGL + X.  Instead, it was sleeping a lot.

I think that frame-rate would start out upwards of 10fps,
but then would drop off to about 1 to 3 fps after the errors.
I suspect that the problem is at least in part that PCMCIA may not
be up to 100Mbits/sec (I don't know what its capabilities are, but
have heard that assertion before).  I might do better if I forced
10Mbits/sec via ifconfig.


The drive noise in your case suggests to me that your system isn't able
to keep up with the card while doing disk activity.  My system, I think,
is just plain not able to keep up.  (^&

You might try throttling the data rate back, if you use that card to
talk to a "mere" DSL modem or cablemodem.  (Here, I have two networks:
One is a hub that my DSL plugs into, the other is a switch that only
my own machines have access to.  The latter goes a lot faster than what
the modem could ever keep up with...  Although I haven't had a lot of
problems with RealTek, personally, I've heard enough bad things about
them that I keep the RealTek cards on the hub and use tlp and fxp
interfaces for the switch.)

(My laptop, alas, has but the one interface.)


-- 
  "I probably don't know what I'm talking about."  http://www.olib.org/~rkr/