[vortex] problems w/3c59x on 3CCFE575BT

[dean gaudet]

On Mon, 23 Dec 2002, dean gaudet wrote:

> On Mon, 23 Dec 2002, Malcolm Scott wrote:
>
> > Just an obvious question - are you sure the problem's on your machine,
> > and not on your webserver ("neemlark")?
>
> i'm pretty certain it's not neemlark... here's a full explanation of what
> i've tried to isolate the problem down to the 3c59x driver:
>
> 	mewlark (router)+- eth0  <- 192.168.1/24 -> neemlark
> 	                \- wlan0 <- 192.168.2/24 -> doomlark

> i should have a newer eepro100 kicking around somewhere, i can give it
> a try.

i couldn't find a newer eepro100 pci card kicking around my apartment...
but doomlark has a built-in i8255x which i know is much more recent
(probably an i82559).  so i moved doomlark to the wired 192.168.1
segment...

doomlark is a win2k box and the best i can whip up for sending bandwidth
out of doomlark is the following (in cygwin):

Administrator@DOOMLARK /c/temp
$ time sh -c 'dd if=/dev/zero bs=1024 count=8192 | ssh -l dean neemlark "cat >/dev/null"'
8192+0 records in
8192+0 records out

real    0m2.824s
user    0m0.791s
sys     0m0.670s

Administrator@DOOMLARK /c/temp
$ time sh -c 'dd if=/dev/zero bs=1024 count=1024 | ssh -l dean mewlark "cat >/dev/null"'
1024+0 records in
1024+0 records out

real    1m22.771s
user    0m0.300s
sys     0m0.140s

note that's 8MB transferred to neemlark, but only 1MB transferred to
mewlark... i got bored waiting for 8MB to transfer to mewlark.

(i'm using ssh-agent for authentication so that there's no human
passphrase delays in this.  and i actually used the numerical addresses
for mewlark and neemlark in the above commands because i wanted to be
sure it was going to the right interfaces.)

i've also checked a bunch of other things -- including disabling mewlark's
eth1 which is a USB ethernet dongle that talks to my dsl modem, and
has NAT and traffic shaping (wondershaper).  i've disabled the traffic
shaping entirely -- no change.

i've also tcpdump'd the 192.168.1 segment to look for any suspicious
traffic (i.e. ICMP routing foo or ARP noise or anything...) all i saw
was the usual TCP traffic.

and i just connected mewlark and doomlark's ethernets with a crossover
so that i could eliminate even the switch from this configuartion
and the doomlark->mewlark transfer rate is similarly slow as above.
(or worse -- i got tired of waiting and ^C'd it, but tcpdump indicated
some pitiful number of bytes were making it through.)

i'm stumped!

anyhow, next time i'm at a computer store i'll pick up a new cardbus
ethernet and i'm willing to mail this bizarre one to someone (donald?) if
you want to investigate it further.

-dean