collisions 200% of Tx on 10BaseT (21140, .90q, Samsung SC1200-TX)

[Metod Kozelj]

Hello,

On Sun, 4 Apr 1999, Monty wrote:

> My net here is 10baseT; I bought the 100Mbps capable Tulips as they're
> reputed to be cheap and hardy.  However, the Samsung card displays one
> extremely annoying behavior; when doing heavy receiving (for example,
> FTP downloads from another machine on the same subnet), it collides on
> just about every transmit (the TCP acks).  The "collision" light on
> the hub is on solid.  ffconfig confirms the diagnosis; it collides
> more often than it successfully transmits (all this documented below).
> The reverse does not seem to be a problem; heavy uploads (mostly large
> transmits, very few receives) show no collision behavior. I've used
> .90q and .90p.  Note that the de4x5 driver shows the same behavior.

Just a couple of days ago I ran into the same problem. I was thinking
about which forum would be appropriate to discuss the topic when I
received this message.

Now, my setup is a bit different, here's my story.

I have an Intel box (P120 if that matters) with a 3c509B NIC and thin coax
network. The system was running excelently with Slackware 3.3 and various
2.0 kernels.
A week ago I did a completely new install, this time RedHat 5.2 with
updates for kernel 2.2. Now it's running 2.2.4.
There are several boxes connected to the same coax cable, which in turn is
connected to some 10/100 switch with several othe branches connected.
The system can't deal with only two speciffic boxes on the same coax, it
deals fine with other boxes. When I try to 'ping' the box, I don't get any
reply. The collision light on the hub I have also connected to the coax,
goes on for each and every packed sent to that host. I can't get even an
ARP reply from it (the /proc/net/arp entry has all zeroes for HW address).
If, by a chance, I get the ARP reply, I only get about 50% pings echoed.

The two boxes that have problems wit hte box in question, are linux boxes.
One is AMD-K6/linux 2.2.4 box ant the other is Alpha/linux 2.0.36 box.
The first one has another 3c509B NIC and the second one has a 21140 tulip
NIC.

I can ping the box in question from any other machine, be it fro mthe same
coax branc or from any other branches via the switch. Other boxes on
the same coax are mainly Win boxes with slower TCP stack, and switch
seemingly adds enough latency that the faster boxes on other branches
don't have problems.

It seems important to me, that the same equipment (including the 3c509B
NIC) worked fine under 2.0.x kernels.
So it seems to me, that it's the tcpip driver in the 2.2 kernels that is 
a bit too fast or something.

Regards,
  Metod

Metod Kozelj

mailto:Metod.Kozelj@rzs-hm.si            /\  Ne posiljajte mi smeti ker grizem!
http://www.rzs-hm.si/                   /  \  Don't spam me for I bite!
_______________________________________/    \__________________________________

---- perl -e 'print $i=pack(c5,(41*2),sqrt(7056),(unpack(c,H)-2),oct(115),10);'