Very Computer

by **Noe** » Sun, 22 Dec 2002 20:35:25

Hi,

I am building a firewall with 3 nics (fxp0 - fxp2) using 3.1-stable.
fxp0 is the to the internet. fxp1 is to the private network and fxp2
is to the DMZ.

The problem I am trying to track down is too many ethernet collisions
on the private and DMZ subnets. I see these collision occur when I
transfer files from the firewall to the private network. These occur
even though I am still in test phase and have only a laptop and the
firewall on the private network! In my opinion having only two boxes
on this network should mean that collisions should be very rare.

Anyone out there run into similar problem? Are there any utilities I
can run that will report on collisions?

FYI: I am using Intel Pro/100 S Desktop adapters on the firewall. The
laptop is using a Xircom Cardbus Ethernet100 adapter. The hub is a
3com 3c16700 (10Mbps only).

Noel

by **erik** » Sun, 22 Dec 2002 21:23:57

> Hi,

> I am building a firewall with 3 nics (fxp0 - fxp2) using 3.1-stable.
> fxp0 is the to the internet. fxp1 is to the private network and fxp2
> is to the DMZ.

> The problem I am trying to track down is too many ethernet collisions
> on the private and DMZ subnets. I see these collision occur when I
> transfer files from the firewall to the private network. These occur
> even though I am still in test phase and have only a laptop and the
> firewall on the private network! In my opinion having only two boxes
> on this network should mean that collisions should be very rare.

> Anyone out there run into similar problem? Are there any utilities I
> can run that will report on collisions?

> FYI: I am using Intel Pro/100 S Desktop adapters on the firewall. The
> laptop is using a Xircom Cardbus Ethernet100 adapter. The hub is a
> 3com 3c16700 (10Mbps only).

Collisions are a perfectly normal thing on a non-switched ethernet. Even
with only two devices connected you will see collisions since both devices
will try to talk at the same time. Ethernet is _designed_ this way. There
is nothing to worry about.

You describe that you see too many collisions, as well as that these are a
problem. Please state the number instead of 'too many', and please state
the problem they cause. 10% of all traffic being a collision is still
perfectly normal.

If you do not like collisions, use a switch instead of a hub.

EJ
--
Remove the obvious part (including the dot) for my email address

by **John Smit** » Mon, 23 Dec 2002 09:59:46

or if you only have one PC (your laptop) on your internal network, and plan
to keep it that way, use a X-over cable between the 2. no collisions, and
if both cards are 10/100, you should get 100mb

> > Hi,

> > I am building a firewall with 3 nics (fxp0 - fxp2) using 3.1-stable.
> > fxp0 is the to the internet. fxp1 is to the private network and fxp2
> > is to the DMZ.

> > The problem I am trying to track down is too many ethernet collisions
> > on the private and DMZ subnets. I see these collision occur when I
> > transfer files from the firewall to the private network. These occur
> > even though I am still in test phase and have only a laptop and the
> > firewall on the private network! In my opinion having only two boxes
> > on this network should mean that collisions should be very rare.

> > Anyone out there run into similar problem? Are there any utilities I
> > can run that will report on collisions?

> > FYI: I am using Intel Pro/100 S Desktop adapters on the firewall. The
> > laptop is using a Xircom Cardbus Ethernet100 adapter. The hub is a
> > 3com 3c16700 (10Mbps only).

> Collisions are a perfectly normal thing on a non-switched ethernet. Even
> with only two devices connected you will see collisions since both devices
> will try to talk at the same time. Ethernet is _designed_ this way. There
> is nothing to worry about.

> You describe that you see too many collisions, as well as that these are a
> problem. Please state the number instead of 'too many', and please state
> the problem they cause. 10% of all traffic being a collision is still
> perfectly normal.

> If you do not like collisions, use a switch instead of a hub.

> EJ
> --
> Remove the obvious part (including the dot) for my email address

by **Noe** » Mon, 23 Dec 2002 19:58:30

Hi,

I went in yesterday to determine what the collision counts were and
ran into problems of a different nature.

I checked the collison counts prior to running tests that involved
transferring a large file on the subnet. Those counts looked good
assuming counts less than 10% of the total number of packets sent are
okay. However the file transfer hung when I tried to ftp a 4MB file
between the firewall and a workstation. This only occurred when I had
a 10 Mbps hub in place. If I swap out the 10 Mbps with a cheap Asound
100 Mbps hub the transfer worked and ran quickly - collision counts
changed very little too.

In addition to swapping out the hub I have also tried new cables. Same
results.

I took the hub home and found performance to be good there so I
believe it can be ruled out as the source of the problem. Using 100
Mbps second hub is not an option. The adapters on the workstations are
old 10Mps types.

I have searched the groups and have found a couple of similar, but
different, problems wrt to the Intel Pro/100 S. I beginning to wonder
if these lan adapters are incompatible with OpenBSD when operating in
10BaseT mode. I plan on trying some old Linksys ethernet cards today.

Anyone out there have a similar experience?

Noel

by **Noe** » Tue, 24 Dec 2002 05:12:38

Hi,

I put in the old linksys cards this afternoon and the box worked just
fine.

Seems to me the drivers in 3.1-stable must be at fault.

I brought one of the nics home and plan on trying it on an 3.1 base
machine.

What do I do next? Replace all three cards? Report the problem? I
would prefer finding a way to fix the problem with a software update
if at all possible.

Noel

by **erik** » Tue, 24 Dec 2002 12:11:06

> Hi,

> I put in the old linksys cards this afternoon and the box worked just
> fine.

> Seems to me the drivers in 3.1-stable must be at fault.

> I brought one of the nics home and plan on trying it on an 3.1 base
> machine.

> What do I do next? Replace all three cards? Report the problem? I
> would prefer finding a way to fix the problem with a software update
> if at all possible.

You could try to run 3.2 since this is the current version. 3.1 is already
old. I haven't seen problems with intel (fxp) cards, not with 10 Mbit, nor
with 100 Mbit. For a loooooong time I had the webserver connected to a 10
Mbit hub ( the webserver is an Apple with integrated 10 Mbit network i/f).
The firewall was connected to the same hub with a fxp interface. No
problems whatsoever. Not with OpenBSD 2.9, 3.0, 3.1, 3.2.

Current uptime webserver: 04:09:30 up 245 days, 4:06, 1 user, load
average: 0.00, 0.00, 0.00

EJ
--
Remove the obvious part (including the dot) for my email address

by **Noe** » Tue, 24 Dec 2002 23:05:34

Hi,

Ran more tests at home this time. The box at home is a 3.1 base. I
installed the intel ethernet card in it. Used the hub at work and the
same laptop I used to test throughput with at work. I started ftp on
the laptop and transferred a file from the 3.1 box to the laptop. I
got the same results as at work - ftp reports a miserable 8KBps for a
10baseT link.

Next I ftp-d to the 3.1 box using a different home computer and
tranferred the same file. The results were much better. I get anywhere
from 700KBps to 900KBps.

Given all this you might assume the laptop is the problem. Maybe it is
but I notice the number of collisions is above the 10% mark mentioned
earlier.

When using the computer that gets 900 KBps I get the following report
from netstat:
Total Packets in: 177
Total Packets out: 218
Colls: 149

When using the laptop (8KBps throughput) nestat reports:
Total Packets in: 165
Total Packets out: 222
Colls: 30

Constrast this when I replace the 10baseT hub with a 100baseT hub and
get 0 collisions on either computers when I run the test.

My guess is the laptop begins to struggle when the collisions exceed a
certain threshold.

Do you have any ideas why the number of collisions are so high and
what I can do to reduce their number?

Noel

by **Kris Kielhofne** » Wed, 25 Dec 2002 00:45:55

> Hi,

> I am building a firewall with 3 nics (fxp0 - fxp2) using 3.1-stable.
> fxp0 is the to the internet. fxp1 is to the private network and fxp2
> is to the DMZ.

> The problem I am trying to track down is too many ethernet collisions
> on the private and DMZ subnets. I see these collision occur when I
> transfer files from the firewall to the private network. These occur
> even though I am still in test phase and have only a laptop and the
> firewall on the private network! In my opinion having only two boxes
> on this network should mean that collisions should be very rare.

> Anyone out there run into similar problem? Are there any utilities I
> can run that will report on collisions?

> FYI: I am using Intel Pro/100 S Desktop adapters on the firewall. The
> laptop is using a Xircom Cardbus Ethernet100 adapter. The hub is a
> 3com 3c16700 (10Mbps only).

> Noel

Try forcing the Intel Nics into 10mbps half duplex mode.

--
Kris Kielhofner

by **erik** » Wed, 25 Dec 2002 11:52:58

> Hi,

> Ran more tests at home this time. The box at home is a 3.1 base. I
> installed the intel ethernet card in it. Used the hub at work and the
> same laptop I used to test throughput with at work. I started ftp on
> the laptop and transferred a file from the 3.1 box to the laptop. I
> got the same results as at work - ftp reports a miserable 8KBps for a
> 10baseT link.

Which, to me, seems to be an issue with one side full-duplex and the other
half-duplex. One would expect speeds like this. I've seen those speeds.

Quote:
> Next I ftp-d to the 3.1 box using a different home computer and
> tranferred the same file. The results were much better. I get anywhere
> from 700KBps to 900KBps.

Fair enough.

Quote:
> Given all this you might assume the laptop is the problem. Maybe it is
> but I notice the number of collisions is above the 10% mark mentioned
> earlier.

So the problem you have is properly described by having collisions, but by
a remarkably slow network.

Quote:
> When using the computer that gets 900 KBps I get the following report
> from netstat:
> Total Packets in: 177
> Total Packets out: 218
> Colls: 149

> When using the laptop (8KBps throughput) nestat reports:
> Total Packets in: 165
> Total Packets out: 222
> Colls: 30

> Constrast this when I replace the 10baseT hub with a 100baseT hub and
> get 0 collisions on either computers when I run the test.

Again, I think that you have a half-duplex/full-duplex issue at hand.

Quote:
> My guess is the laptop begins to struggle when the collisions exceed a
> certain threshold.

Nope.

Quote:
> Do you have any ideas why the number of collisions are so high and
> what I can do to reduce their number?

Fix the link as being half-duplex. It seems that autonegation fails, or
settings are incorrect.

HTH,

EJ
--
Remove the obvious part (including the dot) for my email address

by **Noe** » Thu, 26 Dec 2002 08:45:16

You two were right. I had one side on half-duplex the other on
full-duplex. Changing this to half-duplex on both sides improved my
performance to very reasonable levels - ftp reports 900 -1000 KBps.
However my collisions are still occurring. This really disturbs me.
netstat reports aproximately 3000 collisions for 2100 packets. This on
a cross-over cable! What is it going to be like when I introduce more
workstations? What could be the reason for these collisions?

Noel

by **Eirik Se** » Thu, 26 Dec 2002 08:54:06

Quote:> You two were right. I had one side on half-duplex the other on
> full-duplex. Changing this to half-duplex on both sides improved my
> performance to very reasonable levels - ftp reports 900 -1000 KBps.
> However my collisions are still occurring. This really disturbs me.
> netstat reports aproximately 3000 collisions for 2100 packets. This on
> a cross-over cable! What is it going to be like when I introduce more
> workstations? What could be the reason for these collisions?

Change to full-duplex on both sides. This will not only eliminate
the possibilities of collisions, but also give a theoretically
double speed, as both ends can transmit at 100 Mbps at the same time.

- Eirik
--
New and exciting signature!

by **Joe Doupn** » Thu, 26 Dec 2002 07:34:35

> You two were right. I had one side on half-duplex the other on
> full-duplex. Changing this to half-duplex on both sides improved my
> performance to very reasonable levels - ftp reports 900 -1000 KBps.
> However my collisions are still occurring. This really disturbs me.
> netstat reports aproximately 3000 collisions for 2100 packets. This on
> a cross-over cable! What is it going to be like when I introduce more
> workstations? What could be the reason for these collisions?

> Noel

---------
It turns out you are worring about nothing. That's convenient
to know. The assumption is a collision is some horrible thing, taking
up lots of time etc. It isn't, it is normal, it takes next to no time.
A typical ftp transfer will show a collision on every other incoming
data packet, it's normal, it causes a nearly immeasurable slowdown.
They are not "bad," they are the normal way of sharing the medium.
Joe D.

Very Computer

Too many Collision on subnet

Too many Collision on subnet

Too many Collision on subnet

Too many Collision on subnet

Too many Collision on subnet

Too many Collision on subnet

Too many Collision on subnet

Too many Collision on subnet

Too many Collision on subnet

Too many Collision on subnet

Too many Collision on subnet

Too many Collision on subnet

Too many Collision on subnet