Late Collisions

[Mike Hassell]

I have two servers that have become quite unhappy over the past few days since I've upgraded the NIC in the second server too 100MB. Both servers are talking on the network and can "speak" to each other, however their transfer rates are incredibily poor, since I upgraded. Both servers are attached to a Cisco Catalyst 4006 switch.

The newly installed NIC is running at 100MB Full Duplex, while the older server is running at 100MB Half Duplex, again both on the same switch.

I tested transfer rates using ftp and transfering a 60MB file to and from the machines, both only getting around 50Kb/s, which is pretty horrible considering they are on the same switch. We noticed the problem during our normal backups when the clients were taking forever to backup.

Here is the output from the server where the newly installed 100MB FD NIC is installed:

Alignment Errors = 0
FCS Errors = 0
Single Collision Frames = 1018
Multiple Collision Frames = 44
Deferred Transmissions = 4
Late Collisions = 338341
Excessive Collisions = 2
Internal MAC Transmit Errors = 0
Carrier Sense Errors = 1
Frames Too Long = 0
Internal MAC Receive Errors = 0

What do I need to do to avoid the "Late Collisions" and increase the speed to normal 100MB transfer rates?

I'm assuming there should be no problem with one card talking half duplex and the other talking full duplex as the switch should take care of these issues, but I may be wrong here.

Where should I start to troubleshoot and what steps do I need to take next? Any help would be greatly appericated, thanks.

-Mike

[A. Clay Stephenson · Accepted solution]

Hi Mike:

After much frustration, I have learned to never trust Auto-negotiate on either end when connected to these servers. Hard set to whatever value you like on the servers and then hard-set the values on the switch as well and I think your problems will completely disappear.

Clay

[Patrick Wallek]

Make sure that the ports on the switch that each system is plugged into are set to the appropriate speed and duplex for that server. It really sounds like a duplex mismatch between the server and the switch port.

I have found much less troublesome to hard set both the switch ports and the cards on the servers to the speed and duplex you want. I hate auto-negotiation. It hardly ever works.

[James R. Ferguson]

Hi Mike:

Make sure that auto-negotiation is turned OFF and that your configuration file settings match your switch. To see the current setting, do:

# lanadmin -x

Regards!

...JRF...

[Sanjay_6]

Hi Mike,

Suggest you have the port on the switch configured for 100MBPS fullduplex, no-autoneg for the server with 100FD card and the other port configured for 100MBPS half duplex, no-autoneg. This is to avoid auto neg which is always a problem.

Hope this helps.

Regds

[Darrell Allen]

Hi Mike,

Ditto on the auto-neg and duplex mismatch between a server and a switch.

Also, I'd set 100FD on the NIC (and switch port) that is currently at 100HD. It should support full duplex.

Darrell

[Mike Hassell]

Thanks everyone for the responses. You all hit it on the head with being a duplex mismatch issue, however I'm a bit confused about the configuration file for my A3514A dual 10/100 NIC that is in this Kbox. I knew better than to leave auto negotiate on, on the switch, as both of these ports that I'm referring to have been hard set from the Cisco side of things. I also thought I had my side (NIC configuration) hard set as well. Take a look at the following:

/etc/rc.config.d/hpgsc100conf

#####################################################################
# @(#) hpGSC100conf PHNE_13940 $Revision: 1.2 $ $Date: 98/01/29 12:00:01 $
# hpGSC100conf : contains configuration values for HP 100Mb interfaces
#
# HP_GSC100_INTERFACE_NAME Name of interface (lan0, lan1...)
# HP_GSC100_STATION_ADDRESS Station address of interface
# configured through lanadmin(1m)
# Prefixed with "0x".
# HP_GSC100_DUPLEX Half or full duplex mode (half is default)
#
# HP_GSC100_SPEED Manually set the driver's speed. Can be one
# of: 10HD, 10FD, 100HD, 100FD, auto_on
#
# The interface name, major number, card instance and nmid may be
# obtained from the lanscan(1m) command.
#
# The station address and duplex are set through the lanadmin(1m) command.
#
#####################################################################

Now here is where I configure this instance of the card:

HP_GSC100_INTERFACE_NAME[2]=lan2
HP_GSC100_STATION_ADDRESS[2]=0x001083361313
HP_GSC100_SPEED[2]=100FD


Does 100FD in the speed setting not clearly state 100FD? Well in lanadmin it shows the following:

Description = lan2 Hewlett-Packard 10/100 TX Half-Duplex Hw
Rev 0. TT = 1500

So I'm assuming it's set to half duplex, while the switch is set to full causing the late collisions. Lanadmin also shows the speed for this instance of the NIC to be '100', while on my other server where I have the same NIC running at half duplex shows the speed as '100000000'. This configuration file seems a bit flaky to me, as I'm passing the same parameters on the other host and I get different results. Both of these servers are running HP-UX 10.20 and the March quarterly patches. Any ideas? Thanks.

-Mike

[Patrick Wallek]

HP_GSC100_INTERFACE_NAME[2]=2 things.

1) Are you sure that the card instance is lan2? Do a lanscan and see what cards you have.
2) I would get rid of the value that you have set in the station address field. Let the card set its own MAC address.

I think the difference you are seeing with the speed (100000000 vs. 100) is a patch issue if I remember correctly. The speed on both is 100 Mb. The lanadmin output should show Full Duplex once the card is set to FD.

How is the [1] instance of that card set up?