NIC speed change

[FlatFive]

I have an IBM e series model 240 with a built-in IBM fault tolerance 10/100 NIC. I'm running 2k server, SP4, all the updates installed. Generally, it runs great. In fact, I have two, one running 2k server as a domain controller, backup server (Veritas 8), file server. The other is a terminal server, same version and updates.
My problem is this:
On my DC server, the NIC likes to drop from 100mbps to 10mbps. It does this regularly, usually overnight. I suspect it is happening during backup. It runs a full local backup and uses the agent for a full backup of the TS and another machine, running 2k pro and a Spam filter product.
I have changed switches, from a cheap D-Link to a Cisco Catalyst 3550. I thought that auto-negotiate might be the problem. I have tried auto and forced settings with the Cisco switch and in either case, it starts off at 100 and drops to 10 within one to two days. This problem never occurs on the Terminal server with identical hardware.

Has anyone exerienced a similar problem? If so, have you had any success in solving it? This has vexed me for months and I feel like I am running out of things to try. While not a huge problem, it is a real nuisance and does impact perfomance for clients.

Thanks

 

[Stiddy]

Have u tried hard coding the port on the swith to 100mbps/full duplex and setting the same on the servers nic card?

 

[FlatFive]

Yep. I've tried hard coding just the NIC or just the port or both the port and the NIC. Presently, I'm back on auto for both. In every case, it appears to start out at 100 (based on the Win2k "port status"), but the status shows it as dropped down to 10 within the first or second day, regardless. Port status does not show the active Duplex mode, but the management console for the switch shows it as "full" in every case.

 

[technome]

First update your Nic drivers.
A switch or Nic cards will lower speed or change flow control methods with a build up of errors

Not familiar with the switch but it appears to a managed switch, go into the management and check for errors on the involved ports, enabling or disabling flow control may help, check your spanning tree setting, turn off spanning tree if you do not need it for routing on the network. While your at it, check all ports for errors, a few error per month is normal on any port, more on server ports..small networks I would say very roughly 25 per month is normal, if your ports have hundreds or thousands it is most likely cable problems, sometimes network drivers/NICs. Flow control does not need to be enable on the switch and on the NIC card, enable it on the switch, remove it on the NIC setup.

After you check the switch settings..
Has your cabling been tested ?
If the cable to the switch is a long length I would replace it with perhaps shielded cable. Small patch cable, get a shielded cable. Check the cable is not near electrical interferance; cat5 is very resistant to emf but it is not perfect.

Lastly if you have a great deal of errors, you should have all your cables tested with a cat5 tester such as produced by Fluke. I have seen one cable create havoc for an entire network; at one site for years.

 

[FlatFive]

Drivers are the latest.

STP and flow control are disabled. I have swapped cables to see how that works. It's a clear, short run from the server to the switch, but I changed from a handmade to a premade jump.

I see surprising few error on the port in the switch mgmt console.

Thanks for the responses.

Now I wait and see...

 

[Skywalker305]

i have the same issue aswell but i can figure it out and it really annoys me beacause that is my webserver as well

 

[FlatFive]

Latest update - I have replaced the cable with a shorter run, using an out-of-the-package CAT5. It's not sheilded, but it is running thrugh an area with no electrical components within a foot of it. I have also enabled flow control (receive) on the switch port. It still does the same thing. I have also been watching the log on the switch and it didn't register any errors over 2 1/2 days.

I am considering changing the NIC, but past experience is that doing so screws up the functionality of DNS (running on this box).

Skywalker305 - is your problem with the same model box or NIC?

 

[technome]

Interesting problem...

I see you have been busy posting at other sites.

Good on the lack of errors, one possibility ruled out.

Any wild chance you have power management in bios or on the nic card enabled, I know you have probably checked this.
The reason I question this..problem general appears at night, this when the power utils do power grid switching. I have had numerous problems associated with this due to "iffy" power supplies, either rolling the machine over, even with BBU or causing other problems

At this point I would change the NIC. I would not use the same make /model NIC if possible.
Intel has the edge on NICs

Do not uninstall the first NIC, until your new NIC is installed. Your machine setting will be messed up if you uninstall the bad boy first. DNS,DCHP,WIN etc are initialized without any setting if the card is removed.

Instead of swapping the NIC, get the new NIC installed, clear the settings in the first NIC, immediately duplicate to the new NIC, uncheck all the protocol/clients on the first then disable the original NIC, after the new card shows a connection. If possible reboot.