Connectivity/Speed problem

michael.spence · ‎07-12-2005

After installing a catalyst 4507 I have users complaining that it takes longer for their computers to bootup. One user's computer takes 1 minute 45 seconds to bootup. This user, as well as others, also gets this error, "Your computer was not able to renew its address from the network (from the DHCP Server) for the Network Card with network address xx.xx.xx.xx.xx.xx. The following error occurred: Your computer will continue to try and obtain an address on its own from the network address (DHCP) server."

If I do a show interface I see a high number of input and CRC errors. We tested the cables and they seem to be fine. We have reimaged the computer, but the problem persisits. We have moved a select group of users back to the original switch the problem got worse for the users in the old switch.

We have not tested fiber or analyzed the traffic being generated in that building. The entire building is on its own VLAN segment.

I am running out of options, so any help will be appreciated, greatly. thank you.

thisisshanky · ‎07-12-2005

Enable spanning-tree portfast on each port to which a PC is connected. try to hardcode the nic speed and duplex settings on the switch to see if the errors clear out.

Sankar Nair
UC Solutions Architect
Pacific Northwest | CDW
CCIE Collaboration #17135 Emeritus

michael.spence · ‎07-13-2005

All non-trunking ports are set to portfast enabled. Also all computers and switchports have their respective speed and duplexes hardcoded. If the device can do 100MB Full, then the switchport is hardcoded the same, if not then both are set to 10 half. After doing this our problem continued.

Now we are wondering if there may be a hardware problem with the gbics or a problem with the Fiber.

I appreciate all the input keep it coming. Thank you.

daniel.price · ‎07-12-2005

If the switch ports are at AUTO negotiate and the PC

are set to static FULL/100 there will be CRC due to

the switch not receiving the AUTO signal from the PC and defaulting to the standard of 100/HALF. This could be very problematic in a large single VLAN where the CRC packets will begin causing the switch CAM to timeout early, and upon the reset of the CAM the switch will unicast flood all the frames. Check the duplexing in the switch, if its AUTO, and has AUTO detected HALF, and you are incrementing CRC. Then solve that issue first. Anywhere the switch AUTO DETECTS 100/HALF, set static to 100/FULL, or locate the PC and set it back to AUTO AUTO.

Thanks

Dan Price, CCIE6747

michael.spence · ‎07-13-2005

VLAN has roughly 150 users. Ports and Nics have their speed and duplex settings hard coded. Are you saying that I should set all my devices to Auto for their speed and duplex settings? I just want to be sure I understand, because I am under the opinion that hardcoding both the device and the port is the best practice, if not let me know. Thank you.

Kevin Dorrell · ‎07-13-2005

IMHO, they are both "best practice". But you should try and stick to one or the other: either auto or hard-coded. Personally I go for auto, because that is the way PCs are delivered. But some people prefer to pin everything down.

I really don't know why it is implemented in this way. If you hard-code one side only, I see no reason why it should not still be able to negotiate, but only allowing the hard-coded values. But as it is, hard coding one side disables the autonegotiation. I'm sure they had their reasons.

Kevin Dorrell

Luxembourg

burleyman · ‎07-13-2005

We had a very simular problem. Could you post your config?

Mike

b-ulrich · ‎07-13-2005

Hi, we had similiar issues twice when we pulled out a 3500XL switch and installed a 3550-3560. Everything had to be set to auto/auto.

Even though the cabling was tested and verified for Cat 5, we had to completely re-cable two branches to get this to work correctly as intended. The weird thing was that the 3500XL would connect at 100/Full but the 3560 would not. Maybe the newer switches are more susceptible to cabling. That is all we could gather from the situation.

Bill

michael.spence · ‎08-12-2005

config attached.

brammathorn · ‎07-14-2005

On the client ports, have you ran the macro "Service Host"? Also, do you have this 4507 connected to other switches? If so then check your STP configuration. There may be an older switch thinking that it is the STP root.

michael.spence · ‎08-12-2005

We do have another switch connected to the 4507 in question. But STP shows the correct root bridge.

Here is the wierd thing, no the users connected to the switch state that their computers are working much faster and swear that we have done something to fix the problem.

burleyman · ‎08-12-2005

Do your modules have POE (power over ethernet)?

bbaillie · ‎08-15-2005

The behavior of asking for a new IP address is suspicious, a workstation (Windows 2000 or XP I assume) will attempt to ping its default gateway and if this fails, it will attempt to get a new IP address. Number one, have a workstation release its IP address and see if the renewal is timely, two ping from the workstation to the DHCP server to attempt discovery of any problems between them, three post the show interface outputs from the workstation port and the uplink port to the rest of the network from this switch. If "show controllers" for the interfaces works on this switch it would also be enlightening. The users now seeing a speed improvement is cold comfort when the issue may be the path to the DHCP server (IP addresses will expire eventually). I have had problems with GBIC-LX using multimode over a longer than normal distance so if you can confirm the distance of the fiber run including patches may be worth the effort.