Cisco Switches, Broadcom NICs, Random Network Semi-dropping.

[justinrhodes]

'ello folks...

So here's the situation. I'm in a company where roughly 100

workstations, 15 printers, and 8 servers all drain into one closet in

which we have a series of 3 Cisco 3500 XL switches (I forget exact

model at the moment). The switches are chained together via GBICs with

copper firewire. The workstations in the building are almost all using

Broadcom NetXtreme Gigabit Ethernet cards (HP machines). At completely

random intervals, a machine will drop part way off the network - for

example, it will be able to ping servers A, B, and C, but not D, E, F.

All servers are running fine and everyone else is doing beautifully.

If I go to the switches and change the port the workstation is

connected on, it can then see all the servers. If I plug a different

machine into the original "problem" port, the new machine can access

everything just fine.

Any ideas? The ports on the switch are all auto-negotiate. The NICs

are all auto-negotiate. The network cables shouldn't be the issue - it

happens with new and old cords.

We never had this problem until we moved to the Cisco switches. Before

this, we had the same network setup, but with 3Com switches.

I'm at a loss to what this could be. Any suggestions would be great!

Cheers

[Josef Oduwo]

Check your VLAN setup, or if possible post your running configs.

Cheers,

Josef.

[justinrhodes]

The only vlan configured is the default. I'll post configs when I get back to work on Wednesday.

[justinrhodes]

------- show running-config -------

Building configuration...

Current configuration:

!

version 12.0

no service pad

service timestamps debug uptime

service timestamps log datetime msec localtime show-timezone

no service password-encryption

!

hostname company1

!

no logging console

logging source-interface VLAN1

logging 10.200.1.24

enable secret 5

!

!

!

!

!

spanning-tree uplinkfast

!

ip subnet-zero

!

!

!

interface FastEthernet0/1

random-detect

spanning-tree portfast

!

interface FastEthernet0/2

spanning-tree portfast

!

interface FastEthernet0/3

spanning-tree portfast

[ Ports 4-47 are all the same... ]

!

interface FastEthernet0/48

spanning-tree portfast

!

interface GigabitEthernet0/1

!

interface GigabitEthernet0/2

!

interface VLAN1

ip address 10.211.162.250 255.255.255.0

no ip directed-broadcast

no ip route-cache

!

ip default-gateway 10.211.162.1

access-list 62 permit 10.200.1.20 0.0.0.3

access-list 62 permit 10.200.1.24 0.0.0.3

snmp-server engineID local xxx

snmp-server community RW

snmp-server community RO

snmp-server community RO 62

snmp-server community RW 62

snmp-server trap-source VLAN1

snmp-server system-shutdown

snmp-server enable traps snmp authentication linkdown linkup coldstart

snmp-server enable traps vlan-membership

snmp-server enable traps config

snmp-server enable traps entity

snmp-server enable traps bridge

snmp-server enable traps hsrp

snmp-server enable traps c2900

snmp-server enable traps mac-notification

snmp-server enable traps vtp

snmp-server enable traps cluster

snmp-server host 10.200.1.24 trap

!

line con 0

transport input none

stopbits 1

line vty 0 4

password

login

line vty 5 15

password

login

!

ntp clock-period 11259130

ntp server 10.200.1.1

end

[daniel.fischer]

hello justin,

we have a similar problem with Broadcom NetXtreme Gigabit Ethernet nics. when running in 100 Mbit/s half duplex mode, the connection between a c2940 switch and a host was "lost" in a strange way. the link was up, line protocol up and runing, there was even an entry for the mac address of the client in the switch but it was not possible to comunicate with that client. not even with another host connected to the same vlan on the same switch. we captured some traces with a network analyzer and figured out that the nic produces runts (frames shorter than 64 octets) and misaligns.

the cisco switch does not seem to like these runts.

the behaviour of the 2940 switch depends on the ios version, with 12.1.19, communication is possible, although having runts, with 12.1.22 not.

i still have an open case with cisco concerning this problem, but one workaround could be to fix the speed to 100 full. make shure you do not get a duplex mismatch.

best regards

daniel

[justinrhodes]

Daniel,

Thanks for your response! I've been tracking some things over the past couple of days and thus far what you said fits in rather well. Haven't seen any runts, but I've noticed the only switch where the NICs are mysteriously dropping off is on 12.0(5.2)XU. The other two switches are on 12.0(5)WC10 and 12.0(5.3)WC(1). (Yeah, I know it's old... Still trying to get the ok to bump up the ioses and make them consistent).

I've just now set up my workstation (which dropped off this morning) to be 100 full with matching settings for it's switch port (on the problem switch). Now to see if it drops again.

Thanks for the input!

Cheers,

- Justin