Poor network performance in 3524XL

[lardum]

I have very poor performance in my Cisco 3524XL switch. There are 24 RJ-45 10/100 ports and 2 FDDI ports. I've attached 2 servers to the FDDI ports, both running at 1000Mbits/s and full duplex. On the RJ-45 ports i have 24 Mac stations connected in various speeds/duplex. Generally i have G3 Mac's and G4 Mac's.
The Mac's have very bad performance to the FDDI ports. Seems like there are timeouts and stuff. All the Mac clients have been complaining. The reason why we run some of the stations in half duplex only is that it runs much better for some reason. The Mac stations are running against the FDDI ports but we have an uplink through the FastEthernet0/24 port. I'm also getting error messages in the console about duplex mismatches:
1. duplex mismatch on GigabitEthernet 0/1 (not full duplex), with cisco2950-1 FastEthernet0/1 (full duplex)
2. duplex mismatch on GigabitEthernet 0/1 (not full duplex), with cisco2950-2 FastEthernet0/1 (full duplex)
3. duplex mismatch on GigabitEthernet 0/1 (not full duplex), with cisco 3524 FastEthernet0/24 (full duplex)

These error messages are repeated all the time. I've set the ports on all uplinks so they aren't autonegotiating. and the cisco 2950's aren't connected to the 3524 where these error messages appear.

Anyone have any ideas?

My Cisco 3524 config is:


!
version 12.0
no service pad
service timestamps debug uptime
service timestamps log uptime
no service password-encryption
!
!
!
no spanning-tree vlan 1
ip subnet-zero
!
!
!
interface FastEthernet0/1
duplex full
speed 100
!
interface FastEthernet0/2
duplex full
speed 100
!
interface FastEthernet0/3
duplex half
speed 100
!
interface FastEthernet0/4
duplex half
speed 100
!
interface FastEthernet0/5
duplex half
speed 100
!
interface FastEthernet0/6
description KRISTIAN
!
interface FastEthernet0/7
duplex half
speed 100
!
interface FastEthernet0/8
duplex half
speed 100
!
interface FastEthernet0/9
duplex half
speed 100
!
interface FastEthernet0/10
duplex half
speed 100
!
interface FastEthernet0/11
duplex half
speed 100
!
interface FastEthernet0/12
duplex half
speed 100
!
interface FastEthernet0/13
duplex half
speed 100
!
interface FastEthernet0/14
duplex half
speed 100
!
interface FastEthernet0/15
duplex half
speed 100
!
interface FastEthernet0/16
duplex half
speed 100
!
interface FastEthernet0/17
duplex full
speed 100
!
interface FastEthernet0/18
duplex half
speed 100
!
interface FastEthernet0/19
duplex half
speed 100
!
interface FastEthernet0/20
duplex full
speed 100
!
interface FastEthernet0/21
duplex half
speed 100
!
interface FastEthernet0/22
duplex half
speed 100
!
interface FastEthernet0/23
duplex half
speed 100
!
interface FastEthernet0/24
duplex full
speed 100
!
interface GigabitEthernet0/1
spanning-tree portfast
!
interface GigabitEthernet0/2
duplex full
spanning-tree portfast
!
interface VLAN1
ip address 192.168.1.10 255.255.255.0
no ip directed-broadcast
no ip route-cache
!

 

[krr]

Hmm, there is more than one thing wrong in this setup. First, your uplink ports are GigabitEthernet, not FDDI. GigabitEthernet is always full duplex. There is something called flow control negotiation for GE, but no autonegotiation like with FastEthernet.

The messages you see tell me that there is other non-Cisco network equipment connected to port Gi0/1. These are CDP messages (Cisco Discovery Protocol). You have to disable CDP on ports that connect to non-Cisco switches. Otherwise CDP well "see through" the other switch and find the next Cisco switch along the line, which results in false messages.

When you explicitly set speed and duplex on a port, make sure that the connected MAC interface also has the same settings defined (NOT autonegotiate).

Hope this gets you a step further. To give you more hints I'd have to know what exactly is connected where and how it is configured.

Cheers *Rob

 

[lardum]

Thanks for your response.

Here is the setup:
FE0/1-0/23 - all Mac OS 9.2 stations
FE0/24 - uplink to non-Cisco equipment
GE0/1-0/2 - Compaq servers with Intel fibre cards

All Mac stations are set hard to either full/half duplex. Based on the research we've made we found out that some of our Mac stations had better performance running in half-duplex.

The Compaq servers have some settings for flow control but these don't match the names in the Cisco switch. However in the switch i have assymetric or symetric flow control. The options on the Intel NIC are: Off, Respond, Generate, Both on, Hardware Default.

Basically you want me to run no cdp enable on FE0/24, GE0/1 and GE0/2? or maybe on all ports since i don't have any Cisco equipment connected to this particular switch. The 2950's are connected to another Intel stack.

 

[krr]

The Intel cards will respond to CDP. This explains why you see port Gi0/1 in the messages.

Could it be that the server on Gi0/1 has an additional FastEthernet card connected to some other switch and has "adapter teaming" with load balancing enabled?

As for the flow control: Hardware Default should be ok and correctly negotiate with the switch.

And please remove the "duplex full" from Gi0/2 because this might disable the negotiation (GigabitEthernet is always full duplex anyway).

Good luck *Rob

 

[mandar22]

Hi Gurus,

Lardum could u pls. let me know what is your status now as i am also facing somthing like u only...

Mandar

 

[Zelandakh]

Lardum, what timeouts are you getting? Slow transfers or slow updates on Chooser?

It could well be appletalk seed router issue.

Either set up a seed router or set all macs manually. This solved our speed issues.

 

[lardum]

Ok this is what i've done:

1. Turn off flow control on switch and Gigabit Interfaces.
2. Turned off CDP on both Gigabit interfaces.

Result: A little better performance however still occasional hangings on Mac clients.


Zelandakh: I'm not a Mac expert, but can you give me more details on what you mean?

 

[Zelandakh]

ok, appletalk 101 now convening...

appletalk kinda works like ip for broadcasting. They use something like dhcp to get an address normally.

There are 2 components, network ID and node (client ID). If you have 2 Macs on different network IDs they takes ages to show up on each others chooser list.

Set the network ID manually on each Mac at (say) 65240 then set the node of each to something unique. Once all Macs are set, restart all printers and PC servers. These will start up and pick a node on an existing network ID.

You should find this makes things smoother. Though because you've got some set manually and some set automatically, you can get clashes when a manual comes on to an address taken by an automatic.