Resilient topology

[trismegistus]

Hi

I have 2 identical clusters, which I want to mesh for resilience. I have 2 catalyst 3508 switches at my core, off of which hang 2 catalyst 2950 switches, which feed out to the user stations. I want to connect both 2950 switches to both 3508 switches to form this mesh.

I am unsure of what technology I should use to allow this to function correctly. Both clusters are in use all the time, so I don't have a redundant side, but I would like 1 side of the core to feed traffic around in the event of the other side going down.

I realise this is not a snap answer, and would be grateful for any advice forthcoming. A point to the correct technology solution would be a great start.

 

[svermill]

Can you give a little more detail regarding your L2/L3 topology. Is this strictly a L2 network? If so, spanning tree will do the trick for you. We can get into the specifics of what to hook to where once clear on the topology details. CCO has some great LAN design guides also.

 

[trismegistus]

Hi Svermill

The situation changed after another chat with my client. He requires 2 x 3508 routers, each coming from a different server farm. Hanging off each of these is a 2950, so we have 2 identical topologies. The next step down is that both plug into a Lucent ? switch.

The links from the 3508's to the server farm are fibre, the uplinks between all switches are etherchannel. The client would like a fibre link between the 2 x 3508 as well.

I have hunted down (no pun) how to configure the Lucent switch for etherchannel, and I am confident the config should be straightforward, but I am worried that spanning tree is going to close down one side of the topology, pushing everything across the fibre link to the other 3508, leaving one side redundant.

It is all layer 2, no vlans etc, so apart from this it is no problem. I have very little switch config experience, i'm more a security man, any advice is helpful.

Also, will the lucent switch cause a problem with the stp?

Thanks in advance

 

[svermill]

Well...

You mention routers but say it is all L2. Routers are L3. So, are the server farms in different subnets? Presumably they are. Which means the routers, once interconnected, will route between these subnets at L3. In this case, two completely separate instances of spanning tree will exist and will not cross the L3 router boundary. I'm still not sure I fully understand your topology though. 3508 to a 2950 to a Lucent? Or are the Lucents on the other side of the 3508 (Servers->2950->3508->Lucent or Servers->Lucent->2950->3508)?

 

[trismegistus]

Sorry, it was a typo. I meant switch.

To visualise it; In a line on top are Server>3508 Switch>3508 Switch>Server. Then coming down, each 3508 has a 2950 hanging off of it. The 2 x 2950 then both link into a single Lucent Switch, forming a ring.

Sorry about the confusing explanation earlier regarding this. My main concern is that STP will close one of the uplinks between the cat 2950's and 3508's, sending all traffic across the two 3508's and down the other side.

Also, I need to sort out the STP for the Lucent but i'm sure thats not going to cause an immense problem.

Thanks for your patience

 

[svermill]

I'm on the road on a short-notice trip so I'm pressed for time. But if the 3508s are L3-enabled, the 2950s hanging off of them won't be in the same spanning tree. Is there a subnet between the 3508s or is everything in a single subnet? Also remember that you can deterministically set the root to influence which ports are blocked and which aren't.

 

[svermill]

OK, I'm back in town for a while. Are you making progress? If I read your description correctly, you could have the 3508-3508 link blocked. Set the Lucent as the STP root and see to it that the 3508-3508 link is the one that gets blocked. That works as long as there isn't more L2 below the Lucent that we haven't discussed. I think.

 

[trismegistus]

Hi

Right, this is the situation. I got everything going, and set the lucent to priority zero to make it root. I changed the priority on the catalysts to more but I don't really have much idea about the Lucent, so could only suppose it had become the bridge. I asked about but the guy who admins the Lucent didn't know what spanning-tree is!!!!

Anyway, everything connected up fine apart from the link between the 3508's which went turbo. The interface was flappiong like mad and recalculating spanning-tree every 10 seconds.

I am stuck on this one. I am going to go back and mess around with the lucent next week. I don't know of all the reasons why the 3508 interfaces would do this. Any suggestions?

I also had the idea that it might be better to link each 3508 to the opposite sides 2950 for redundancy. I really need a better input from the switch admin regarding the Lucent.

Thanks for you rattention on this one. I realise its a bit difficult to envisage etc, but any ideas are welcome.

 

[svermill]

Hmmm. The flapping link would require that you run some debug. You did set the duplex and speed to match on both ends? Cisco to Cisco autonegotiation normally works fine, but I still prefer to set speed/duplex manually.

Yes, the additional links to the 2950s sounds good. But you need to get a firm grip on your spanning tree topology first. If the Lucent is root, the link between the 3508s should probably be in a blocking state with the way you have things set up now.

Anyone else care to jump in?