topic Re: CSS and spanning tree in Application Networking

CSS and spanning tree

kejeusa — Fri, 22 Apr 2005 15:37:02 GMT

i'm slowing building the CSS ASR configuration and currently have

Private

Switch

/ \

master> CSS----CSS <backup

\ /

Switch

Public

i plan to add 2 more L2 switches to the 1 in pub area and 1 in priv area. i have the CSS vr peering configured and showing up correctly as master and back up front and back.

My question is that will some additional spanning tree configuration have to be (done/looked at) with the additional switches in

Re: CSS and spanning tree

dtodd — Sat, 23 Apr 2005 18:10:50 GMT

Hi:

What are you stating as the additional switches? The css's?

==DMT>

Re: CSS and spanning tree

kejeusa — Sat, 23 Apr 2005 20:42:34 GMT

in the private area will be 3550 or 3750.. on the public area it will either be 3550 or a 4000 series.. 2 switch connections in front and back for redundancy.. the css will be using ASR those are in the middle using 11506.. from all my checking it works in th current one access switch level but the concern is when spanning tree picks the best path that is pick the master css..

i noticed from a trace the master was sending vrrp messages which i think should resolve the prob but i'm unsure

Re: CSS and spanning tree

Gilles Dufour — Mon, 25 Apr 2005 08:21:45 GMT

if you double the number of switches I would recommend not to connect each CSS to each switch.

This has proven to be source of trouble due to spanning-tree.

SW1 ------ trunk -------SW2

|.......................|

CSS1...................CSS2

|.......................|

SW3------- Trunk ------SW4

If the link between SW1 and CSS1, let the CSS fails over instead of relying on spanning-tree.

This solutions works better from past experience.

Gilles.

Re: CSS and spanning tree

kejeusa — Mon, 25 Apr 2005 17:54:39 GMT

if i'm understanding you correctly..

only have 1 switch front and back per CSS?

the Backup CSS would have the same configuration the only common link the CSS's would be is the isc port..

once it (the backup)notices the master goes down via isc the ports for the back up would become active

Re: CSS and spanning tree

skumar1969 — Mon, 25 Apr 2005 23:42:46 GMT

Even I do not understand what Gilles says, assuming Gilles said that, what if the pri css goes complete dead?

What we did was that connect the Server interfaces to L2/L3 switches using NIC teaming. An L2/L3 port is connected to each CSS. Between the L2/L3 run the Etherchannel or Stack GBIC/trunk. It works fine since for a while ago. We had issues intialy like the MAC adress table got flodding on the L2/L3 switches, belive me we came out of that after enabling 'portfast' on the CSS ports and on the connected L2/L3 SW ports, I know its foolish but for some reason it works fine now.

Server1

/ \---->NIC Teaming

/ \

Sw1---Etherch---Sw2

/ \

master>CSS---isc---CSS

Switch

Public

Re: CSS and spanning tree

Gilles Dufour — Tue, 26 Apr 2005 07:38:37 GMT

if the primary CSS goes down completely, the backup will take over and it will have access to all servers through the switch it is connected to.

This is the easiest solution and a good solution.

I spent too many hours trying to explain to some people why their highly redundant network failed because of spanning-tree.

The solution I suggest will provide good redundancy and is quite simple so avoiding problems.

Gilles.

Re: CSS and spanning tree

Gilles Dufour — Tue, 26 Apr 2005 07:48:02 GMT

the ISC port is not used to detect if the master is active or not.

It is just used to send stateful information.

The CSS will use VRRP on each link to establish mastership and detect that devices are alive or not.

Gilles.

Re: CSS and spanning tree

kejeusa — Tue, 26 Apr 2005 12:12:44 GMT

i'm posting a drawing just to be clear on what you've said. if i'm incorrect please modify it beacuse these stick lettering is killing me..lol

Re: CSS and spanning tree

Gilles Dufour — Tue, 26 Apr 2005 17:32:23 GMT

could you resend as jpeg ?

THanks,

Gilles.

Re: CSS and spanning tree

kejeusa — Tue, 26 Apr 2005 19:20:09 GMT

not a prob

Re: CSS and spanning tree

Gilles Dufour — Wed, 27 Apr 2005 06:24:49 GMT

so you use 2 vlans on the public side and 2 vlans on the private side ?

This is ok, just more complex.

You'll need routes pointing to each vlan ip address for the vips.

All I can suggest is to find a solution where you can turn off spanning-tree on the CSS and avoid any routing protocols.

If you can do this, you will avoid potentially a lot of trouble.

Gilles.

Re: CSS and spanning tree

kejeusa — Wed, 27 Apr 2005 12:09:12 GMT

"so you use 2 vlans on the public side and 2 vlans on the private side ? "

i believe i will have to due to the use of terminal services used to manage servers. so i would have to have an internal vip address for terminal svc to access the servers directly. bypassing the public 'www'vip that outside users would use to access the stuff.

now when creating that internal vip setting to a vlan it.

would have to a redundant interface?

is there no command to turn off spanning-tree on css?

Re: CSS and spanning tree

Gilles Dufour — Wed, 27 Apr 2005 12:46:55 GMT

To turn off spanning-tree :

CSS11503(config)# bridge spanning-tree disabled

Gilles.

Re: CSS and spanning tree

kejeusa — Wed, 27 Apr 2005 13:13:01 GMT

thanks for all the help

i will try the spanning-tree command if after adding the extra switches creates a problem

Re: CSS and spanning tree

s.srivas — Fri, 29 Apr 2005 11:24:19 GMT

I looked at the diagram.

May be if you remove the cross connect (X, X) you will still get redundancy and probably avoid spanning tree. Let me know the outcome.

Re: CSS and spanning tree

kejeusa — Wed, 04 May 2005 14:07:49 GMT

i hadn't gotten quite to the ether channel connections but i deceide not to go with it and also disabled spanning-tree on the css after noticing alot of spanning tree anouncements every 2-10ths of a second

Re: CSS and spanning tree

kejeusa — Wed, 04 May 2005 14:09:29 GMT