IRF setup / STP / loop detection

[Trukosh]

Hello,

i am relatively new to networking topics and plan some bigger changes in our environment.
I home some of you experts can give me a thumb up/down to my plans.

We do have 2 * HP 5800 (was 3COM) as core switches + 6 * HP 2900AL48.

Current setup is:
- The 5800 are connected with 2 cables with LACP/trunk.
- Every 2900 is connected with 4 cables, 2 to both 5800. The 2 Cables/ports are trunked.
- Rapid Spanning tree is enabled on all devices. The 5800 have been given highest STP priority.
- There are also some servers connected to the 5800 switches (one cable to each 5800), each with an active passive bonding.

What i don't like:
- Even with RSTP enabled (Perhaps there is sth wrong) every STP change (because of changes, upgrade or unplanned downtime) affects the whole network.
- I want the servers to have active active bonding to higher bandwith (not, that we can really use it ;-) - i just like to get the most possible out of the setup)

Planned changes:
- physical topology stays the same
- enable IRF on the 2 5800 switches and stack them (2 * 40Gig cables/adapter are available)
- set up (irf)trunks over switch border on 5800er switches
- trunk all 4 cables/ports on the 2900 switches (connected with 2 to each switch)
- servers trunk active active 
- disable STP on all switches
- enable loop detection on the 2900 switches for all ports

Expected improvements:
- 5800 switches bandwith from 20 Gbit to 40 (or is it even 80?) Gbit
- 2900 switches bandwith from 2Gbit to 4 Gbit 
- Server bandwith from 1 to 2 / 10 to 20 Gbit
- No more spanning tree changes
- the 5800er switches are totally redundant, so only effect of a downtime would be, that half of the bandwith has gone, but with disabled STP no disruption of network traffic
-  With 2900 downtime only this segment is affected - again no STP, so no disruption in network traffic

There is one exception:
Behind one of the 2900 switches there is another 2900 switch connected with fibre optics (2 trunked cables). (We can't change that and connect directly to core). Will i need STP here? In my understanding i do not, as there is only one (trunked) uplink, so loop detection to protect clients (or admins) from doing sth wrong is enough.

It's a long post, hot somebody will read it and give me some input, for errors or possible improvements.

Thank you very much!

Regards

[Vince-Whirlwind]

That all seems very sensible, except for one bit, and also I have two queries.

First query,
You say, "The 5800 have been given highest STP priority" - I hope you mean the two 5800s have different priorities, and both of them have lower priorities than the Access switches?
I don't understand why STP changes are an issue - STP is only there to prevent loops. Once a loop has been detected, it should very, very rarely ever have to recalculate its topology, unless you are constantly adding loops to the network.

Second query,
You don't mention your routing. This is importaqnt. Is it occurring on the 5800 with the lowest STP priority? (that would be ideal).

Bit that isn't very sensible:
" disable STP on all switches
- enable loop detection on the 2900 switches for all ports"

You need to keep STP enabled. If you disable it, then these "disruptions" will turn into network-killing broadcast storms. STP doesn't *cause* issues, it protects you from them.

Also, Loop detect has a different purpose: you put it on switch Access ports so that any device not under your management that causes a broadcast storm is detected and shut off.