[c-nsp] Splitting root bridges (was Migration to Rapid-PVST)

[Matt Stockdale]

I've seen the concept of splitting your roots across your 2 core
switches like this mentioned before, but something about this
conventional wisdom escapes me - 

  If your goal for this setup (2 routers, 2 core switches, 2
distribution switches, N access switches) is redundancy instead of
performance, does it make sense to split the load like this? Isn't it
possible to reach > 50% load on both sides, such that were one half to
fail, the remaining portion couldn't pick up the slack? How about a
situation where you are at less than 50%, but different vlans on
different sides are using different features, resulting in uneven CPU
load - You could end up with plenty of fabric, but the hamsters would be
falling over exhausted when you put them all in one wheel.

 What would the drawback be to just pushing it all through 1 side all
the time? (aside from not knowing if the standby gear is actually
working, since it's not "live")

Matt

--
Matt Stockdale
Network Engineer
Logicworks

On Mon, 2006-05-15 at 09:53 +0200, Per Carlson wrote:
> On Sat, May 13, 2006 at 02:50:27PM +0200, Robert Hass wrote:
> 
> >Basing on my first post with network diagram how many instances you
> >recommend to have ? And which switches should be root-bridges for these
> >instances.
> 
> The number of instances doesn't matter that much, a couple would
> suffice. I would choose the switches connecting to the 7304's as
> primary respective secondary root. It's the forwarding path
> towards those switches that should be optimized. 
> 
> Are you doing any load-balancing on the 7304's, i.e. each 7304 is
> the primary HSRP gateway for half of the Vlan's? If not, I would
> purpose such a scheme. The easiest (to operate and maintain) way
> is to allocate a large contigous Vlan range to each 7304. Like
> Vlan 1-2047 to one 7304, and 2048-4094 to the other.  Put each of
> the ranges into separate MST instances, and let the switch
> connecting to the 7304 be primary root for one of the instances,
> and secondary for the other.
> 
> >We have configured bpdu-filter and SPT portfast at customers ports.
> >Most of customers using static routing, so We're terminating them on
> >both our 7304 routers providing redundancy with HSRP.
> 
> That will make the STP converging quite fast. 
> 
> >Most of VLANs carring up to 2-6 MACs. I worst case it will be 50 VLANs
> >* 6 MACs = 300 MACs.
> 
> That's a low number of MAC-addresses. No problem here either.
> 
> >>Be aware that there are two different MST implementations in IOS,
> >>one pre-standard and one standard. As far as I have seen, those
> >>will interoperate without any problems. But running the same type
> >>in the whole network is probably a Good Thing (tm).
> >
> >I have 12.2(25)SED and 12.2(25)SEE on 3550/3560/3750 and 12.1(22)EA7 on
> >2950. Is MST implementation in 12.1(22)EA7 is same that in
> >12.2(25)SE{D/E} ?
> 
> Not sure, we don't have any 2950's in our network. If you dig up
> the Config Guide for the newest IOS version and checks the "MST
> Config Guide", you will find info about from which IOS version the
> standardized MST is used.
>