[c-nsp] Failover with 3550-SMI

[Marco Matarazzo]

Hi all,

I manage a colo-like network, quite simple:

Two borders (7206/NPE400/256Mb) each talking eBGP with two upstreams (double
feed each upstream) and iBGP between them.
Those two 7206 trunk each into a different 2950T-24, that are also connected
together. Then finally a couple of 3550-48-SMI, each connected with both
switches, that talk RIPv2 with the borders (they announce the connected
routes, and take the default gw from the borders). All the ports are layer3
(except the links to the switches that are layer2, configured for spanning
tree).
In this setup I can only see the 3550 as Single Points of Failure.
Now some customers want dual ethernet for failover. I can see three
scenarios:

1) Put them on the two 3550 layer2 ports, and route the vlan directly on the
border, using HSRP on the border for default gateway failover.
2) Upgrade the 3550-SMI to EMI, and run HSRP between the two customers
facing interfaces.
3) Don't upgrade the 3550, add little router to the customer and advertise
the default route to this router via RIP.

Is there something I overlooked? Could I use some other failover method?
Which of the three would you use? I prefer the third solution, as it keeps
the border and the distribution completely separated, and it seems more
manageable. But that's also the most waste of IP space (need two additionals
/30). Number 1 could also be feasible, but if number of customers wanting
dual ethernet rises, I think that 30 or 40 HSRP istances together with two
full BGP feeds and three downstreams, netflow, RIP and some access-lists
could kill the router (transit traffic is 40Mbps, router is as 25/30% CPU).
Number 2 is expensive (the upgrade to smi to emi), and again, could
theorically a 3550 handle 48 HSRP instances easily?

Thanks!
]\/[arco