Expanding an HP IRF topology

Question

I'm examining options for a future expansion of our infrastructure. We rent space at a colo so we don't have a dedicated datacenter of our own. This means that our next major expansion will likely involve two discrete cages not immediately adjacent. I'm looking at ways of bringing my network into a new cage.

Our current topology has a pair of HP A5820's in an IRF pair, and all of our distribution switches linked to both routers through a 10Gb BAG/Trunk port. That's all well and good.

The question I have comes when we add another full cage. The way I see it, I have two major options:

1. Continue as I am, and connect the distribution switches back to the pair I already have.
2. Pick up another pair of A5820's and connect the new cage's distribution switches to that, and link that pair to the pair I already have somehow.

It's the second option that I'm getting some confusion with.

This is a pretty standard topology, in that I link the new switches back to the other pair in a linear topology. Anything in Cage 2 needing to go to the other switch will have to transit the full stack, though. I know this will work.

The question here is, can I short-circuit that with a ring topology? Does IRF allow that?

This would allow a shorter transit of packets. That said, I don't know if IRF allows this kind of setup. If it doesn't...

Or I could create a second IRF device in the new cage, and link the two together with aggregation ports and not bother with IRF at all between them.

Considering each cage would have all of my distribution switches connected to both pairs, the IRF links may not actually get that much traffic across them. If so, I'm over-thinking this.

Am I over thinking it? And if not, does IRF allow a ring topology?

Answer 1

I'm not HP savvy but I did some quick Googling (hp irf ring topology) and found a couple of docs that indicate it does support a ring topology.

• c02648772.pdf

Quote from page 5:

IRF topologies

Create an IRF virtual device in daisy chain topology, or more reliably, ring topology, as shown in Figure 2. In ring chain topology, the failure of one IRF link does not cause the IRF virtual device to split as in daisy chain topology. Rather, the IRF virtual device changes to a daisy chain topology without affecting network services.

This page has the same quote so at least that seems like a possibility.

HP seems to be recommending the ring but beyond that I can't speak to HP gear.

Answer 2

IRF definitely supports a ring topology, and as long as that setup makes sense for your environment, then it's definitely the preferable way to expand.

Answer 3

IRF is nice technology. I have experience with them. Yes, you can expand the stack pretty well. Go with a ring topology by adding new switches like you show in the second picture. However you have to keep in mind several considerations: stack is nice, but it can fell apart with, so in this case two IRFs with MAD (Multi Active Detection) will work better, but provide more overhead (you'll have to provision two stacks instead of one, may be you'll need to reconsider your failover scenario etc)

I work with IRFs for 3 years now and I have seen several split brain cases, but those all stacks were configured without MAD. In all cases networks were pretty dead :) So MAD is really a must, when you configure stacks :)

Answer 4

Yes you can make ring topology for IRF For instance:

Sw1: irf-port 1/1 Port group (interface 1) mode normal

Irf-port 1/2 Port group ( interface 2) mode normal

Sw2: irf-port 2/1 Port group (interface 1) mode normal

Irf-port 2/2 Port group ( interface 2) mode normal

Sw3: irf-port 3/1 Port group (interface 1) mode normal

Irf-port 3/2 Port group ( interface 2) mode normal

Then connect every irf-port x/1 to x/2

And from irf-port 3/1 to irf-port 1/2

Ring topology is great because it provides full redundancy.

I hope this helps.