Would you implement a Proprietary L2 ECMP Protocol today ?

When FabricPath came out a few months ago, I didn’t pay it much attention. In the the presentation that I got, the Cisco Product Manager assured me that it was actually a strategy that encompassed the proposed standards and then “extended existing technology” with new and enhanced features. Good, I thought, TRILL / SPB are both covered.

I was doing a review of the FabricPath documentation. It’s now clear that FabricPath is a fully closed, Cisco only implementation of L2 ECMP routing. It’s not a compatible extension or enhancement or a fancy product name. No, it’s an ‘EIGRP style’ we own it and you can’t have it. What’s that all about ?

OK, in practical terms it is functionally similar to TRILL or even 802.1aq SPB, but it isn’t compatible with either. It looks like it COULD be compatible without too much work. Ron Fuller did a nice comparison here.

Where is the standards compliance then ?

Well, in a very slimy bit of marketing, it is fully compatible with Data Centre Bridging (DCB). Of course, DCB has absolutely nothing to do with L2 ECMP. That’s like saying that FabricPath will “be compatible with Ethernet”. So I was suckered by a fast talking product manager.

Stupid me.

So what the problem ?

Here’s the problem ? As a customer, I like Cisco tech support and my relationship, but their core switching products are not very exciting. The Nexus 7000 is missing features ( and others are running very late against predicted delivery times) , has low performance compared to competitor products and comes with a stunning price tag. Lets not mention power consumption. The constant whining noise about UCS from my account manager is all nice and shows some exciting promise, but I need big switches today that will last for five years or so. With what I know today, I’m not convinced that Cisco has it.

Network World has been consistently predicting that Cisco is losing share in the Ethernet switch market which is proof that the shipment delays, high pricing, and poor software is starting to impact Cisco sales. With HP and Juniper planning to undercut Cisco pricing in the year ahead against a backdrop of tight economic conditions, it’s not clear that Cisco is going to come back either.

Meaning ?

This means that you should probably be planning for a multivendor LAN backbone. Maybe not today, but sometime in the future it’s likely that you will have more than one vendor in your data centre core. Gosh, even Gartner is predicting it in a report Debunking the Myth of the Single-Vendor Network (Gartner, Nov 2010). Given that Gartner is usually about a year behind is picking major trends (helps with getting those ‘predictions’ correct eh ? ) you can bet that the prevailing attitudes must be changing. Gartner only writes what someone else has already told them.

This is a return to the policies of Networking Design from pre-2000 era. Many companies had a dual vendor policy to protect themselves from a bad product or a vendor failure.

Standards matter. They REALLY matter.

I’m going to sound all hoity toity, but standards matter. They really matter. Having choices means holding vendors to account. If you’ve ever worked in a company that uses EIGRP routing, you’ll realise what a bad decision that was.

So, time to ask your Cisco Account Manager when they are going to support TRILL and 802.1aq SPB. In fact, demand a response in writing. It doesn’t have to have a date on it just a promise that it will be supported on Nexus products in the future.

Accept no substitutes.

Postscript

I’m told that the current silicon from Cisco will be compatible with TRILL. Reading between the lines, this suggests that Cisco might be planning TRILL in the near future. But I sure as heck don’t want to end with PVST migration to MST migrations all over again for FabricPath to TRILL.

I say, Death to FabricPath. Hello TRILL. Ask your account manager when it will be delivered, keep up the pressure.

In the meantime, don’t deploy FabricPath unless you must. And you would have to be desperate to dead end your design this way.

Postscript Postscript

While its true that FabricPath is shipping today, and you might possibly need the extra bandwidth that FabricPath provides, I think that Cisco should be shipping 40GbE and 100GbE by now so that we could hold on a bit longer.

I still think you would have to be nuts to deploy FabricPath in a data centre and take the risk of migrating to TRILL in the future knowing that will cause numerous outages during the migration. TRILL and FabricPath are highly unlikely to be interoperable. They look similar in a number of ways.

  • Glenn Flint

    Greg, The L2 ECMP concept originally came from Cabletron in 1995. Securefast product using OSPF at L2. This was very successful and is still used today. An RFC was also posted later to try and make it an open standard, but at the time wirespeed routing was the future? It looks like it has taken 15 years to catch on and finally get rid of STP?

  • Pingback: 7 Trends of Data Center Network « Pronto Systems