FCoE and the Return of Spanning Tree

One of the most amusing parts about Fibre Channel over Ethernet(FCoE) is that Spanning Tree is making a triumphant comeback. And I am talking a Roman style parade after the gates to the city have been built and lined with gold.

The IEEE is working on a standard 802.1AQ named Shortest Path Bridging and described as an ‘incremental’ advance on Multiple Spanning Tree Protocol. Well, if you do a bit of reading, ‘incremental’ means it uses IS-IS to locate Layer 2 paths through the network using CLNS (yes, those of us old enough to remember GOSIP and OSI protocols are having conniptions….).

I love the metaphor that ‘incremental’ means adding a fully fledged routing protocol supporting spanning tree to determine layer 2 paths through the network.

Oh how we laughed.

IS-IS is a good protocol, and runs some of the largest Service Provider backbones in the world. There have been many improvements in the last five years that bring IS-IS into mainstream. The idea of extending CLNS to carry Layer 2 path information is conceptually a great idea.

But very very hard to sell.

Oh how I laughed.

You see, I had a meeting recently with the storage people and they instantly had apoplexy. “My goodness, Layer 2 spanning tree!” We cannot allow that.

Of course, they seem to have forgotten that FC is a Layer 2 protocol, and that is why it doesn’t scale….but enough about that for now.

And then the Managers got wind of this. If there is anything that strikes fear into the the heart of a Data Centre manager, Spanning Tree is it. It seems to me that the least seven or eight years has seen constant innovation to remove Spanning Tree. And now I am proposing bringing it back !

Oh how we laughed.

The standards track from the IEEE is continuing on, and they expect the 802.1Qbb, 802.1Qau, 802.1Qaz and 802.1aq standards to be complete sometime in 2010. What ? 2010 for the standards only !

Oh how we laughed.

There is an awful lot of noise about FCoE, but it really faces some huge challenges before its is going to get onto many roadmaps. Even though Cisco is pushing it along like Drott-D9 bulldozer and making enormous promises, I can’t get my fellow workers to stop laughing.

Oh dear, how we laughed.

  • Dmitri

    Greg, if I’m not mistaken, FC had always had a routing protocol built into it, FSPF.

    Also, your statement about 7 years of innovation to remove spanning tree puzzled me. I admit I am not so close to the happenings in the DC arena, but I have been doing Carrier Ethernet for these last 7 years and have not noticed much in the department of removing STP. Yes, where possible, LAG/LACP is used, but what else? (I’m not talking about PBB or MPLS, as I would have thought they are not directly usable for the DC internal LANs).

  • http://blog.ioshints.info Ivan Pepelnjak

    You forgot to mention the good old days (around 15 years ago) when someone had a great idea to run SPF on MAC addresses and call the chimera “brouter”. Those were the days when a “switch” was not a networking term, “bridge” was a four-letter-word and “router” was the hero-of-the-day, so those that were not competent enough to program a router tried to repaint a bridge.

    It’s so funny to see the same idea reappear, this time using IS-IS as the SPF (obviously we’ve progressed a long way standard-wise in the last 15 years :).

  • Dirk-Jan van Helmond

    I believe that FCoE has a future in the datacenter, but I am in no way an FCoE evangelist.

    And I think you are misconstruing things. Both STP and SPF algorithms are used to create a loop-free topology. The difference between these is that STP and deriviates create a single (or in case of MSTP of PVST some) spanning-trees that all nodes must abide by. This does not ensure the best path for every node in the tree. SFP creates a shortest path tree per node, which DOES ensure a best path for every node.

    Besides that, path learning for 802.1d is done by flooding of frames and closing interfaces, which in no way scales, while 802.1AQ used link-state updates, that does scale and makes Layer 2 multipathing possible.

    • http://etherealmind.com Greg Ferro

      Does this mean that L2 SPF will replace STP then ? Have I misunderstood the fundamental purpose of 8021.AQ ?

      Also, how does IS-IS converge in milliseconds ? Won’t that be a requirement in DCB ?

  • Dirk-Jan van Helmond

    By the way, we are on the brink of completely getting rid of spanning-tree.

    Nortel ofcourse has SMLT and are developing hard on PLSB, which is a combination of mac-in-mac and ‘l2-isis’. L2 isis is tightly coupled to L2 multipathing. Woven is releasing L2MP products, Cisco has VSS and vPC, which is a first step to L2MP, and that is roadmapped for the Nexus platform.

    Therefore I do believe that l2-isis or L2MP or 802.1AQ is the way to go.

  • http://nohype.tumblr.com Stuart Miniman

    You focused on the IEEE L2MP activity, but what about IETF TRILL? Radia Perlman, the inventor of spanning tree is an active participant in these updates, surely you don’t think that Ethernet as deployed today is perfect?

    • http://etherealmind.com Greg Ferro

      My case would be that Ethernet is completely imperfect and no amount of fudging would fix it. Ethernet is the lowest and least LAN technology created in the last thirty years.

      However, we, the world, chose it over others. Perpetuating it is a disappointment, and I would prefer that IP was used for everything and spanning tree would not be required. That is, Ethernet is a locally significant technology only (see IPv6 local scope).

      It is not suitable for building a data centre fabric.

  • Bob

    Your comment “You see, I had a meet≠ing recently with the stor≠age people and they instantly had apo≠plexy. ìMy good≠ness, Layer 2 span≠ning tree!î We can≠not allow that.

    Of course, they seem to have for≠got≠ten that FC is a Layer 2 pro≠tocol, and that is why it doesnít scale.Öbut enough about that for now.” is way off the mark. You seem to be hinting or insinuating that the storage guys are unaware of the fact that FC is an L2 protocol. They (we) are well aware of that fact, but we also don’t suffer from loops the way Ethernet does. All L2 protocols are not created equal. Just different. FC doesn’t loop. FC does multiple paths. FC scales, just differently than Ethernet.