The Monster Memo
Quite a debate about MPLS has been raging on our message board since I wrote a little article about Light Reading’s August Research Poll last week (see Poll: Is MPLS BS?).
More than 100 messages have been posted after the article, and more than 1,000 people have now taken the poll to indicate where they stand on whether MPLS is just another big ball of hair in the making (see MPLS - Just Kidding?)
It’s time for me to come clean (and give the MPLS discussion pot a really good stir).
When I wrote the article last week, I was sitting on a dynamite message from Mike O’Dell, the outspoken former network architect of UUNet. And I must admit that O'Dell's forthright views about MPLS and standards bodies influenced the way in which I pitched the story.
In his memo, O’Dell slams not just MPLS, calling it a “Frankenstein monster,” but also the Internet Engineering Task Force (IETF), which he says “has largely become a negative force for progress." He also takes a dig at the believers of IP-everywhere networks, describing them as “the 21st Century version of the Flat Earth Society.”
O’Dell signed his message “UUNet Chief Scientist Emeritus & Resident Crank,” although he actually left UUnet early this year.
His memo starts out by claiming that he played an indirect role in the birth of MPLS by providing the incentive for Dr. Yakov Rekhter -- then at Cisco Systems Inc. (Nasdaq: CSCO) -- to invent Tag Switching. Tag Switching turned into Multiprotocol Label Switching (MPLS). O'Dell's memo turns into a very illuminating and amusing rant.
Here it is:
- Yakov's creation of Tag Switching was Cisco's response to my architecture for the UUNET DS3 backbone which used a layer of Cascade frame switches under the routers to provide a
transit fabric richer and more complex than could be managed
with pure IP-level routing. That's one of the principle bogosities
in MPLS - it uses IP-only routing technology to produce the
paths, and that's the part which is busted in the first place.
Note that MPLS itself - the forwarding mechanism - is an L2 technology, and the miscreants trying to build "MPLS VPNs" are basically doing a very lame job of simulating the Cascade frame relay technology.
Again, tag switching (renamed "MPLS" when the IETF decided it needed to mark its territory) was created as a way to try and replace the L2 switches used in UUNET's backbone. Later the frame switches were transparently replaced with ATM switches because for this purpose, ATM and Frame are isomorphic.
UUNET is now using MPLS as a transit technology simply because nobody is doing OC192 Frame Relay as such. The story of LDP vs RSVP signalling is pretty entertaining as well. LDP was created as a red herring, thrown to IETF to keep it busy gnawing on a harmless bone while the people who actually understood the (limited) utility of MPLS simply got on with using it.
The fact it's been turned into some kind of Frankenstein monster ("It's a dessert wax! No, it's a floor topping!") is somewhat entertaining if you like black comedy, but at the same time it's an amazing waste of effort.
More importantly, as long as I'm venting, the IETF has largely become a negative force for progress because it's busy promulgating all kinds of happy horseshit which has been created to salve the ego of the participants, not to solve a problem people really have. The result is that new routers must implement all kinds of crap which will never be debugged because it will never get used in any serious way. But it's still in there festering, reducing the reliability of the boxes and increasing the brittleness of the networks built with those boxes.
Of course the vendors like this virulent featuritis - it gives them new things to hawk. To me, though, it's like going to your doctor and he says, "The drug salesman was just here and he has this really amazing new drug, so I'd like you to contract this rare disease so I can cure you!"
The vendors are solving problems they want the customers to contract; they are not solving the problems customers have, and the IETF is an extremely willing participant in this.
Finally, I've been saying for many years that the "IP-ONLY" crowd is the 21st century version of the "Flat Earth Society". They insist that the networks have only an IP layer and a completely degenerate Layer 2 below that, and then wonder why trying to fix everything at Layer 3 turns out to be exponentially complex. This IP-only 3G thing is a great example of leading people down the primrose path, very possibly to destruction. Running an RF-based network requires a LOT of local knowledge ("local" in the sense of link-specific) and IP has proven itself incapable of the task more than once.
Applications don't mind that the IP packets got carried by a good L2 infrastructure (that's what switched ethernet is all about), but they do mind if an IP-only infrastructure does a bad job of delivering most of the packets in a timely manner with a low drop rate.
I'll shut up now, but I've grown very tired of seeing a simple idea get turned into an incredible mountain of bullshit. And don't get me started on "Diffserv" - another brilliant example of defeat snatched from the jaws of victory. And worse, the MPLS and diffserv people have gotten together to breed. Oy!
The best way for me to follow this is with an excerpt of another memo - one that I myself posted in the Article Chat following the orginal story on the poll:
- This seems to happen over and over again in technology. Something succeeds in solving a specific problem and then swarms of people descend on it, Life of Brian style, and exclaim it's the answer to everything.
A few years later, they find something new. By then, they've manage to make the previous technology turn into a big ball of hair, so they declare it a hopeless dud and move on.
Meanwhile, the folk with the original problem carry on finding the technology that meets their needs - which is why ATM is so widely deployed in carrier nets.
We wouldn't have much to write about on Light Reading if the industry wasn't so fickle. Coming to think of it, a lot of us wouldn't have jobs.