Avoiding Death By A Thousand Cuts
Mark Hoover, President, Acuitive, Inc.

In the July, 1999 edition of Business Communication Review, David Passmore authored an article entitled "Why ATM Must Die". http://www.netreference.com/PublishedArchive/Articles/ArticleDirectory.html

As with all Passmore writings, the article was well researched and the position well articulated. I happen to agree with nearly everything in the article. I'm not here to rebut the article, but to add a supplemental sidebar of "yes, but in some application areas it will be a long, slow, death."

In some areas the death has been quick. ATM-to-the-desktop was stillborn. ATM in the campus took root a few years ago as a substitute for FDDI, just as a way to get higher speed interconnections between collapsed backbone routers for pure data traffic. Many of these ATM backbones still exist, but they are going away one by one as enterprises reach the stage where they need to migrate to their next generation backbone. As Dave rightly states, most new backbones being put in are Layer 3 switch oriented. The cost/performance is just too compelling to ignore. It's hard to find an ATM bigot in corporate IT organizations any more. There is a lot of back-tracking and revisionist history going on. "Daddy, what did you do in the ATM vs. Router war?" "Ahem, I was on vacation that week." Or, "We lived for awhile in Canada during that war."

Where ATM has remained in place and in fact is growing, is in the traditional carriers. That's where you can still find proud ATM bigots as well as rational economic decision-makers.

Part of the reason ATM remains important with the carriers is mass times velocity. Energy is needed to deflect an object from its present course. Significant energy is needed if that object is massive, like a traditional carrier, even if it is slow moving. The course the carriers have been on for a number of years is ATM. Much of the ATM in carrier networks is in place simply to transport Frame Relay, but more and more native ATM services, voice trunking services, some voice switching services, and related Operational Support Systems have been put into place. After many years of planning, trialing, and training, there is finally a set of people out there who know how to market, sell, deliver, maintain, and operate ATM networks and the variety of services supported over them. For the large and incumbent carriers, that inertia will override the promise of newer and "better" technologies. It's the "turn the battleship" phenomena, only in this case, the orders being given to the helmsman are more like "Turn! No don't turn! It's just a little ice cap, let's ram it! Turn the other way! No, your other left hand! Ahh forget it - let's just merge with somebody. Where's the remote for the cable?"

In the meantime, despite the best efforts of IP technologists everywhere, ATM cells are seeping onto backbone networks like radon into a New Jersey basement. DSL services are just beginning to take off and most of them are based on ATM cells. New integrated access offerings from Accelerated Networks, CopperCom, JetStream, Mariposa, and others are based on an underlying cell infrastructure. MAN multi-service access offerings from Atmosphere Networks and others are cell-based. What is a carrier architect to do but accept these cells and incrementally grow the already established ATM network?

Of course we all know that the right thing for these architects to do is go into their bosses office and say, "Rather than adding a couple of more ATM switches to the network to handle the projected load for next quarter, I think we should rip the whole thing out and replace it with an IP-over-optics infrastructure. The investment will be only about 25% of the total expenses we've had over the past five years building the network we have now. And, as IP matures, we can probably achieve equivalence to many of the ATM features we leverage in our present network. And finally, if we work really hard, we can migrate our Operational Support System over to that new infrastructure in a couple of years". Of course, the architect should only do this if she has her resume updated or, even better, has a job offer from a core IP router company in hand.

In any reasonable planning window, incrementally adding to the ATM network will always be more cost effective than a massive swap to a different technology, vendor base, management system, and operational way of life. Keeping the ATM network intact does not inhibit delivering IP-oriented services. IP services can generally be supported over the ATM infrastructure and in a sense enhance that architecture by reducing the need for PVC configuration and eliminating the need for SVCs altogether. There is some cost inefficiency and there is some question about whether QoS and security mechanisms independently initiated with-in the different layers will complement one another or offset one another. But you can basically make IP services work over ATM. So the ATM-ites get ATM-er. And every new switch added increases the cost of rolling back the ATM network at some future time.

Nonetheless, it's clear to me that IP-over-optics is the future. I do believe that the equipment costs are ¼ of the equipment costs of IP over ATM over Sonet over Optics. And I do believe that sooner or later IP and MPLS will replicate all of the services and functions of ATM. And I do believe that IP over ATM will be a difficult environment to define and maintain QoS functions in, due to decisions being made at two different levels in the network with minimal or no coordination between the levels. But initially the approach will gain a toehold only in greenfield applications. For existing networks, incremental growth will always be less expensive than switching over to the new technology, as long as the existing network can be scaled to meet the demand.

With the greenfields as a starting point, ATM in the carrier infrastructure will start to die a death of 1,000 cuts. IP equipment price/performance will be 25% or less of ATM/Sonet equipment. Core IP routers will be faster and cheaper than core ATM switches. MPLS will replicate and enhance some of the path control functions of ATM. IP will prove to be much easier to provision for mesh private networks, be they presented to the network as Frame Relay, ATM, or native IP. IP QoS will meet or exceed the capabilities of ATM QoS. IP OoS provides much more granular application-level traffic classification, which can only be obtained in the ATM world by those who can afford the luxury of an ATM circuit per application. The IP backbone network will be able to support circuit emulation, CBR, and VoIP as well as ATM has supported the equivalent in the past. Integrated access using pure IP will prevail. IP-level authentication and encryption will become standardized and ubiquitous. Private network and internet access can be more easily merged.

In the past people have talked about the ATM "cell tax," but what will really be its eventual undoing is the "non-ubiquity tax." Since ATM does not exist natively on any end points (accept for a few ADSL end points), the problem always exists that somewhere a cost must be incurred to convert traffic to ATM, and no matter how good that ATM network is, it can't guarantee end-to-end service parameters. The only truly ubiquitous end-to-end technology is IP. As innovations pile up in IP functions and standards, and they become supported in a wide variety of networking and end point devices, they can potentially be taken advantage of on an end-to-end basis. While ATM can still be used as a basic transport functions with-in the carrier infrastructure, ultimately the cost of carrying a set of ATM features which have become either irrelevant or harmful to the operation of the same functions at the IP layer will spur carriers to start to downsize the ATM network and move more and more traffic over to the IP/Optics network.

At some point the greenfields will achieve a big competitive advantage over the incrementally grown ATM-based service infrastructures. This tension will cause many carrier architects' and marketeers' heads to explode. The thing that would change this situation would be a set of products that aid migration from today's 4-tier networks (IP/ATM/Sonet/optics) to tomorrow's 2-layer networks (IP/Optics). Such products would need to:

• Connect to both the existing ATM oriented backbone and a new parallel IP/Optics backbone.
  
• Allow traffic to be directed to the new backbone, the old one, or both. Enable "roll backs" in case expected innovations in the pure IP world don't quite work out and service levels need to be restored by leveraging the old tried and true backbone.
  
• Blur the distinction between ATM and IP by supporting MPLS and MPLS-based enhancements. I figure MPLS is a god technology because the ATM bigots point to it as the victory of ATM over IP and the IP enthusiasts point to it as the demise of ATM. Of course, as with all technologies whose usage is in the future rather than in the now, stay tuned for more information on robustness, inter-operability, scale-ability, manageability, freaky side effects, and The Next Big Thing which solves all those presently undefinable issues.
  
• Enable the Operational Support System to be agnostic to line, trunk, and backbone technologies. A service is a service, whether the bits traverse a cell-oriented portion of the network or not.
  
• Interwork key functions such as QoS, security, provisioning and troubleshooting so that the ATM functions complement the IP functions in the IP-over-ATM-over Sonet-over optics world, with the goal being that services are equivalent whether packets happen to travel over ATM or not.
  
• Support legacy services such as private lines, Frame Relay, and ATM over the pure IP backbone, with the goal being that service levels and features are maintained over the pure IP backbone. If this is achieved, then growth can be shifted to the new backbone, allowing investments in the old backbone to be capped, and allowing an immediate return on investment for the IP-over-optics backbone (rather than waiting for the demand for new IP-oriented services to ramp up).
  
• Add new service options for the legacy services (QoS policies, security policies, virtual networks, outsourced services) that perhaps previously had been thought of as unique to IP VPNs only.
  
  By integrating the worlds more closely, alleviating the either-or planning tension, carriers could afford to ramp up their investments in IP-over-optics much more rapidly but at the same time maintain and benefit from all the sunk cost investments in their ATM network. Similar to campus backbones discussed previously, the ATM networks can then be rolled back and retired either rapidly or slowly, depending on the best economic choice at every future planning point. That's a more humane demise for all concerned.

(volume 2, number 10)

 

Home

Clients

Services

Hoov's Musings

Research Reports

About Acuitive

Send email to info@acuitive.com with questions or comments about this web site.
Copyright ©1997-2001 Acuitive, Inc. All Rights Reserved