ATM: Another Technological Mirage

It is common knowledge that the Internet is in trouble. Stories about Internet traffic jams have become a staple of major media technology reports. The backbone Internet Service Providers (ISPs) are scrambling to upgrade the capacity of backbone links, which is straightforward enough. However, the major problem of limited capacity of switching equipment remains.

The conventional approach to native IP routing, as embodied in routers by Cisco Systems which presently dominate the Internet, is no longer adequate. Routing IP packets is a rather complicated process, requiring traversal of a tree-like routing table, processing IP options, and other relatively complicated actions. The reason for the complexity is that every IP datagram is routed separately. Implementation of complete IP routing in hardware at best can be called impractical.

Routing ATM cells, on the other hand, is very simple because all cells have a small fixed size and because they are not independent. Rather, virtual circuits are formed, and all cells sent by parties in a virtual circuit follow the same path. This means that intermediate ATM switches keep tables of virtual circuits, so routing of a cell is reduced to a simple retrieval from an array indexed by the virtual connection identifier. The simplicity allows ATM switching to be implemented completely in hardware, thus making ATM switches outperform existing IP routers by an order of magnitude.

However, ATM has fundamental problems that are not usually understood, or discussed, by anybody but a few computer scientists and backbone engineers. Those obscure problems are not evident in lab tests, benchmarks or "pilot projects"; they affect the quality known as scalability. In fact, most leading Internet backbone engineers agree that all-out ATM replacement for the Internet simply won't fly. In this article we will attempt to explain why, without going into excessive technicalities.

The following chapters explain the concepts and argue that native IP routing is the only currently existing technology that can hold a global data network together. Please note that this text discusses only the applicability of the different technologies for global communications; the suitability of technological solutions for private networks is certainly different. For example, it is our opinion that ATM can be a perfectly good technology for LANs and medium-size corporate backbones, providing that it can compete on a price-performance basis with other approaches, such as gigabit Ethernet.

1. Traffic Patterns: The Internet Is Entirely Unlike The Telephony.
2. Existing Internet And ATM "Clouds".
3. Route Flap Explained.
4. "Fat State" vs "Lean State" Packet Switching.
5. Exchange Points Are The Key.
6. Dynamic Rerouting.
7. Best-Effort Delivery vs Back-Pressure Flow Control.
8. Resource Reservation Considered Harmful.
9. Economic Considerations.
10. Conclusion.

We will appreciate your questions, critique and suggestions for improvement of this text. Please feel free to contact us by e-mail avg@pluris.com.

Copyright 1996 by Pluris, Inc. All rights reserved.

Cisco, the "bridge" logo, and Cisco Fusion are registered trademarks of Cisco Systems.
DEC and GigaSwitch are registered trademarks of Digital Equipment Corporation.