NAT, ATM, decentralized search – and other outrageous opinions from the 1990s

The predictions that turned out correct ... and the ones we wished were right

Systems Approach The end of the year is often a time for people in tech to make predictions, but rather than making our own, today we’ll look back on some of the bold predictions of the past – specifically the inaugural Outrageous Opinion session held at SIGCOMM in 1995.

One of my most lasting contributions to the SIGCOMM community would have to be my 2003 Outrageous Opinion talk: MPLS Considered Helpful. Twenty years later I still run into people who remember the punchline, “I’m not bitter.” I suspect this will be remembered much longer than my four-year tenure as SIGCOMM Chair.

A less well-known fact is that I chaired the first SIGCOMM Outrageous Opinion session at SIGCOMM 1995. (Another fun fact from that conference: Marc Andreessen was scheduled to give a tutorial at the same time as me, but he bowed out at the last minute, citing business pressures – the Netscape IPO had taken place two weeks earlier. I picked up a bunch of disappointed attendees at my tutorial on smartNICs as a result.)

While outrageous opinion sessions subsequently took a turn towards stand-up comedy, in that first year there were a bunch of quite serious (and nevertheless funny) talks that have stuck with me.

I often refer back to David Clark’s talk, We Should All Become Economists. Already well established as “the architect of the internet,” David had started to branch out into areas of economics and policy, with notable works including Realizing the Information Future, which made the case to a broad audience that the internet was the appropriate architecture for the Information Superhighway. This might seem an obvious truth today, but at the time there was plenty of support for alternative architectures based around evolved versions of either the telephone network (think broadband ISDN) or the cable TV network.

His later paper [PDF] with Marjory Blumenthal – Rethinking the Design of the Internet: End-to-End Arguments vs. the Brave New World – is a wonderful examination of the tension between the idealized architecture of the internet and the commercial pressures that came to bear on it once it became a mainstream communications platform. While there are few things more annoying than tech people discussing economics from a position of ignorance, David was making the case that we should do more to educate ourselves about economics, and I wish more of us had taken his advice.

Equally memorable was a talk by Paul Francis, another internet pioneer whose influence spans topics as diverse as scalable multicast, IPv6, and distributed hash tables.

At that point we were in the early days of internet search – Google was still three years away from being founded, and AltaVista was an internal project to index the entire web at DEC (released to the world later that year). Paul approached me with two topics he was interested in presenting: his latest research on scalable internet search, which he had named Ingrid [PDF], and Network Address Translation (NAT).

I was rather more favorable towards the latter, because it was a very hot topic at the time, Paul was credited with its invention, and he had an amusing way to present his argument. Creative person that he is, Paul found a way to weave the two topics together in one talk.

The gist of the NAT part of his talk was that the prevailing IETF view on NAT at the time – don’t do NAT – should be viewed as analogous to abstinence-only sex education. No matter how much the people giving the advice might believe in the correctness of their position, they were not going to have much impact on the outcome. In hindsight, he was obviously correct: most of the world’s internet users today sit behind one or more NAT devices, but his position was much more controversial at the time.

Most of the world’s internet users today sit behind one or more NAT devices, but his position was much more controversial at the time

In the end, living with NAT became more important than preventing it, and today there is a whole body of work on NAT traversal that allows us to deal with it fairly painlessly.

I have had reason to refer back to the other part of Paul’s talk more often, because of the way it resonates today as we deal with the centralization of the internet and the recent efforts to re-decentralize it.

While early search engines were centralized – for example, AltaVista was apparently developed to show off the capabilities of a powerful DEC database server – Paul argued that a decentralized approach to search was going to be necessary as the web took off. (Recall this was the same year as the Netcape IPO.)

Again, he was right, but Google would eventually deliver a hugely successful distributed system to index and search the web, and put it behind a logically centralized front end. So while the technical solution was indeed decentralized, the user experience is centralized: just go to and ask your question. And of course that is where the web mostly sits today, 25 years later: distributed systems do the work under the covers but the average user interacts with a handful of centralized entities, such as the social media titans and streaming services.

I’m cautiously optimistic that we are seeing a reversal of this trend, especially with federated social media, but I do look back nostalgically on the era when it seemed possible that search itself might be decentralized. Since it is now increasingly apparent that search is getting worse, we can still hope. Also, the idea that decentralized technologies alone do not protect us from the perils of centralized control of technology (eg, the ownership of search or social media by a small number of companies) is something that I developed further in my Sixty Years of Networking talk earlier this year.

I can remember that there were multiple talks about the relative merits of ATM and IP. This seems hard to fathom today

Finally, I can remember that there were multiple talks about the relative merits of ATM and IP. This seems hard to fathom today, where the internet reaches something like half the world’s population and ATM is little more than a historical footnote in the large collection of OSI layer-two technologies that IP has accommodated. As my comments above suggest, it was far from clear in 1995 that this is how things would play out.

I was placing an each-way bet at this point, having worked on ATM at Bellcore (owned by telcos) but believing by 1995 that ATM would be adopted as a substrate for part of the Internet, not a standalone networking technology that would replace IP. Indeed, it was my focus on IP-over-ATM that provided me with the opportunity to join Cisco later that year as they increased their investment in ATM switching. That would ultimately land me in the team that developed MPLS – another technology that, like NAT, faced a fair amount of opposition at the IETF, but is widely deployed today.

The fact that I remember so much of that one evening in 1995 (whereas I remember almost nothing of the tutorial I gave) has a lot to do with how provocative, forward-looking, and accurate many of the opinions were. I think some of the talks in later years may have been funnier, but the sheer amount of predictive accuracy of those I remember from 1995 is striking. (No doubt I’ve forgotten some talks that were totally off base.)

David Clark’s comments about economists continue to resonate as we argue about network neutrality and hear demands from telcos that they be compensated not only by their customers but by the content providers. The increased centralization of the internet is also partly an economic (winner-takes-all) phenomenon. In the current debates about the future of the SIGCOMM conference, and more broadly as we look to shape the internet of the future, I hope we keep a place for outrageous opinions. ®

Larry Peterson and Bruce Davie are the authors of Computer Networks: A Systems Approach and the related Systems Approach series of books. All their content is open source and available on GitHub. You can find them on Mastodon, their writings on Substack, and past The Register columns here.

More about


Send us news

Other stories you might like