[Editor’s note: This is part of a series examining the internet’s first 50 years and predicting the next half century. Join Xconomy and World Frontiers Forum on July 16 for Net@50, an event exploring the internet’s past and future.]
Bob Metcalfe co-invented Ethernet, the communications standard still used for most local on-site networking, but Radia Perlman made it work. Her creation of the Spanning Tree Protocol while working for Digital Equipment Corporation in 1983 made it possible to link individual Ethernet networks into a vast interconnected system—that is, the thing we now call the internet.
After Digital, Perlman worked for Sun Microsystems and Dell EMC, where she’s currently employed. She’s written fundamental textbooks on networking, including Interconnections in 1992. Along the way, Perlman landed the honorific “the mother of the internet,” in recognition of her work on network routing and bridging, but she dislikes the title.
I reached Perlman in Washington state while researching my feature article “Special Report 2069: Predicting the Internet’s Next 50 Years.” Here’s an edited version of our conversation.
Xconomy: When you consider what the internet might look like in 2069, you have a chance to think about the underlying technology, the big forces at work, and which trends we can expect to continue in the future. For example, do you expect the internet’s underlying protocols to have changed by then?
Radia Perlman: This obsession with TCP and IP and these little components—it’s sort of like talking about, if we want to build a civilization should we base it on Italian, Greek, German, or English? It doesn’t matter. It’s like saying, we have wars and stuff, so what if we switch to Italian? Would that help? The bigger issues have nothing to do with the exact format of the underlying things.
X: I get what you’re saying: Changing to Italian wouldn’t change human nature. But once you have a language like English in place, or a protocol like TCP/IP, and it becomes a global standard, you might start to notice ways in which it could be improved, but the fact that it’s so embedded and it has so much momentum makes it horribly tricky to actually do that.
RP: Absolutely. I’m sure we could come up with a better language than English. English is a horrible language, if you look at spelling and pronunciation. But we can do anything we need to do with it. And it can change in small ways. Every year, we add new words, we remove words, we change the grammar rules, and stuff like that a little bit.
One of the things that people think about is: Is there a whole new way of thinking about how to build a network, other than sort of this layered thing where there is this underlying infrastructure that knows how to signal a bit on the wire—which is kind of that Layer 1—and then Layer 2 says, “How can I send the whole message to my neighbor?” And Layer 3 finds the whole path.
I think that basic premise is flexible enough that it’s unlikely the internet will ever change. Nor do I think it would benefit from doing anything other than having a low level that just moves data around and higher layers that do the authentication and stuff like that.
Now, just as I’m listening to those words coming out of my mouth, there are other issues like denial of service that people didn’t foresee. We have all of these incredibly buggy end nodes, and we’re asking humans to do absurd things like “Don’t click on suspicious links,” whatever that means. We’re blaming users when their machines get infected, which I’m appalled at. It’s not their fault. But given that it is possible to infect so many machines, it could be that the very simple notion of “Just let the underlying layers move the data around” might not actually be sufficient. Because it could be that sufficient numbers of bots controlled by bad guys can completely starve out what’s going on.
Other than those sorts of things, I think that the simple things they’re doing today are just fine.
X: Well, let’s go at this from a couple of other angles. You started programming in the ’70s, and you’ve been involved in this field ever since. And I’m wondering if I traveled back in time and interviewed you in, say, 1983, would you believe me if I told you that basically the same internet that was coming together then would still be around in 2019? Did you expect that the protocols would be this long-lived?
RP: I had no idea how long things would last. So, for instance, Spanning Tree Protocol. It was a hack that I thought would live for, like, six months. The reason for Spanning Tree was that when Ethernet came out, it was really just intended to be something where a bunch of nodes, like maybe 100 or 200 within a building, would all share the same wire, so that if anybody talked everyone else would hear them. It was a nice, simple design. And the cleverness about it was how do you manage to make it so that only one guy is talking at once, without having a leader that calls on people and gives them permission to speak. So, that was what Ethernet was.
What I was doing was Layer 3, which was the thing that hooks a bunch of links together and finds a path across the network. Ethernet came along, and I said, “Oh, here’s a new type of link, and I have to modify how Layer 3 works to be more efficient with this new kind of link that has hundreds of neighbors.” I had to do certain things to the routing protocol to make it more efficient because this new kind of link had different properties. Fine. I did that. But because Ethernet was bright and shiny and widely hyped, people thought that was the new way of doing networking. They built applications leaving out my stuff, the Layer 3 stuff, and directly worked on Ethernet. That’s fine if your application is only going to work on a single Ethernet. But the Layer 3 routers that move data from link to link wouldn’t work unless you cooperate with them and implement, for instance, IP, and that’s going to be your Layer 3 protocol.
I was trying to say, “Build your applications on top of Layer 3, and then you can have Ethernet be your link,” but people ignored me, as they always do. A few years later, in 1983, my manager said, “Hey, people want to have their applications work across networks, from one net to another.” The right way to do that was to have the computers at the end nodes put in Layer 3, but that was going to be a lot of work. As a quick fix, I designed this magic box that would move Ethernet packets around, even though the end node thinks it’s just speaking on a single Ethernet. That design required there to be no alternate path. There had to be exactly one way to get from one place to another. That’s what the Spanning Tree was.
I expected this to last for about six months until we had a chance to upgrade all of the end nodes to put in Layer 3. But within a year of that the original design of Ethernet didn’t exist anymore. It’s all the thing I had designed, basically, which was moving Ethernet packets around with Spanning Tree constraining the current routes being used for data.
X: This hack that you envisioned might last six months is actually still in place 36 years later. It’s probably because it actually works. If it ain’t broke, don’t fix it, right? But I think we have this assumption that technology moves at a certain pace and that things will be improved and replaced over time. Perhaps our expectations are wrong. Maybe fundamental technologies actually wind up staying around a lot longer than we’ve come to expect.
RP: Right. But also, when does it become a new protocol? For instance, are we still using Ethernet? People think Ethernet is a great success, but it has nothing to do with what was originally designed; it just has the same name. It has the same packet format. But the real cleverness was this contention protocol for sharing a link.
Now, that’s another interesting thing. If we’re looking backwards rather than forwards, the way networking is taught—and this is very politically incorrect of me to say, but I will—it’s taught as if TCP/IP arrived on tablets from the sky in its awesome perfection. Students are supposed to memorize, “This is the way networks work,” so that graduates of the college can immediately start configuring Cisco routers, but without any sort of critical thinking about it.
As it turns out, there were competitors to IP, and a lot of them were better. In particular,
