January 17, 2020

Software inside

Hedy_Lamarr_in_The_Conspirators_2.jpgIn 2011, Netscape creator-turned-venture capitalists Marc Andreesen argued that software is eating the world. Andreesen focused on a rather narrow meaning of "world" - financial value. Amazon ate Borders' lunch; software fuels the success of Wal-Mart, Fedex, airlines, and financial services. Like that.

There is, however, a more interesting sense in which software is eating the world, and that's its takeover of what we think of as "hardware". A friend tells me, for example, that part of the pleasure he gets from driving a Tesla is that its periodic software updates keep the car feeling new, so he never looks enviously at the features on later models. Still, these updates do at least sound like traditional software. The last update of 2019, for example, included improved driver visualization, a "Camp Mode" to make the car more comfortable to spend the night in, and other interface improvements. I assume something as ordinarily useful as map updates is too trivial to mention.

Even though this means a car is now really a fancy interconnected series of dozens of computer networks whose output happens to be making a large, heavy object move on wheels. I don't have trouble grasping the whole thing, not really. It's a control system.

Much more confounding was the time, in late 1993. when I visited Demon Internet, then a startup founded to offer Internet access to UK consumers. Like quite a few others, I was having trouble getting connected via the Demon's adapted version of KA9Q, connection software written for packet radio. This was my first puzzlement: how could software for "packet radio" (whatever that was) do anything on a computer? That was nothing to my confusion when Demon staffer Mark Turner explained to me that the computer could parse the stream of information coming into it and direct the results to different applications simultaneously. At that point, I'd only ever used online services where you could only do one thing at a time, just as you could only make one phone call at a time. I remember finding the idea of one data stream servicing many applications at once really difficult to grasp. How did it know what went where?

That is software, and it's what happened in the shift from legacy phone networks' circuit switching to Internet-style packet switching.

I had a similar moment of surreality when first told about software-defined radio. A radio was a *thing*. How could it be software? By then I knew about spread spectrum, invented by the actress Hedy Lamarr and pianist George Antheil to protect wartime military conversations from eavesdropping, so it shouldn't have seemed as weird as it did.

And so to this week, when, at the first PhD Cyber Security Winter School, I discovered programmable - - that is, software-defined - networks. Of course networks are controlled by software already, but at the physical layer it's cables, switches, and routers. If one of those specialized devices needs to be reconfigured you have to do it locally, device by device. Now, the idea is more generic hardware that can be reprogrammed on the fly, enabling remote - and more centralized and larger-scale - control. Security people like the idea that a network can both spot and harden itself against malicious traffic much faster. I can't help being suspicious that this new world will help attackers, too, first by providing a central target to attack, and second because it will be vastly more complex. Authentication and encryption will be crucial in an environment where a malformed or malicious data packet doesn't just pose a threat to the end user who receives it but can reprogram the network. Helpfully, the NSA has thought about this in more depth and greater detail. They do see centralization as a risk, and recommend a series of measures for protecting the controller; they also highlight the problems increased complexity brings.

As the workshop leader said, this is enough of a trend for Cisco, and Intel to embrace it; six months ago, Intel paid $5 billion for Barefoot Networks, the creator of P4, the language I saw demonstrated for programming these things.

At this point I began wondering if this doesn't up-end the entire design philosophy of the Internet, which was to push all the intelligence out to the edges, The beginnings of this new paradigm, active networking, appeared around the early 2000s. The computer science literature - for example, Activating Networks (PDF), by Jonathan M. Smith, Kenneth L. Calvert, Sandra L. Murphy, Hilarie K. Orman, and Larry L. Peterson, and Active Networking: One View of the Past, Present, and Future (PDF), by Smith and Scott M. Nettles - plots out the problems of security and complexity in detail, and considers the Internet and interoperability issues. The Road to SDN: An Intellectual History of Programmable Networks, by Nick Feamster, Jennifer Rexford, and Ellen Zegura, recapitulates the history to date.

My real question, however, is one I suspect has received less consideration: will these software-defined networks make surveillance and censorship easier or harder? Will they have an effect on the accessibility of Internet freedoms? Are there design considerations we should know about? These seem like reasonable questions to ask as this future hurtles toward us.

Illustrations: Hedy Lamarr, in The Conspirators, 1944..

Wendy M. Grossman is the 2013 winner of the Enigma Award. Her Web site has an extensive archive of her books, articles, and music, and an archive of earlier columns in this series. Stories about the border wars between cyberspace and real life are posted occasionally during the week at the net.wars Pinboard - or follow on Twitter.