Wednesday, August 15, 2007

What's old is new again

While looking for some info on vi, I found the following interview with Bill Joy, from around 1984:

REVIEW: You mention everything but disks.
JOY: You might want to page over satellite telephone... Page fault, and the computer makes a phone call. Direct broadcast or audio disk - that's the technology to do that. It's half a gigabyte - and you get 100 kilobyte data rate or a megabyte or something. I don't remember. You can then carry around with you all the software you need. You can get random data through some communications link. It is very like Dick Tracy. Have you seen these digital pagers? You can really communicate digital information on a portable.

I don't think you need to have a disk with you. There are so many people who believe that you need to have a disk that you'll be able to have one because they'll make it cheap. That's the way things work. It's not what's possible but what people believe is possible...

What's particularly striking about this quote is that he was extraordinarily prescient and wayyyy off simultaneously.

He was talking about how people would use computing devices in the future, and his idea was that instead of mass storage in a small format, we'd have thin clients where even the operating system was served from a remote machine, very much like the Unix consoles he was working on at the time. (at 1200 baud, apparently)

So let's look at what happened. Just as he predicted, hard drives have gotten cheaper and smaller and now everyone has a portable one (iPod). However, the network latency and throughput has not yet caught up to to hard drive speed to where it's feasible to do all of your computing over a network link. You still need an operating system. But he knew that would probably be the case, because so many people wanted it to be like that.

So it seems like he was right on, except that it's obvious that he was thinking that big servers and thin clients would continue to exist. Even though he talked about Apple's new Mac in the interview, I don't think he predicted the huge increase in the performance of cheap hardware.

It's easy to see why. A million times faster is an easy concept to understand. If I told you computers will be a million times faster in 20 years, you'd shrug and say, "sure, that sounds about right." But what neither of us can envision is what people will actually do with that horsepower.

I guarantee you that if you went back 20 years and told people that everyone will own a multi-core CPU that runs at 3GHz, and that with all that horsepower, the most popular application development environment is an interpreted scripting language that makes asynchronous network calls to grab data and display it on a tree-based document display program that runs on top of an graphical windowing system (which, coincidentally, already has all of the widgets to do all of those things the browser does), which is implemented in multiple layers that span multiple processes that all run on top of the operating system, they'd look at you like you were crazy. And yet that's exactly what we're doing with Javascript, which in effect, is a way to make today's lightning fast hardware behave just like the network computer Bill Joy described.

He was exactly right, but he never predicted there would be five or six layers in between pages of RAM and the user's data.

Here's the big question: if you were develop a system to do what web browsers and Javascript do from scratch, would it look anything like a web browser and Javascript? My guess is no. I'm not sure if that would be a good thing or not.

So. . .here's my prediction, taking into account the pace of hardware development and the history of software development. In 20 years, cheap hardware will be ridiculously fast, but it will still look very much like Intel hardware today. We'll have many, many CPU cores to work with, but nobody will use a parallel programming language designed to take advantage of multiple cores. Instead, virtualization (i.e. VMware, Xen) will be integreted into the operating system, and each process will run on its own virtual machine.

Each web browser will have been expanded into a full blown widget toolkit and have merged in something that looks like Flash, but there will still be multiple incompatible browsers. The latest craze will be a browser compatibility layer written in a programming language that compiles to "raw Javascript", and it will reduce the performance of applications but allow you to use them anywhere.

People will set about re-writing a version of Photoshop in the new compatibility layer, and everyone will wonder why they'd do that, when the current version of Photoshop runs in Internet Explorer just fine.