Letter to 20 years ago (06 Sep 2020)

I noticed that I have not posted anything here in 2020! There's a bunch of reasons for this: the work I'm doing at the moment does not lend itself so well to blog posts, and life intervenes, leaving less time for personal projects. But in order to head off the risk that I'll post nothing at all this year I pulled something from one of my notebooks. 2020 is a round number so I decided to do some reflection and this was a letter that I imagined writing to myself 20 years ago. It is very much a letter to me! The topics are going to be quite specific and if you weren't paying attention to the computing industry in the year 2000 I’m not sure how much of it will make sense. But these are the points that I think me of 20 years ago would have wondered about.

You must be thinking that computers will be crazy fast by now. Yes…ish. It's complicated, and that's going to be a theme here. You've been hearing from Intel that the NetBurst chips will hit 10GHz in a few years, so with another few doublings what will have by now? 50GHz? Actually common values are around 3.5GHz. Some reach 5GHz, but only in bursts. Intel never hit 10GHz and nor did anybody else. It’s better than it sounds: instructions per clock are up a lot, so each cycle is worth more. (Although maybe we'll get to some of the issues that caused!) More importantly, all systems are multiprocessor now. It's physically a single chip, but inside is often 8- to 32-way SMT. Yep, that's cool. And yep, it only helps for certain sorts of workloads. Multithreaded programming is not going away.

Memory? 10s of gigabytes is common. Hard drives? It's nearly all flash now. You can still buy hard drives and they’re huge and cheap, but the speed of flash is pretty sweet. Computers really are quite substantially faster — don't be too put off by the clock speeds.

Your day-to-day life is a bunch of xterms and a web browser. Nothing's changed; you are dramatically underestimating the importance of path dependence. Terminals are still emulating a fancy VT-100 and sometimes they get messed up and need a reset. No fixes there. It's still bash or zsh; nearly unchanged from your time. The kernel has been fixed a little: you can now get a handle to a process, so no more PID races. You can open a file relative to a directory descriptor and you can create an unlinked file in a directory and link it later. Yes it's good that these things are possible now, but it is not a fundamental change and it took a long time. Actually you know what? Windows grew a much smarter shell, leapfrogging Linux in several respects. They had hardly moved forward since DOS so it was easier there, perversely.

So innovation must have happened higher up where there was new ground and little legacy, right? What about the semantic web? How did that turn out? Not well. We don't have lots of data in machine-readable formats and fancy GUIs so that anyone can create automation. Information is somewhere between impossible and a huge pain to access. You’ve read The Dilbert Future by now and its ‘confusopoly’ concept is much closer to the mark. The Semantic Web stuff failed so badly that nobody even tries any longer. (I'm afraid Scott Adams won’t seem so wholesome in the future either.) The closest you'll get is that your web browser can fill out your name, address, and credit card details. And it has to work really hard to do that because there’s almost no assistance from web pages. Go find Weaving the Web and throw it away.

Something more positive: bandwidth! You are using a dial-up that tops out at 5 KB/s and charges by the minute. You use a local proxy that keeps a copy of everything so that viewed pages are available offline and it lets you mark missing pages for batch fetching to reduce the cost. This problem is now solved. You can assume that any house in a city can get an always-on, several 10s of Mb/s connection. It's not as cheap as it could be but it's a standard household expense now. (Note: past me doesn't live in the US! —agl.) Also, everyone carries an impossibly fancy PDA that has that level of connection wirelessly and everywhere. I don't need to equivocate here, connectivity is solved in the sorts of places you're likely to live. But … there's a second edge to that sword. This connectivity can be a bit … much? There are some advantages to having the internet be stationary, metered, and behind 30 seconds of banshee wailing and static. Imagine your whole social life getting run through IRC, and that you're always connected. It’s tough to explain but there's a problem. But these PDAs? They have GPS and maps. Nobody gets lost anymore. Nobody carries paper street maps in their car. Connectivity can be pretty sweet.

This next bit is going to upset you a little: the whole Palladium / trusted boot stuff never took off on the desktop, but these PDAs are pretty locked down. One type of them is completely locked down and you can’t run non-approved software. The other will make you jump through hoops and, even then, you can't access the data of other programs. On the latter sort you can install a completely custom OS most of the time, but there's attestation and some things won't cooperate. This is still playing out and people are fighting over the details (because of money, of course). It remains a concern, but you underestimate the benefits of this sort of system. Your idea that people should own their own computers because they’re critical tools isn't wrong, but it is elitist. For the vast majority of people, their desktops degrade into a fragile truce with a whole ecosystem of malware and near-malware. Maybe it's their “fault” for having installed it, but these PDAs are so popular, in part, because they're hard to screw up. Bad stuff does get through the approval process, but it cannot mess things up to the wipe-and-reinstall level that desktops reach. The jury is still out about whether we will regret this, but you're wrong about the viability of giving people Windows XP and getting a good result.

Back on a positive note: the music industry switched to a $15 a month stream-whatever-you-want model and it works fine. You were completely right about this. Music still exists and it still pays a few at the top large sums and the rest very little. The music industry itself didn't sort this out though, other companies did it for them. What you're missing is that you’re not taking things far enough: companies also did this for TV and (many) movies. There are still rips of this stuff on BitTorrent, but it's not a live issue because people pay the subscription for the ease, by and large. In fact, access to scientific papers is a hotter issue now!

Basically, rates of change are really uneven.