In late 1997, we started out with a ropey demo and about five customers. A grand total of something like 40 credit-card transactions went through the system in November '97 to January '98.

By March-April 98, if I remember correctly, we had a buggy production system, which had been rushed into service far too soon ... but which could deal with two banks rather than one, and supported a bunch of whizzy extra features (whenever I got time to implement them). Features like the ability to handle multiple merchants, each with multiple EPOS terminal IDs, and in some cases trading in currencies other than Sterling. One early roadblock was the discovery that the banks' payment services expected each EPOS terminal ID to be in use for only one transaction at a time; concurrency reared its ugly head very early, because what we were doing was processing transactions on behalf of client software somewhere on the web. We dealt with the concurrency headache by bolting on an extra layer of Virtual terminal IDs — a pool of terminal IDs would be managed by a scheduler — to hide the concurrency problem from the merchants. And we added transaction velocity checking, to prevent end-users mashing their thumbs repeatedly on the "Submit" button and pounding on their credit card. Along the way, we moved from using an ISDN dialup to talk to the banks (basically behaving like a very fast modem dialup) to an X.25 box on BT's commercial backbone; my server, running on a dedicated machine, would open socket connections to the PAD (as many as necessary) and throw transactions at the banks over the dedicated line.

Some time in this period I wrote what was basically a limited-functionality clone of expect in Perl. And then I wrote a server framework in Perl, NetServer::Generic, to make it easier to write pre-forked multiheaded server processes. (NetServer::Generic is on CPAN because I talked Dave, my boss, into agreeing that it wasn't part of our core business, but we could do with having lots of free beta testers to whack on it for bugs. It worked well; I still get badgered by the occasional bug report to this day, even though I stopped maintaining it in 2001.) And I wrote a whole bunch of other stuff; much of it over-hasty, or buggy, or both, because I was being ridden hard by feature creep from one end and maintenance headaches from the other. I was busy: the old truism (first expressed by Tim Berners-Lee, I think) that five years pass on the web for every year in the real world was creeping up on me.

(I gather that Datacash retired the last of the code I'd worked on some time in 2008. As I left the company in 2000, I contemplate this news with mixed emotions. I'm a better author of fiction than I ever was a programmer; and software is notoriously short-lived. I suppose the fact that any of my code survived my departure by more than a month, and that the company kept it running for another few years, speaks for itself.)

By the summer of 1997, we'd begun acquiring new customers. If we had five in December 1997, we had maybe fifty in December 1998. We were still losing money — charging £0.5 per transaction, you need a lot of transactions to pay for two small offices and four salaries and a couple of collocated servers. And I was still a contractor, paid by the day. But things were hotting up. And then they went nuclear.

1998 was the year that the dot-com bubble began to inflate for real. And it nearly killed us. At one point in September, I realized that our transaction throughput was growing at a compound 30% per month, and it kept that up for six months before it began to slacken off. Some time in early to mid 1998, we handled 100 transactions in a month. By the same time in 1999, we had many individual customers who put that volume of credit card transactions through in a day. By late 1999, we were looking at transactions hitting our servers every few seconds; around the time I left, in early 2000, it was looking like credit card transactions per second were the way to measure throughput. Whatever his other faults, Gavin was an excellent salesman. But at the back end, it kept me on my toes. The more your software gets worked, the more likely obscure bugs are to rise to the surface. Sam was having headaches too; his web-based reports were growing like crazy, and the number of customers who needed them was growing too.

I was stressed out for most of two years. I'm an alpha-type personality to begin with, but this wasn't funny. I was writing fiction (and articles for Computer Shopper) as a therapeutic distraction. Around the middle of 1998, I figured that the novel I'd written in 1995-96 Singularity Sky was about ready, and mailed it off in the direction of Tor in New York, where it sat on a certain editorial director's desk for the next eighteen months. I wrote and sold a couple of short stories, and began work on a project which I was workshopping with some other local writers; a strange humorous horror novel/spy thriller about a hapless geek who's fallen into a government department for dealing with ... look, you probably know where this is going, right?. This was strictly a weekend activity, to distract me from the weekday stress cycle: compartmentalising my life helped me deal with Datacash. But it probably didn't help enough.

For most of the end of 1998 and the first half of 1999, the uttermost bane of my life was an ecommerce subsidiary of Bank Paribas called KLELine. They were offering a credit card solution over SSL, which had certain attractions for some of our customers, being (a) French, and (b) able to do some funky and useful things, or so they said. The trouble from my point of view was ... well, they weren't terribly clear on open source, for starters, or on public APIs, which was somewhat more serious. And when I got in deeper, I discovered some horrifying shortcuts in their API. Like, oh, once a credit card transaction hit their servers they'd process it, but the acknowledgement might well disappear into the bit bucket if the poor-quality leased line between London and Paris chose that particular moment to crap out. And the exchange rate for the transaction in question would be pulled out of a hat in accordance with the phase of the moon or something, and a subsequent refund or cancellation request wouldn't go through at the same exchange rate if there was a currency fluctuation.

We were in early-to-mid 1999. We had about four banks up and running, and another three or four in the pipeline to hook up and get ourselves certified by. A couple more programmers had arrived, all (except for Sam and I) being smokers: I was coughing my lungs out by an open window, but still juggling the entire server development and debugging workload. Dave and Gavin had figured out that the one-room model wasn't working; later that summer we were going to move to a new purpose-built office suite with room for 20-40 people, in a refurbished 18th century warehouse in Leith. KLELine was giving me sleepless nights; insomnia, depression, stress. I'd written a wrapper for their broken DLL that worked, most of the time, and took appropriate defensive action when things went pear-shaped, but I had no confidence in them. And Dave's management style was to treat me as an interrupt-driven resource. "Shoot at the monkey's feet; make the monkey dance," as one of my colleagues put it. When you've got one overstressed programmer trying to hold together the mission-critical system the whole company runs on (a company which is now up to about twelve people and mushrooming rapidly) that's not a terribly smart way to manage them — but we were all learning on our feet: it was Dave's first management job, and he was learning by making a lot of mistakes.

One particularly bad Wednesday I came into work, did stuff until I redlined, then sat at my desk for an hour. I got up and went over to Dave's desk. "I'm taking tomorrow and Friday off work," I announced.

"But what about —" Dave looked up and saw my face. "No problem," he said hastily.

I emailed my partner, then booked return tickets to Amsterdam and a weekend in our favourite hotel there. (Amsterdam is cheaper and easier to get to than London, from Edinburgh, not to mention more fun.) We flew out, checked in, and I lay awake until 4am mentally drafting and redrafting my resignation email (for delivery on Monday at 9am). This being the pre-wifi dark ages, I had neither the means nor the opportunity to send it; so we spent Friday morning wandering around Amsterdam in the rain.

Then my mobile rang. I looked at the screen: it was Dave. "Yes?" I asked.

"It's about KLELine," he said. A pause. "They're being shut down by Paribas!" I heard a tinny sound of applause through the phone earpiece. "So you can cross them off your to-do list ..."

It was the applause — from the rest of the office — that convinced me not to send the resignation email. Instead, Feorag and I headed for a pub, where we proceeded to get drunk. And I managed to sleep solidly that night for the first time in a month.

The pub we got drunk in, incidentally, is the Cafe in de Wildeman. And the concentrated sense of future shock I'd been dealing with in my job for the past year — of the pace of change that has accelerated until it's all a whirling blur that you can barely keep track of — came out in the fiction. That weekend I began doodling the intro to a short story on my Psion: "Manfred's on the road again, making strangers rich ..." It was the beginning of the story "Lobsters", which in turn grew into the novel Accelerando, and kept me on the Hugo award shortlists for most of the first half of the past decade.

Then on Sunday we flew home, and it was shortly thereafter that Dave finally twisted my arm into signing on as a full-time employee rather than a contractor; and at about the same time, Jeremy Spencer, then editor of Computer Shopper, offered me the monthly Linux column.

(To be continued ...)