On April 1st at 8pm, tickets for the Glastonbury Festival 2004 went on sale. Potential buyers had two options; the phone line or the Web site. To reduce the chances of touts getting their hands on tickets, all sales were centralised; one phone number, one Web site.

It's become fairly obvious that this centralisation was not a good idea. The Web site has collapsed under the strain. OK, according to the official Glastonbury site, it "has never gone down", but since it's inaccessible to the vast majority of people, that's a technicality. It's been hit by an unexpected and unintentional Denial of Service attack.

Many people will have redialled the phone number over 1,000 times since yesterday evening: redial, engaged, hang up, redial, engaged, hang up, etc. Sometimes, perhaps once in 20 times, they'll get a recorded message along the lines of "thanks for calling, all the lines are busy, please try again later." Nice.

Doing some rough calculations based on the quoted figures for site accesses, there are probably around 10,000 people trying to get through to each of the 200 phone operators at any given time. And probably hundreds of thousands trying to buy tickets over the Internet.

Commercially speaking, this isn't really a problem. The tickets will quickly sell out and I'm sure that a lot of people will have a great time. But many times that number will be annoyed, having spent a day trying to buy tickets and failing, and feeling that they've wasted their time.

Michael Eavis, the event's organiser, asked in a radio interview whether people would really have been happier had the operation been scaled up and all the tickets sold within two hours. I suspect the answer's yes, if only so they didn't waste so much time.

More importantly, if the operation had been scaled up it might have been more fair. Server load balancing is a tricky business; if many requests hit a server simultaneously, which ones does it reject? How fast are retries sent? It's possible that people whose ISPs are topologically closest to the ticketing server may have a greater chance of getting tickets.

The same may be true of the phone lines. I tried three different phone providers and the results (i.e. the number of times that I got through to the recorded 'busy' message) varied considerably between them.

So, depending on your ISP, your phone provider and your geographical location, you may (although I'm theorising here) have more chance than other people of getting your tickets. With this kind of load, milliseconds matter.

This isn't really anyone's fault. I doubt even the optimists would have predicted this level of demand. But it does show that centralising ticket sales will only work if the single company involved has the same resources as all of the other usual outlets put together. Something to consider for next year, perhaps.