Europeana website overwhelmed?

The 20th November saw that launch of the Europeana portal and its almost immediate technical crash. I was struck by the excitement with which this was hailed as an overwhelming success, evidently demonstrating the high user demand for such a service.

You can read the official notice here.

Upon closer examination, it seems that they had more that 3,000 simultaneous users, which was what brought the system down.

3,000 ā€¦ not 10,000 not 100,000 but 3,000.

This does not seem to be a very high number (the 24HM site regularly gets more than 7,000 in any one day of which there must 3,000 at any one time on a regular basis?).

I am shocked that a system of this scale, with substantial investment, had such low targets for its audiences ā€“ I also wonder what will have happened to those that tried to visit the site in the first flurry of PR coverage, will they return?

Hmmm ā€¦

4 thoughts on “Europeana website overwhelmed?

  1. A different kind of “simultaneous”, perhaps. If 7000 simultaneous users means 7000 requests being made/handled at the same time, as opposed to 7000 user sessions open at the one time, it’s more of a challenge than perhaps you give credit for, and probably an order of magnitude or two more than that faced by a server serving 7000 users a day – even bunched up they are not going to be requesting server resources at the same time at anything like that level. If you’re experiencing around 10 million hits per hour, as Europeana did that first day (well, 13M at its actual peak) that translates to coming on for 3000-4000 per second (10M or 14M/3,600), a fairly big load. Whereas if you bunch up 7000 users so that perhaps 3000 of them make between them 300,000 requests in an hour, that’s 80 per second and your server can breathe easy. And if you flip the calculations around, you could guesstimate that they were expecting perhaps a maximum of maybe 200-400K users per day, not that low an ambition.
    Of course it’s true, perhaps the big bang launch was asking for trouble, although perhaps that was unavoidable given the nature of the project. How much capacity they allowed for had to be a balance – it’s a prototype and the service is hosted by KB, who they have a business decision to make about how much hardware they’re able to dedicate to it. I await the relaunch with a positive attitude: I think when we get to try it out this blip will be history.

  2. Maybe not every day but I was referring to 7,000 unique users, who tend to be grouped heavily into the UK working hours, so I am sure there are 3,000 simultaneously occasionally šŸ˜‰

  3. Jane,
    I agree with much of what you say here (I also blogged about this: Europeana, numbers and scalable architectures.

    The “launch with a big fanfare” strategy was particularly ill-advised, as it invites this kind of disaster, as well as setting up a high profile target for bad press etc. For some years now, successful web services have adopted a quiet or “soft” launch approach, growing gradually. Of course, there is the random element of the possibility of being “slash-dotted” but, as I mention in my blog post, this can be planned for with an architecture which gives you options for “scaling-out”.

    BTW – I’m not sure that 7,000 (users? hits?) a day indicates a likelihood of 3,000 concurrent users at any one time….


Comments are closed.