There’s some excellent stuff about Wikipedia’s architecture floating around as well. There was a bunch of fascinating talks at some wiki conference in Berlin, if I remember right (lost the link) – some detail on the architecture and some fascinating insights like how the need to “diff” word by word, not line by line.

A quote from a fairly lightweight article here: http://www.wired.com/news/culture/0,1284,66210,00.html

“One of the mysteries of scale is that there’s no such thing as scaling well,” said Clay Shirky, who writes about culture, media and technology. “You can make something 100 times bigger, and if it works, you think you’ve got it licked. But the next power of 10 can kill it. So I don’t know whether or not openness and co-creation are incompatible at Wikipedia scale.”

So that’s the opinion of someone who has scaled.

In fact I wonder if PHP (and MySQL, Apache, Squid and everything else involved) needs to bang the wikipedia drum more often. I love to hear more from those involved.

When you look at alexa’s top500, Wikipedia is number 19;

http://www.alexa.com/site/ds/top_sites?ts_mode=global&lang=none

Now looking at the sites ranked higher, only myspace.com (coldfusion?!?) and ebay (claims to have “gone J2EE” but still much C++ around) might be considered “many-to-many”: sites which are having to cope with lots of reads and writes. The rest are effectively read only : updates are restricted to the few. What would be a really interesting statistic is the number of GET vs. POST requests these sites are having to handle.

The point is I think wikipedia certainly has unique scaling requirements and perhaps the most challenging – no one really has an architecture on their drawing board to handle this (no matter how much they get paid to invent ones). So far the LAMP stack has worked well for wikipedia. Or as a snide turn around – have Java ever scaled to anything like this?

Also wonder if Java really would be a good choice for interplanetary navigation – at least if writing such systems ever became close to mainstream – strikes me you’d want a language specialized to that task such as data structures that makes it easy to work with 3d + time. Also a long might need to be very very very long – perhaps longer than what Java offers right now? And what about relativistic data types

How do you know that Gosling was talking about web apps? It’s not possible to determine that from the article. He said:

“But none of them attempt any serious breadth in the application domain and they both have really serious scaling and performance problems.”

Since that was paired with a statement about Ruby and PHP lacking breadth, maybe he meant that they have scaling/performance problems in certain domains, not necessarily web apps. Surely you’ll agree there are some areas (desktop GUI applications, long-running system daemons, etc.) where PHP *would* have scaling/performance problems and Java would not? Perhaps that was what Gosling was referring to; who knows? This seems to be supported by his other statement:

“You can do web presentation stuff really well in PHP…”

so I think it’s likely that he didn’t mean web apps when he said that PHP has “serious scaling and performance problems”. Just my interpretation…