Most what was told in all the thread was about 'rewriting' in another languages, but it totally lacked ideas on how wiki should be implemented architecture-wise. Right now, being pushed to LAMP model, we have several issues. (warning, rants on crack.. :)

We do use relational data models. They are quite ok for content website, not great for massive collaborative editing. In order to make really large site work we'd either have to split everything and decrease funcitonality (no shared watchlists, Timwi :), or start using some other concepts of scale - messaging middleware, event brokers, whatever. With more and more data out there, various tasks could be moved away. Async aggregated feeds Though, as we're running on simple 6-disk dual-opteron db servers we're far from that.

Our current trouble is too much of data on DBs. With too much of data, more i/o has to be done for some or another task. This is partially solved by 1.5 mediawiki. Could be solved more by relocating lots of data to other external stores.

And our current trouble is single media host, though, it is supported by squids. External storages will help here as well some day.

I don't really understand what's all that fuss about. The only real need to rewrite Wikimedia software is to prepare it for 300+ server infrastructure. Then we'll need more intelligent solution than single LAMP script-framework. We need ideas on how to improve architecture of whole wiki-process, than ideas which language is better and which language is worse.

With PHP5 we can easy convert our mediawiki into daemons (HTTP, FastCGI, whatever), that would gain performance. Sure, we _could_ do same with Python. Sure, Python has JIT right now, but there are commercial JITs for PHP out there (which means there will be opensource some day, Parrot, whatever :)

And sure, there's one more thing that can be done... Simply replacing Apaches with Lighttpd (http://www.lighttpd.net/). We've got great support by lighttpd folks, and have been running it for our downloads service, and I've been running for my projects 'at home' with great success. Ah, that could provide at least 10% (eh, say... 50%).

Domas

P.S. I didn't participate in main discussion as I was away, and I did later did an intentional break, just to calm everything down :)