Hi,

That is a technical issue, not a policy one. My suggestion is for the German chapter to act on one of their hosting offers, buy a bunch of servers to use as squids and then we can work out the ownership issues later.

A bunch of squids? We have a bunch of squids in Europe. We do not need more. Serving the web site is not just about caching :)

Not a shortfall. A squid in Germany will negate the need for buying one in Florida. The budget is based on needs; it is even likely that if the Yahoo! servers go online this quarter, then that will negate the need for buying a whole bunch of servers. The money saved would then roll-over into the next quarter's budget.

We do not buy more squids for Florida. We buy more database and application servers. Squids ain't any magic, they cache what is cachable, all other web content has to be generated. Including your stub colors. On the other hand, we're growing much faster than we could ever split.

Korea would be our testbed for splitting (at least we would not understand user troubles in their native languages :) off Asian wikis.

Please note, that there are different load patterns out there. Instead of having multiple sines (local load patterns) covering each other, we'd have to maintain separate sines, with lots of unused capacity. We have to really understand how splitting our website would affect our performance. And how manageable it would be afterwards.

The only way to really start e.g. European deployment, is to put at least 200k$ in there. But that won't be worth 200k$ spent in Florida. Or maybe even 100k$. Wikimedia cluster is already a platform of scale. We just do not want to have more resource-bound Deployments, as they are real PITA to manage.

Or as I told before - we may establish quarterly performance goals and limit that. We may explain concepts of SLA and stuff. You would not want that. And it is impossible to try out new concepts for wikis at all while operating in maxed out environment.

It was rather difficult for me to express what we need for the site with current budgeting practices. But I'd really love to go towards such hardware purchase:

200 dual opteron multi-purpose servers, ~2.5k$ each. 20 data management servers, ~15k$ each. Additional networking/support gear, ~50k$

So by the end of this year we could have a nicely functioning cluster. And I hope it wouldn't be that loaded until then. Or else we'll come up with multi-million budget.

We are running a website after all. A big website. A complex website. There is no magic in there, there are resources and resource shortages. And... We're still growing. If anyone comes with a nice optimization idea (we have heard some, ..), they can sure tell it. But for a while we will need resource expansion, whatever 20% or 50% wins are gained. So, if we can, let us not forget that.

Cheers, Domas