On Sat, Sep 17, 2011 at 02:42:08PM +0300, Strainu wrote:
I believe the most difficult part would be to replicate the foundation's "secret sauce", i.e. the configuration files that are not made public, if such thing exists.
Special:Version has always been good enough for me ;-)
Then would come the whole traffic balancing/caching/optimization settings, which would greatly depend on the actual traffic a fork would have.
My first instincts for de.wikipedia would be to note down de.wikipedia's usage statistics, get a bunch of techies together, and all go have a nice chat with say hetzner.de, to figure out roughly what things will cost. You can always start a bit small and work your way up.
sincerely, Kim Bruning
( To help jumpstart maths:
Renting a 49 unit 19" rack with 1TB traffic, 1GB/sec costs around Eur 200/month these days. You still have to buy equipment to mount in that rack and set it all up, of course, which might cost you around 50KEur[1] Rent+obsoleting over 3 years gets in the ballpark of about 20K/year amortized. Extra traffic allowance can typically be ordered separately.
You may need more, or less, than a single rack, depending on traffic.
* http://www.nedworks.org/~mark/reqstats//trafficstats-hourly.png * http://www.nedworks.org/~mark/reqstats//trafficstats-monthly.png For the entire cluster(all projects), it looks like traffic can peak to 6Gbit/s at the moment
* http://stats.wikimedia.org/EN/Sitemap.htm enwiki: 7,976,862 views per hour dewiki: 1,054,677 views per hour
[1] @~Eur 1000/system. Note that you also need to mount switches, UPS, etc, so you can't use the whole rack just for computation. Also note that things like blade servers or NAS servers can fit more processor power or storage into less rackspace, where required.
)