On 22/02/16 18:45, Erik Moeller wrote:
The numbers for "very active editors" appear to have stabilized at a slightly higher level than previously. I can't find any firm conclusion on what has caused this in Wikimedia's public communications, but the HHVM rollout, long-planned and implemented in December 2014 under Ori Livneh's leadership seems like a plausible hypothesis:
https://blog.wikimedia.org/2014/12/29/how-we-made-editing-wikipedia-twice-as...
I don't think it is plausible, given the data collected at:
https://meta.wikimedia.org/wiki/Research:HHVM_newcomer_engagement_experiment
25,000 new users were put into an HHVM bucket, so the whole site was twice as fast for them. Then they were tracked for a week. There was no improvement in engagement or productivity.
I'm sure the performance improvements we did in 2004-2005 had a big impact, especially initial batch of 9 Tampa servers in February 2004. There must be a scale effect: going from 20s to 10s is much more important than going from 2s to 1s.
-- Tim Starling