Hi,
Oops. I meant "counting unique users just like we do with directory fetches". The argument given was that most users are already behind NAT and hence counting unique IP addrs would not be accurate anyway. It was suggested that we count per connection country statistics (that is, total number of connections coming from a country) and divide that by average number of connections a user makes, to arrive at estimated number of unique users. I cannot find where I got to know this. Maybe on IRC, but I don't have logs of an year ago.
Also, the Metrics team has(?) to come up with a proposal on this IIRC. Until then it would not be considered a valid project?
Karsten would have something to say on this.
cc: karsten