If there was a meeting with 1,000 participants from one location (say Stockholm), and one participant from a very distant location (say, Sydney Australia), then this argument would put half of the meetings in Stockholm, and half of the meetings in Sydney Australia. Another possible criteria would be to minimize the total cost paid for travel, without regard for who is paying. With this model, if there were 1,000 participants from Stockholm, and 999 participants from Sydney, we would have all meeting in Stockholm. Of course, in this case a change of two participants could cause all meetings to switch to the other location. In practice we compromise between these two considerations, plus others (such as where companies are willing to sponsor a meeting). Thus we mostly have meetings in locations proportionately to where people are coming from. Ross -----Original Message----- From: ietf-bounces@xxxxxxxx [mailto:ietf-bounces@xxxxxxxx] On Behalf Of Olaf Kolkman Sent: Monday, August 30, 2010 3:58 PM To: IETF-Discussion list Subject: Optimizing for what? Was Re: IETF Attendance by continent The recent remark on bias against individuals[*] made me think about weighing the location preference by number of participants from certain regions. Suppose an individual from Asia attends all IETFs then her costs are that for attending 6 IETFs she gets to travel 1x regional and 5x interregional. While an individual from the US travels 3x regional and 3x interregional. Clearly there is a bias agains our Asian colleague in with respect of the costs. Using participation/contribution numbers to weigh locations minimizes the global costs (total amount of miles flown, carbon spend, lost hours by the collective, total amount of whining) but nothing of that flows back to the individual engineer that attends every time. If you want to be fair to the individual participants you have to optimize in such a way that attending 6 meetings costs the same for every individual that regularly attends the IETF. Obviously one can only approximate that by putting fairly large error bars on the costs but isn't the X-Y-Z distribution where X= approx Y= approx Z the closest optimum? (or finding one place that sucks equally for everybody) Am I missing something? --Olaf (strictly personal) [*] Independent consultants, somebody not financially backed up by big corporations. ________________________________________________________ Olaf M. Kolkman NLnet Labs Science Park 140, http://www.nlnetlabs.nl/ 1098 XG Amsterdam _______________________________________________ Ietf mailing list Ietf@xxxxxxxx https://www.ietf.org/mailman/listinfo/ietf _______________________________________________ Ietf mailing list Ietf@xxxxxxxx https://www.ietf.org/mailman/listinfo/ietf