On 4/2/07, Daniel R. Tobias dan@tobias.name wrote:
Looking at the Alexa stats for Wikipedia: http://www.alexa.com/data/details/traffic_details?q=&url=wikipedia.org
I prefer our own stats. The graphs are much nicer, and there's other data (which cluster the request is to, for example) which proves instrumental in my analysis.
This is the latest weekly graph for requests:
http://hemlock.knams.wikimedia.org/~leon/stats/reqstats/reqstats-weekly.png
...and here's the latest monthly graph:
http://hemlock.knams.wikimedia.org/~leon/stats/reqstats/reqstats-monthly.png
Note that the week (maroon line) begins on a Monday. The highest traffic is Monday to Thursday, and sometimes to Friday, with lower traffic on weekends. The weighted peak each day, across all clusters, is around 12:00 to 18:00 UTC.
Take a look at the weekly graph. You'll see that the knams and knams-img clusters peak earlier in the day, maybe 12:00 UTC, while the pmtpa and images clusters peak later, maybe 18:00 UTC. It's hard to see when the yaseo and yaseo-img clusters peak, but if you look closely, they peak at maybe 06:00 UTC. Lopar doesn't make it into the graphs
This might not make much sense until you consider that the pmtpa and images clusters are in Florida (pmtpa stands for Powermedium, Tampa), the knams and knams-img clusters are at Kennisnet in the Netherlands (knams is Kennisnet, Amsterdam) and the yaseo and yaseo-img clusters are in Seoul (yaseo is Yahoo!, Seoul).
My theory is that more people are using Wikipedia at work and during school hours than they are in their own time. The various clusters serve traffic closest to them at different times in the UTC day, but always corresponding to the local middle-of-the-day period.