10 Sep 11, 10:05AM
during a few IRC-talks that went slowly I did some python prototyping myself, a while back. you might remember me quoting some percentages on map/mode-combos and the used-servers to registered-servers ratio ;-)
Well, the code was hardly thought out, just on-the-fly .. the polling bit was of more interest to me anyway - I even included some geoIP stuff. Retrieving and displaying nicely (monospace FTW!) the maprot and stuff like that; wanted to even go as far as querying the whole range of EXTping (score-by-player and stuff)...but never made time for it.
The stats turned out to get more skewed as the errors I was doing on-the-fly accumulated; so I broke off that experiment.
My guess now would be to try to simply collect stats on an individual server - so you'll need some way to uniquely identify them.
You can always provide a "last accumulated version" of the stats for each server, if you combine all of those to one chunk you can then run (weighted!(*)) analysis on the "entire network" with an acceptable range of deviation.
_____
*: what you weigh on is the tough choice, but you can't simply add any specific values coming from a server online 24/7 during your last 3 weeks of polling data from the network with the pendant from one that just popped up yesterday and has been online 7.25 hours out of those 45.5 since you first saw it.
Well, the code was hardly thought out, just on-the-fly .. the polling bit was of more interest to me anyway - I even included some geoIP stuff. Retrieving and displaying nicely (monospace FTW!) the maprot and stuff like that; wanted to even go as far as querying the whole range of EXTping (score-by-player and stuff)...but never made time for it.
The stats turned out to get more skewed as the errors I was doing on-the-fly accumulated; so I broke off that experiment.
My guess now would be to try to simply collect stats on an individual server - so you'll need some way to uniquely identify them.
You can always provide a "last accumulated version" of the stats for each server, if you combine all of those to one chunk you can then run (weighted!(*)) analysis on the "entire network" with an acceptable range of deviation.
_____
*: what you weigh on is the tough choice, but you can't simply add any specific values coming from a server online 24/7 during your last 3 weeks of polling data from the network with the pendant from one that just popped up yesterday and has been online 7.25 hours out of those 45.5 since you first saw it.