Tuesday, February 28, 2012

Hitting Moving Targets

One thing I'm interested in is understanding how accurate any population rankings I observe can be (from the previous post).

Just taking a quick glance at TorStatus, its pretty clear that the population of any given server can be... well... lumpy.  Case in point:  using the same basic methods of TorStatus, but averaging the server status of all East Coast and West Coast servers (US) and across a month, the overall server population looks like this:
Not only would that be an awesome roller coaster ride, it suggests a few things about the populations of servers.  (Numbers in military EST.)
  • Differences between nights aren't as large as you might expect.  I'm pretty sure that if I ditched the weekday labels, it'd be hard to match the charts to days of the week.  (Other than the Tuesday downtime.)  Saturday looks an awful lot like Monday.  And even though Sunday is the heaviest play time... it isn't by much.
  • Time of night matters much, much more.  Just for example, the difference between 7 and 9 pm PST is huge on a West Coast server (22 vs 24 on the charts).
What does that mean? I sampled the data for the last post over three hours on Saturday night, so I'm understandably nervous that the time it took to observe the server populations could have an effect.

Too check this, I took a quick glance at how closely the data I observed matches to the listed server status, to get an idea of how accurate my earlier observations were.  First, looking at the status at the time of sampling (triangles), 15 minutes before (circles), and 15 minutes after (squares):
Thankfully, none of the servers I examined changed their status around sampling time.  Even better - only Infinite Empire (top left, green) shows a mismatch between the server status and the observed population, with more people than would be expected.

This also gives some idea of the cut-points for each server status - or at least a nice guess for where they might be, sitting at around 500 and 1500ish.  (Hopefully I can get a bit more data to pull that together.)

While the observed data broadly match the server status - the same can't be said of the rankings between servers over time. To show this, as I collected the data, I sampled two servers (Prophecy of the Five and Veela) at multiple times as I collected the data.
Because I'm drawing the data from different times, the observed total population of each server moves pretty dramatically across the scale. Unfortunately, this makes answering whether Po5 has a higher/lower pop than Mandalore something I can't tell with any confidence.

But instead of being concerned about server size... what about the balance between factions?  How much does that move over the night?
It shouldn't be any surprise that the server populations are smaller pre-primetime. Curiously, that doesn't seem to have any effect on the balance between factions, which doesn't show much movement between times.

TLDR:  Server populations are consistent across days, but change pretty dramatically over the course of the night. These swings make estimating the relative size of servers pretty difficult. Fortunately, these changes don't influence the estimated balance between factions.


*note:  as with the previous post, The Harbinger isn't a PvP server, its just there for a point of reference.

No comments:

Post a Comment