Daniel Mayer wrote:
When that becomes a significant issue, then the model will need to be updated. I've been told that, so far, pretty much all the hardware we've ever bought is still in use.
That's exactly right.
It might at some point pay to look at selling some of the older servers to buy newer ones. This would have two benefits:
1. we are now exclusively buying dual-opterons for apache webserving, these are significantly more powerful _in the same physical space_ than our single-pentiums -- therefore, we can save a lot of rack space at the colo by replacing. (This space is not especially expensive, though, in the grand scheme of things.)
2. In addition to saving physical space, there is a fair amount of dev time which goes into working with 124 servers (the current count, I believe, except not all of them are installed just now) which could be reduced if we were running on half the number of servers. Wikipedia traffic is growing faster than Moore's Law, but even so, Moore's Law should make it possible for us to do more with fewer boxes.
The first benefit can be quantified, the second cannot. (How much money should we be willing to spend to save the developers some time? My answer is: a LOT. Developer time is not free to us, it is infinitely expensive. What I mean by that is that it is a lot more cost effective to have volunteer devs working in a well-funded and exciting environment where they can play with cutting-edge technology, than it is to eventually be forced to hire devs to work with boring and annoying old hardware.)
----
My suspicion is that by the time we are seriously ready to get rid of some old machines, they will have minimal market value. Dunno.
--Jimbo