Estimating Google's number of servers using the WWII tank method

The German tank method relies on a unique probability density function created specifically for the task of estimating tank production.

It takes into account factors unique to tanks, so you'd have to define your own distribution for Goolge's gadget servers to make a correct estimation.

Factors unique to tanks? I thought it assumed only that tanks would be assigned unique sequential ID numbers.

Someone else pointed out the most likely accuracy issue - Google may (probably does) use a dynamic balancing system to add and remove servers/nodes according to load.

The checks I ran were done over just 4 days. The resulting numbers may have no relevancy at all over other time periods with different load amounts.

I don't know much about the German tank method but it seems interesting that given 62 unique addresses with a maximum of 103 it would give a total number of 104. This doesn't seem to make sense to me. Can someone explain why this works this way?

You need to read the explanation for the German Tank method (http://www.guardian.co.uk/g2/story/0,,1824525,00.html) to see the formula used to get 104.

well there are 42U in a rack, does that help?