Re: Excuse my ignorance.
I hope you don't mind me re-posting my comments above? I think they fit better here:
It's a linear programming problem.
If you build your energy reserve just big enough to cover the specified power-outage duration, it is a lot of cost that sits there doing nothing for nearly all of its life.
If you increase the reserve by just 20%, you can trade power every day, generating an income. You can get the best spot-price of the day, and the wear-out costs for just 20% cycle-depth are minimal. It also keeps the equipment "exercised" at full power, so you can trust it.
You might just choose to decline the peak electricity price and run your datacentre from batteries for an hour a day, using electricity that you've bought at minimum price, minus your electrical losses.
Now run the numbers for [x%] of additional reserve, additional installed cost, and you can see it works better. In fact, for a 20% increase in capacity you'd only be looking at more batteries, the converters would remain the same - so a cost delta of 10% say.
You can even play with the statistics - the power outage duration is a Gaussian, unlikely to coincide (in its worst case) with you being at your minimum of stored power. So, for a small increase in the risk that you won't cover the outage (already a non-zero risk) you can pay back some costs, even with zero increase of your energy reserve.
Biting the hand that feeds IT
