Web Operations

Datacenter Footprint

We know that datacenters do fail, so you're going to need at least two of them. But is that enough? Would three or more be better? That depends on three factors: cost, complexity, and performance. Ironically, availability isn't a factor; once you have at least two datacenters in widely disparate regions, the chances of losing both at the same time are strikingly low.

In a two-datacenter model, you need to provide enough capacity in each datacenter to handle 100% of the load, in case the other one dies during peak traffic. So, in total, you're going to have to purchase twice as much server gear as you would with a single-datacenter solution. That's a 200% increase in CapEx; ouch. If you instead move to a three-datacenter model, those numbers start to go down. Now you need to build capacity for only 50% of the load in each datacenter. Assuming one goes down, you can just split your traffic evenly between the remaining two datacenters to handle 100% of your traffic. Add those three datacenters up, and you have a 150% increase in CapEx. Better! On the other hand, the additional infrastructure, rent, and staffing required to build out the third datacenter may eat up that savings, and more.

A three-or-more-datacenter solution also adds complexity and additional risk. It's difficult to guarantee that, having lost one datacenter, the remaining traffic will split perfectly evenly across the remaining two. Without that guarantee, you may be subject to cascade failure. Let's ...