Underprovisioning Cloud Services

The default pattern in cloud operations is to overprovision your cloud resources. At Fermyon Technologies , the experimentation we did early on led us to an interesting conclusion: It is possible to safely and reliably underprovision cloud services.

I recently wrote a blog post on Fermyon about this.

In this post, I talk about the three criteria for provisioning:

1. The efficiency criterion: don’t pay for resources that are not needed or currently used
2. The peak load criterion: even during peak load, an app should should have sufficient resources
3. The resiliency criterion: When infrastructure fails, the app should (within reason) still have the resources to handle demand

The dominant paradigm says that we can't have all three, so we give up on the efficiency criterion and just spin up enough resources to handle peak load and resiliency considerations without trying to gain efficiency,

Why give up on the efficiency criterion? Because the first two waves of cloud computing, VMs and containers, cannot scale fast enough. Scaling up is slow, and scaling down is dangerous. But the next wave of cloud computing does not suffer from this drawback. As a result, we can employ the efficiency criterion.

But more than that, as we run more and more applications, we can do better than just right-sizing our provisioning. We can underprovision cloud resources. That is, we can provision fewer resources than we would otherwise consume if every cloud service was actively running. And we can do this because we are never running all of the services at the same time.

In my Fermyon blog post, Better Than Overprovisioning: Underprovision Your Cloud Services , I cover this in detail, and throw in there some numbers.