In 2004 I had finally edited down everything that I wanted from my infrastructure into a single sentence and then reused this phrase in a lot of the proposals during the early years of Joyent.
I simply wanted to run applications well and keep data well. And needed
A highly available, redundant and modular setup with a determinable QoS, easily administered, scaleable, and can be right-sized and geographically distributed without issue.
Simple. The “determinable QoS” is the hard, unknown part. And by unknown, I mean that no one in the world actually knew now to do it. The redundancy requirement over the years has shifted into better parts.
The task then became a single question too.
How do I physically and logically design a scalable, cost-effective datacenter independent of knowing what I’m going to run on it and still manage to run these unknown applications well?
Cost-effective is synonymous with “don’t over-engineer it”.
Then it turned out that there wasn’t a vendor in existence that we could buy this from.