When I joined VMware, friends asked why? Wasn’t the future public cloud?
And I rejected that hypothesis.
The list below is an abridged version of a lot of deep thinking. These points have served me well over the years as I think about infrastructure.
One: Capex is more efficient spend than op-ex for things that can survive for more than 3 years
It’s taxed more efficiently, frees up cashflow, etc, etc. As power moves from gas/to solar, and spinning media runs on silicon, hardware can last about 1 decade. The systems don’t fail and the cost of running goes to 0.
If you know your capacity, you can literally buy once capitalize for 3 years, and run it for free for 7
Mainframes continue to survive for that reason.
Two: The Turing machine can run any program, and yet we have all kinds of hardware.
Why?
For any given workload, there is optimal hardware that will deliver the desired performance/reliability at an optimal cost.
Cloud doesn’t offer that hardware.
Three: Any prototype of a new system is best done in a typeless scripting language, any understood system is best done in a typed compiled language leveraging hardware
Every python project ever written that required performance or reliability had to be re-written in C/C++
Cloud optimizes for agility, not optimal execution.
Four: Computer systems that are inherently reliable are cheaper to operate than computer systems that are not
The single most important variable in making a system unreliable is how often it changes. A system that never changes, never breaks. The more people that touch a system, the more unreliable it becomes the more costly it is to operate.
Cloud infrastructure is always in-flux, thefore less reliable.
Leave a Reply