From Fluid Solver to Studio Infrastructure

Flood simulation shown at SIGGRAPH 2003

Fluid simulations could take days to complete.
I built a simple distributed scheduler called Cloud so I could use idle machines around the office.
That let me keep iterating while long-running simulations executed in the background.
At the same time, the studio's render farm had a much bigger reliability problem.

The Problem No One Was Solving Well

My problem

Simulations were expensive and slow.
I needed a way to spread jobs across idle machines.
A lightweight shared-filesystem scheduler was good enough.

The studio's problem

Frantic was using Autodesk Backburner on a roughly 60-machine render farm.
It was unreliable enough that someone worked nights just to restart it when it crashed.
Failed overnight renders meant missed deadlines and wasted artist time.