As we move from statically assigned builders to a general pool, builder performance will degrade substantially, if we can't find a way to safely preserve relevant artifacts from previous builds. If we CAN find a really solid solution, we might be more efficient than we've ever been.

This doc captures limited thoughts on the problem. It needs to be fleshed out, especially with the sizes and performance numbers associated with each type of "hot cache" it mentions.

go/chromeos-builder-cache-management

We need to investigate further, and see how we can best improve.