Comment by shiandow

Not sure about the equation at the end. Looks like you may have incorrectly thought that e^(x/y) = e^x / e^y.

I get wanting to get a sense of the runtime compared to the time spent checkpointing but the equation as stated is wrong. A better one is

Ne^(l/f) / Cf = f/l

Which basically means the ratio between checkpointing and runtime (including retries) should be inversely proportional to the ratio of checkpoints and flakes.

Also really, you're calling the flakiness parameter lambda and the checkpointing parameter f?