This seems like a great place to ask: how does one go about optimizing something like zram, which has a tremendous number of parameters [1]?

I had considered some kind of test where each parameter is perturbed a bit in sequence, so that you get an estimate of a point partial derivative. You would then do an iterative hill climb. That probably won't work well in my case since the devices I'm optimizing have too much variance to give a clear signal on benchmarks of a reasonable duration.

[1] https://docs.kernel.org/admin-guide/sysctl/vm.html