Skip to content

[PERF] Improve runtime when optimizing on disk

Every now and then I will look how to improve the runtime.

Profiling two Rimea tests with Nelder Mead, both times it looked similar to this:

image

  • getNextPosition take ~70% of the runtime
    • about 60% is spent in the Apache code for the step optimization

So in this area of the code is great performance-potential to improve runtime.

I will have a look into the Nelder Mead settings, i.e. how is it configured, maybe some settings can get relaxed without changing behaviour.

It is also interesting how many potential point evaluations are there per optimized step.


Furthermore, I want to evaluate other optimization methods, mainly also taking the derivative.

To upload designs, you'll need to enable LFS and have an admin enable hashed storage. More information