Performance with short lived processes

Open EliaCereda opened this issue 3 years ago • 2 comments

I'm trying to run many short lived command-line processes under FEX (a pre-compiled embedded GCC toolchain) and I'm wondering if there are configuration options that I can tune to achieve the best performance.

I saw a -20% improvement with an unpacked RootFS compared to squashfs, but performance is far from native (6s to compile an hello world). Are there other options that can be tuned for this use case?

It's my (limited) understanding that the transpiled ARM binary is not cached between different runs by default, but I'm not sure if it can be enabled. How are these options supposed to be used?

Thanks!

Jun 23 '22 16:06 EliaCereda

The IR/AOTIR options are semi-broken, and object cache is work in progress. Both should be in working order for august release, no promises there though.

Short term there's not much you can tweak, sadly.

Jun 23 '22 16:06 skmp

Thanks for the quick answer, it's great to know it's being worked on. Happy to test these changes when they'll be ready.

Jun 23 '22 16:06 EliaCereda