MiniVM: A zero-dependency cross-language runtime on par with LuaJIT and C

Re:MiniVM: A zero-dependency cross-language runtime on par w... by isaacimagine on Saturday January 08, 2022 @06:43AM (#29850743)

Personally, I find it pretty impressive that it performs as well as these runtimes despite not having a JIT compiler. I'm pretty sure Shaw's written more benchmarks, but as the README explains, it's really hard to tell what the performance characteristics of a language are without writing a larger application. So far the largest applications written with MiniVM are Paka[0], a self-hosted language similar to Lua that targets MiniVM; os49[1], an operating system built on Paka/MiniVM in the spirit of lisp machines; and xori[2], an online playground for the language.

Edit: to address your edit, the the Tree benchmark starts small, but grows exponentially (the graph is logarithmic). The final runs of the benchmark take about 15 seconds to run, and are run 10 times per language, at which point startup time is no longer a large factor. The fib test compares both luajit with the JIT on and JIT off: as MiniVM is not JIT'd, I think that this is a fair comparison to make.

[0]: https://github.com/FastVM/paka

[1]: https://github.com/ShawSumma/os49

[2]: https://fastvm.github.io/xori

[ Parent | on HN ]

Re:MiniVM: A zero-dependency cross-language runtime on par w... by iainmerrick on Saturday January 08, 2022 @06:47AM (#29850771)

Sure, it’s a nice piece of work!

I just don’t think you should claim it’s very fast unless you can show that it actually is very fast. The current benchmarks look like toy benchmarks. (I don’t mean that in a derogatory way -- they’re exactly the kind of tiny programs you need to build and test while bringing up a new VM.)

[ Parent | on HN ]

Re:MiniVM: A zero-dependency cross-language runtime on par w... by mst on Saturday January 08, 2022 @10:31AM (#29852564)

The disclaimer at the top of the benchmark section - https://github.com/FastVM/minivm#benchmarks - seemed pretty reasonable to me.

It's very clear that microbenchmarks don't tell the whole story, and that whole application benchmarks to give more real results will be desirable later.

I mean, I agree these benchmarks don't mean that much except for being a useful way to verify that the core code is working and reasonably performant - but they're up front about that fact so calling it 'benchmark-gaming' seems a trifle unfair.

[ Parent | on HN ]