Benchmarks for dynamic languages

A comparison between a couple of dynamic procedural languages and their different implementations. Here is what is measured:

• Lua 5.3.3
• Luajit 2.0.4
• Python 3.5.1
• Pypy 2.4.0
• HipHop VM 3.13.2
• NodeJS v6.2.1
• OpenJDK 1.8.0_92 (not dynamic, but here as a reference)
• PHP 7.0.7
• Ruby 2.3.1

Goal here was to compare languages and how their interpreters or JIT compilers perform at simple programs. I took five benchmarks from The Computer Language Benchmarks Game and added another. I took these ones, because not all work with Pypy or Luajit implementations of their respective languages. Problems:

• binary tree
• fannkuchredux
• fasta
• nbody
• spectralnorm

Later I added two of mine problems: bubblesort (highly iterative) and non optimized fibonacci (highly recursive). And here are the results:

Do it yourself

Clone and run init.py. This will download benchmark script from The Computer Language Benchmarks Game, change ini file and copy over programs. Further instructions for actually running the benchmars and generating images are given when init.py finishes.

Conclusion

Here are my thoughts on these results:

• Lua - a very fast interpreter, even faster than some JITs on this list. Its biggest drawback is when (re)alocating heavily - see Binary trees.
• Luajit - crazy fast, fastest dynamic jit. It really shines at quick programs. But when running more time, it gets surpassed by Java. Only problem, that can be said about luajit is its performance when allocating a lot of space.
• Python - slowest interpreter on the list. In the future, PHP5 interpreter will be added to compare this against. Worst case scenario fo plain python is recursive heavy Fibonacci program.
• Pypy - ususally scores inbetween slower (HHVM, Lua and Python) and faster (Node, Luajit, Java) programs. It still has a lot of problems when a lot of allocating is taking place.
• HHVM - tries to beat Pypy-s performance and sometimes it actually does, but still not quite there. Just like native Python implementation, this also soffers at recursive heavy problems.
• Node - a very fast JIT, beaten only by Luajit and Java. Is is a couple of times faster than Pypy. For some reason it shines most when doing bubble sort.
• Java - when shorter programs are executed it is still beaten by Luajit but when execution is longer than a couple of seconds it overtakes Luajit and is fastest.