This post originated from an RSS feed registered with Ruby Buzz by Eigen Class. Original Post: Comparing lightweight threads Feed Title: Eigenclass Feed URL: http://feeds.feedburner.com/eigenclass Feed Description: Ruby stuff --- trying to stay away from triviality. Latest Ruby Buzz Posts Latest Ruby Buzz Posts by Eigen Class Latest Posts From Eigenclass

The Computer Language Benchmarks Game includes a benchmark that measures context switch performance. The entries can be classified into three categories:

1. those that rely on lightweight threads, namely Haskell GHC, Erlang HiPE and Mozart/Oz. I don't know whether Smalltalk VisualWorks and Scala belong here too; their performance is far from the top three entries.
2. those that use POSIX threads, with or without actual parallelism, including most other entries
3. Ruby, a category of it own (?): very expensive user-mode mode threads with no parallelism

The OCaml entry is amongst the fastest pthread-based ones, but still markedly slower than the top entry, by around an order of magnitude. The version I wrote some time ago, based on the Lwt cooperative lightweight thread library, is close in performance to GHC. Some analysis reveals interesting facts about GHC's concurrency support and Lwt.

Performance

Here are the figures I get on a dual-core AMD Athlon 64 X2 6000+ in 32 bit mode (Why 32 and not 64? Because it's faster in this benchmark; 64 bit pointers are heavy and we get nothing in return in this case.)

The Haskell code was compiled with -O2 (GHC 8.8.2); for Erlang, I used erlc +native +"{hipe, [o3]}" (Erlang R12B-1).

GC overhead

The OCaml version is clearly GC bound, and performance increases as the minor heap is enlarged, decreasing the amount of GC work. Whereas with the default 256KB heap the Erlang program is slightly faster, when OCaml is allowed to use comparable amounts of memory, it is over twice as fast.