Recently I've been working on a library for generating large JSON[1] documents quickly. Originally I started writing it in Haskell, but quickly encountered performance problems. After exhausting my (meager) supply of optimization ideas, I rewrote some of it in C, with dramatic results. Namely, the C solution is
* 7.5 times faster than the fastest Haskell I could write (both using raw pointer arrays) * 14 times faster than a somewhat functional version (uses monads, but no explicit IO) * >30 times faster than fancy functional solutions with iteratees, streams, etc I'm wondering if string processing is simply a Haskell weak point, performance-wise. The problem involves many millions of very small (<10 character, usually) strings -- the C solution can copy directly from string literals into a fixed buffer and flush it occasionally, while even the fastest Haskell version has a lot of overhead from copying around arrays. Dons suggested I was "doing it wrong", so I'm posting on -cafe in the hopes that somebody can tell me how to get better performance without resorting to C. Here's the fastest Haskell version I could come up with. It discards all error handling, validation, and correctness in the name of performance, but still can't get anywhere near C: http://hpaste.org/fastcgi/hpaste.fcgi/view?id=16423 [1] http://json.org/ _______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
