[plt-scheme] Futures performance

> On Wed, May 26, 2010 at 11:44 PM, Jon Rafkind  wrote:
>  > On 05/25/2010 08:53 PM, Groshev Dmitry wrote:
>  >>
>  >> What am I doing wrong? Here is the code and some tests:
>  >>
>  >> (require scheme/future)
>  >>
>  >> (define (fib n)
>  >>   (case n
>  >>     [(0) 0]
>  >>     [(1) 1]
>  >>     [else (let ((a (fib (- n 1)))
>  >>                 (b (fib (- n 2))))
>  >>             (+ a b))]))
>  >>
>  >> (define (fibp n)
>  >>   (case n
>  >>     [(0) 0]
>  >>     [(1) 1]
>  >>     [else (let ((a (future (lambda () (fib (- n 1)))))
>  >>                 (b (future (lambda () (fib (- n 2))))))
>  >>             (+ (touch a) (touch b)))]))
>  >>
>  >>
>  >>>
>  >>> (time (fib 30))
>  >>>
>  >>
>  >> cpu time: 546 real time: 539 gc time: 0
>  >> 832040
>  >>
>  >>>
>  >>> (time (fibp 30))
>  >>>
>  >>
>  >> cpu time: 8986 real time: 7814 gc time: 391
>  >> 832040
>  >>
>  >
>  > Is that actually the code you are running? If I take your code verbatim I
>  > get times like these
>  >
>  > ;; I set n to 40 so you can see a real difference
>  > cpu time: 6100 real time: 6448 gc time: 0
>  > 102334155
>  > cpu time: 6132 real time: 3995 gc time: 0
>  > 102334155
>  >
>  >
>  > But if I change the code so that `fibp' makes recursive calls to `fibp'
>  > instead of `fib' then I see `fibp' taking much longer.
>  > ;; n = 33
>  > cpu time: 496 real time: 498 gc time: 0
>  > 3524578
>  > cpu time: 17429 real time: 18720 gc time: 748
>  > 3524578
>  >
>  
>  Last I heard, you can't create futures within a future (the runtime
>  blocks the future that tried).
>  
>  If that's still the case, it explains why you see no speedup. I don't
>  know if that's enough to cause the slowdown you're seeing.
>  
>  > Which is similar to your results. What sort of machine are you on? I'm on a
>  > dual-core amd 1.8ghz laptop.
>  > _________________________________________________
>  >  For list-related administrative tasks:
>  >  http://list.cs.brown.edu/mailman/listinfo/plt-scheme
>  >
>  >
>  _________________________________________________
>    For list-related administrative tasks:
>    http://list.cs.brown.edu/mailman/listinfo/plt-scheme
>  
Yes, it must be "fibp", not "fib".
Runtime blocking is an explanation, but it seems to me that there is still to much slowdown for just blocking.