Hi Zak,
I tried your example on my i7 (4 physical cores, 8 logical); here are
the results -
1:298 user=> (time (do (doall (map fac (take 10 (repeat 50000))))
nil))
"Elapsed time: 54166.665145 msecs"
1:300 user=> (time (do (doall (pmap fac (take 10 (repeat 50000))))
nil))
"Elapsed time: 27418.26344 msecs"
With map CPU usage ~12.5%, with pmap (10 threads) ~50% average.
But when I change the fac function to say -
(defn fac
[n]
(let [n (* n 1000)]
(loop [i 0]
(when (< i n)
(* 2 2)
(recur (inc i))))))
1:308 user=> (time (do (doall (map fac (take 10 (repeat 50000))))
nil))
"Elapsed time: 48507.220449 msecs"
1:309 user=> (time (do (doall (pmap fac (take 10 (repeat 50000))))
nil))
"Elapsed time: 9320.92417 msecs"
With map CPU usage ~12.5%, with pmap (10 threads) ~95% average.
So I think it may be something to do with really BigIntegers in your
original fac, but I can't be sure. The point is that the original fac
somehow doesn't consume the entire CPU even with 10 threads ?! If you
find out the reason why let me know.
Also I didn't spend much time with your npmap function, but it seems
like a nice idea. One observation - partition-all is not very good at
dividing up work to be done. For example suppose you want to divide
up a coll of length 9 into 4 threads, there is no way to do that with
the partition-all directly -
(1:335 user=> (clojure.core/partition-all 2 (range 9))
((0 1) (2 3) (4 5) (6 7) (8))
1:336 user=> (clojure.core/partition-all 3 (range 9))
((0 1 2) (3 4 5) (6 7 8))
Something like this might be better suited (https://gist.github.com/
54313ab02d570204393b) -
1:338 user=> (partition-work 4 (range 9))
((0 1) (2 3) (4 5) (6 7 8))
The above partition-work is based on ideas which I got from MPI
programming a while back.
- Thanks
--
You received this message because you are subscribed to the Google
Groups "Clojure" group.
To post to this group, send email to [email protected]
Note that posts from new members are moderated - please be patient with your
first post.
To unsubscribe from this group, send email to
[email protected]
For more options, visit this group at
http://groups.google.com/group/clojure?hl=en
