Clojure Transients Example - No significant speedup

Question

I copied the code from:

http://clojure.org/transients

but my results differ signifiantly from what was posted.

(defn vrange [n]
  (loop [i 0 v []]
    (if (< i n)
      (recur (inc i) (conj v i))
      v)))

(defn vrange2 [n]
  (loop [i 0 v (transient [])]
    (if (< i n)
      (recur (inc i) (conj! v i))
      (persistent! v))))



(quick-bench (def v (vrange 1000000)))
"Elapsed time: 459.59 msecs"

(quick-bench (def v2 (vrange2 1000000)))
"Elapsed time: 379.85 msecs"

That's a slight speedup, but nothing like the 8x boost implied in the example docs?

Starting java in server mode changes the story, but still nothing like the docs..

(quick-bench (def v (vrange 1000000)))
"Elapsed time: 121.14 msecs"

(quick-bench (def v2 (vrange2 1000000)))
"Elapsed time: 75.15 msecs"

Is it that the persistent implementations have improved since the post about transients here: http://clojure.org/transients ?

What other factors might be contributing to the lack of boost with transients?

I'm using the OpenJDK java version 1.7 on ubuntu 12.04. Maybe that's a lot slower than the (presumed) Hotspot 1.6 version used in the docs? But wouldn't that imply BOTH tests should be slower by some constant, with the same gap?

Answer 1

Your results are consistent with my experience with transients. I've used them quite a bit and I typically see a 2x performance improvement.

I tried this on Ubuntu 12.04, OpenJDK 1.7 with Clojure 1.6.0 and 1.7.0-alpha3. I get 2x performance with transients, slightly less than the 3x I get on OSX with the 1.8 Oracle jvm.

Also the above page is from the time of Clojure 1.2, and the performance of collections has improved significantly since then. I tried the experiment with 1.2 but Criterium doesn't work with it, so I had to use time just like on that page. Obviously the results are significantly variable (from 2x to 8x). I suspect the example in the documentation may have been cherry-picked.