It stands to reason that transients won't help given a small number of operations. Consider a simplified model where conversion to transient and, eventually, back to persistent takes a constant number of cycles T and every operation on a transient takes exactly d cycles less than the corresponding operation on a persistent. The savings from using transients is then nd - T for n operations between making transient and making persistent again. If d is much smaller than T then nd will be less than T for the first several n, and using transients will be slower for n less than some threshold N. (N = ceil(T/d) in fact.)
In the real world, it's more complicated than that, and N could end up not only depending on which transient operations and on vector vs. map but even on details of the system hardware, Clojure version, and JVM version. However it's likely to be a small integer in every case, so if you're doing dozens or hundreds or more of operations, transients will generally help, and if you're only doing one or two, don't bother. On Fri, Nov 9, 2012 at 7:27 AM, Softaddicts <lprefonta...@softaddicts.ca>wrote: > Ubuntu 12.04, java 6 32 bits.... > > Luc P. > > > > This is weird...for just 2 operations I'm getting consistently better > > timings with persistent than transients... > > > > user=> (time (dotimes [_ 1000000] (assoc [1 2 3 4 5 6 7 8 9 0] 3 5 6 8))) > > "Elapsed time: 240.247405 msecs" > > nil > > user=> > > user=> (time (dotimes [_ 1000000] (persistent! (assoc! (transient [1 2 3 > > 4 5 6 7 8 9 0]) 3 5 6 8)))) > > "Elapsed time: 354.122806 msecs" > > nil > > user=> (time (dotimes [_ 1000000] (assoc [1 2 3 4 5 6 7 8 9 0] 3 5 6 8))) > > "Elapsed time: 188.765493 msecs" > > nil > > user=> (time (dotimes [_ 1000000] (persistent! (assoc! (transient [1 2 3 > > 4 5 6 7 8 9 0]) 3 5 6 8)))) > > "Elapsed time: 318.90242 msecs" > > nil > > > > Jim > > > > ps: are you guys on java 1.7 ? > > > > On 09/11/12 02:29, Softaddicts wrote: > > > Roughly... My FPU module is slow tonight.... > > > > > > > > >> My own tests shows that the transient version is twice as fast: > > >> > > >> (time (dotimes [_ 1000000] (assoc [1 2 3 4 5 6 7 8 9 0] 3 5 6 8))) > > >> Elapsed time: 711.848312 msecs > > >> > > >> (time (dotimes [_ 1000000] (persistent! (assoc! (transient [1 2 3 4 5 > 6 7 8 9 0]) 3 5 6 8)))) > > >> Elapsed time: 399.466556 msecs > > >> > > >> Luc P. > > >> > > >>> user=> (use '[criterium.core]) > > >>> nil > > >>> user=> (bench (assoc [1 2 3 4 5 6 7 8 9 0] 3 nil 6 'a)) ;;2 > operations > > >>> > > >>> Evaluation count : 125829120 in 60 samples of 2097152 calls. > > >>> Execution time mean : 488.826554 ns > > >>> Execution time std-deviation : 19.952095 ns > > >>> Execution time lower quantile : 459.910320 ns ( 2.5%) > > >>> Execution time upper quantile : 527.008269 ns (97.5%) > > >>> nil > > >>> user=> (bench (persistent! (assoc! (transient [1 2 3 4 5 6 7 8 9 0]) > 3 > > >>> nil 6 'a))) > > >>> Evaluation count : 97399860 in 60 samples of 1623331 calls. > > >>> Execution time mean : 639.789263 ns > > >>> Execution time std-deviation : 26.265080 ns > > >>> Execution time lower quantile : 597.753127 ns ( 2.5%) > > >>> Execution time upper quantile : 687.838668 ns (97.5%) > > >>> nil > > >>> user=> (bench (assoc [1 2 3 4 5 6 7 8 9 0] 3 nil 6 'a 0 "z" 2 5)) ;;4 > > >>> operations > > >>> Evaluation count : 68275920 in 60 samples of 1137932 calls. > > >>> Execution time mean : 901.407416 ns > > >>> Execution time std-deviation : 35.682637 ns > > >>> Execution time lower quantile : 851.060950 ns ( 2.5%) > > >>> Execution time upper quantile : 970.492949 ns (97.5%) > > >>> > > >>> Found 1 outliers in 60 samples (1.6667 %) > > >>> low-severe 1 (1.6667 %) > > >>> Variance from outliers : 25.4849 % Variance is moderately > inflated by > > >>> outliers > > >>> nil > > >>> user=> (bench (persistent! (assoc! (transient [1 2 3 4 5 6 7 8 9 0]) > 3 > > >>> nil 6 'a 0 "z" 2 5))) > > >>> Evaluation count : 68844120 in 60 samples of 1147402 calls. > > >>> Execution time mean : 893.798444 ns > > >>> Execution time std-deviation : 24.566655 ns > > >>> Execution time lower quantile : 848.307600 ns ( 2.5%) > > >>> Execution time upper quantile : 952.841308 ns (97.5%) > > >>> > > >>> Found 13 outliers in 60 samples (21.6667 %) > > >>> low-severe 7 (11.6667 %) > > >>> low-mild 6 (10.0000 %) > > >>> Variance from outliers : 14.2244 % Variance is moderately > inflated by > > >>> outliers > > >>> nil > > >>> user=> (bench (assoc [1 2 3 4 5 6 7 8 9 0] 3 nil 6 'a 0 "z" 2 5 4 > (atom > > >>> []) 7 "v")) ;;6 operations > > >>> Evaluation count : 39783840 in 60 samples of 663064 calls. > > >>> Execution time mean : 1.480511 us > > >>> Execution time std-deviation : 50.249443 ns > > >>> Execution time lower quantile : 1.425053 us ( 2.5%) > > >>> Execution time upper quantile : 1.591407 us (97.5%) > > >>> nil > > >>> user=> (bench (persistent! (assoc! (transient [1 2 3 4 5 6 7 8 9 0]) > 3 > > >>> nil 6 'a 0 "z" 2 5 4 (atom []) 7 "v"))) > > >>> Evaluation count : 44755020 in 60 samples of 745917 calls. > > >>> Execution time mean : 1.358394 us > > >>> Execution time std-deviation : 49.723547 ns > > >>> Execution time lower quantile : 1.306456 us ( 2.5%) > > >>> Execution time upper quantile : 1.473882 us (97.5%) > > >>> nil > > >>> > > >>> user=> (bench (assoc [1 2 3 4 5 6 7 8 9 0] 3 nil 6 'a 0 "z" 2 5 4 > (atom > > >>> []) 7 "v" 5 1200 8 -1111)) ;;8 operations > > >>> Evaluation count : 31457280 in 60 samples of 524288 calls. > > >>> Execution time mean : 1.907389 us > > >>> Execution time std-deviation : 74.523511 ns > > >>> Execution time lower quantile : 1.822336 us ( 2.5%) > > >>> Execution time upper quantile : 2.076044 us (97.5%) > > >>> nil > > >>> user=> (bench (persistent! (assoc! (transient [1 2 3 4 5 6 7 8 9 0]) > 3 > > >>> nil 6 'a 0 "z" 2 5 4 (atom []) 7 "v" 5 1200 8 -1111))) > > >>> Evaluation count : 37739640 in 60 samples of 628994 calls. > > >>> Execution time mean : 1.623599 us > > >>> Execution time std-deviation : 68.677518 ns > > >>> Execution time lower quantile : 1.529454 us ( 2.5%) > > >>> Execution time upper quantile : 1.777040 us (97.5%) > > >>> > > >>> Found 1 outliers in 60 samples (1.6667 %) > > >>> low-severe 1 (1.6667 %) > > >>> Variance from outliers : 28.6901 % Variance is moderately > inflated by > > >>> outliers > > >>> nil > > >>> > > >>> > > >>> etc etc... > > >>> > > >>> you get the idea...conclusion: there is no real benefit to using > > >>> transients for less than 6 operations... > > >>> > > >>> Jim > > >>> > > >>> > > >>> > > >>> > > >>> On 08/11/12 23:04, Philip Potter wrote: > > >>>> Share the benchmark! > > >>>> > > >>>> In theory, conversion to and from transients should be constant time > > >>>> operations, but it would not surprise me if in practice transients > > >>>> aren't universally faster than persistent. > > >>>> > > >>>> Performance in different environments varies wildly, so I would > doubt > > >>>> results from a small benchmark on a single machine are universally > > >>>> applicable. > > >>>> > > >>>> Benchmarking is hard, particularly on the jvm. Caching and dynamic > > >>>> inlining and Hotspot confuse matters greatly. > > >>>> > > >>>> Phil > > >>>> > > >>>> On Nov 8, 2012 10:49 PM, "Jim - FooBar();" <jimpil1...@gmail.com > > >>>> <mailto:jimpil1...@gmail.com>> wrote: > > >>>> > > >>>> Some quick benchmarking that I did, showed that it is actually > > >>>> more expensive to convert to transient, conj/assoc and convert > > >>>> back to persistent than the regular conj/assoc unless you want > to > > >>>> do more than 8 operations at a time (preferably more than > that). > > >>>> My experiments were on vectors and the vararg version of > assoc/assoc!. > > >>>> > > >>>> any thoughts? would you agree? > > >>>> > > >>>> Jim > > >>>> > > >>>> -- > > >>>> You received this message because you are subscribed to the > Google > > >>>> Groups "Clojure" group. > > >>>> To post to this group, send email to clojure@googlegroups.com > > >>>> <mailto:clojure@googlegroups.com> > > >>>> Note that posts from new members are moderated - please be > patient > > >>>> with your first post. > > >>>> To unsubscribe from this group, send email to > > >>>> clojure+unsubscr...@googlegroups.com > > >>>> <mailto:clojure%2bunsubscr...@googlegroups.com> > > >>>> For more options, visit this group at > > >>>> http://groups.google.com/group/clojure?hl=en > > >>>> > > >>>> -- > > >>>> You received this message because you are subscribed to the Google > > >>>> Groups "Clojure" group. > > >>>> To post to this group, send email to clojure@googlegroups.com > > >>>> Note that posts from new members are moderated - please be patient > > >>>> with your first post. > > >>>> To unsubscribe from this group, send email to > > >>>> clojure+unsubscr...@googlegroups.com > > >>>> For more options, visit this group at > > >>>> http://groups.google.com/group/clojure?hl=en > > >>> -- > > >>> You received this message because you are subscribed to the Google > > >>> Groups "Clojure" group. > > >>> To post to this group, send email to clojure@googlegroups.com > > >>> Note that posts from new members are moderated - please be patient > with your first post. > > >>> To unsubscribe from this group, send email to > > >>> clojure+unsubscr...@googlegroups.com > > >>> For more options, visit this group at > > >>> http://groups.google.com/group/clojure?hl=en > > >> -- > > >> Softaddicts<lprefonta...@softaddicts.ca> sent by ibisMail from my > ipad! > > >> > > >> -- > > >> You received this message because you are subscribed to the Google > > >> Groups "Clojure" group. > > >> To post to this group, send email to clojure@googlegroups.com > > >> Note that posts from new members are moderated - please be patient > with your first post. > > >> To unsubscribe from this group, send email to > > >> clojure+unsubscr...@googlegroups.com > > >> For more options, visit this group at > > >> http://groups.google.com/group/clojure?hl=en > > >> > > > -- > > > Softaddicts<lprefonta...@softaddicts.ca> sent by ibisMail from my > ipad! > > > > > > > -- > > You received this message because you are subscribed to the Google > > Groups "Clojure" group. > > To post to this group, send email to clojure@googlegroups.com > > Note that posts from new members are moderated - please be patient with > your first post. > > To unsubscribe from this group, send email to > > clojure+unsubscr...@googlegroups.com > > For more options, visit this group at > > http://groups.google.com/group/clojure?hl=en > > > -- > Softaddicts<lprefonta...@softaddicts.ca> sent by ibisMail from my ipad! > > -- > You received this message because you are subscribed to the Google > Groups "Clojure" group. > To post to this group, send email to clojure@googlegroups.com > Note that posts from new members are moderated - please be patient with > your first post. > To unsubscribe from this group, send email to > clojure+unsubscr...@googlegroups.com > For more options, visit this group at > http://groups.google.com/group/clojure?hl=en > -- You received this message because you are subscribed to the Google Groups "Clojure" group. To post to this group, send email to clojure@googlegroups.com Note that posts from new members are moderated - please be patient with your first post. To unsubscribe from this group, send email to clojure+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/clojure?hl=en
