Re: Oh, yeah, transients are fast!
Here are my results. The transient version performs slightly better on windows 32bit client JVM, and considerably better on linux 64bit server JVM (they are both 1.06.0_17) -- 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
Re: Oh, yeah, transients are fast!
I believe that this is most likely a symptom of the Apple JVM and not that of transients as the change from persistents to transients is far more substantial on one of our linux servers than it is on my macbook pro (6x vs 2x speedup) I'm not entirely sure as to why this is the case but I suspect that GC has something to do with it as the following example, which simply bumps a counter in a map, exhibits the more impressive performance improvement (defn make-n [x n] (loop [i 0 v []] (if ( i n) (recur (inc i) (conj v x)) v))) (defn make-n2 [x n] (loop [i (int 0) v (transient [])] (if ( i n) (recur (inc i) (conj! v x)) (persistent! v (dotimes [x 10] (do (System/gc) (time (make-n 1 10)) 1)) (dotimes [x 10] (do (System/gc) (time (make-n2 1 10)) 1)) On Nov 23, 9:07 pm, Raoul Duke rao...@gmail.com wrote: i have tried: 1.5 1.6 1.6 -server the last i did both in repl-in-emacs, and in a repl-in-straightup-shell. the numbers i get are weird. it does seem like v2 is faster than v, but never gets stupendously fast (never faster than 500 msec on a dual core macbook pro 2.2ghz core 2 duo 4gb ram), and v does a strange thing where i run v2 a bunch then the first time i run v it is like 2x v2, and then subsequent runs of timing v are like 8x v2. and here i thought hotspot was supposed to get faster, not slower, over time. oy veh. -- 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
Re: Oh, yeah, transients are fast!
I believe that this is most likely a symptom of the Apple JVM and not yeah, given Apple's wonderful treatment of Java over the years, i could believe your theory. -- 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
Re: Oh, yeah, transients are fast!
I believe that this is most likely a symptom of the Apple JVM and not that of transients as the change from persistents to transients is far more substantial on one of our linux servers than it is on my macbook pro (6x vs 2x speedup) Not necessarily so. My times earlier in this thread (4GB Core 2 Duo MacBook Pro, Apple's JDK 6, Regular Leopard) showed a 10x speedup. Most likely it's related to the JVM version (1.5 by default), the amount of memory allocated (I use -Xmx512m), the amount of L2 cache, a HotSpot tuning parameter, or something else along those lines. There are far too many possibilities to consider, and too little evidence, to support a conclusion of it's just Apple's JVM. -- 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
Re: Oh, yeah, transients are fast!
(i love how supposedly we've come so far with our systems, only to have them become overly complex. it is to sigh. on the other hand i guess it is 'job security'.) Heh, true. Reading articles about JVM tuning reminds me what the M stands for -- it's as complicated a topic as optimizing the interface between your code and a real machine. E.g., http://blogs.sun.com/bharath/entry/taming_gc_for_sip_applications http://blog.headius.com/2009/01/my-favorite-hotspot-jvm-flags.html Your point about job security is spot on, though -- every time the topics of VHLLs, autonomic systems, and so on arise, I gently point out that somebody needs to figure out what's going on when they stop working, and the smarter they get, the more fiendish the problems! -- 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
Re: Oh, yeah, transients are fast!
i have tried: 1.5 1.6 1.6 -server the last i did both in repl-in-emacs, and in a repl-in-straightup-shell. the numbers i get are weird. it does seem like v2 is faster than v, but never gets stupendously fast (never faster than 500 msec on a dual core macbook pro 2.2ghz core 2 duo 4gb ram), and v does a strange thing where i run v2 a bunch then the first time i run v it is like 2x v2, and then subsequent runs of timing v are like 8x v2. and here i thought hotspot was supposed to get faster, not slower, over time. oy veh. -- 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
Re: Oh, yeah, transients are fast!
2009/11/21 John Harrop jharrop...@gmail.com: On Fri, Nov 20, 2009 at 5:16 PM, Raoul Duke rao...@gmail.com wrote: Try with a 1.6 JVM... wow. it actually got worse than when i was using 1.5. ... so much for hallowed write-once-run-the-same-anywhere-ish of the jvm, d'oh. Clojure 1.1.0-alpha-SNAPSHOT user= (load-file /tmp/test.clj) #'user/vrange2 user= (. System getProperty java.version) 1.6.0_15 user= (time (def v2 (vrange2 100))) Elapsed time: 329.906 msecs #'user/v2 user= (time (def v2 (vrange2 100))) Elapsed time: 888.734 msecs #'user/v2 user= (time (def v2 (vrange2 100))) Elapsed time: 546.982 msecs #'user/v2 user= (time (def v2 (vrange2 100))) Elapsed time: 517.969 msecs #'user/v2 user= (time (def v (vrange 100))) Elapsed time: 1649.449 msecs #'user/v user= (time (def v (vrange 100))) --- never came back, i had to ^C^C (in emacs buffer)!! You've got some kind of system problem confounding your results, I'll bet. It got slower? One test actually hung? My suspicion, of course, lies with the emacs environment you've just confessed to using. Half the traffic on this list at times is from people having problems with configuring either emacs or vi to work well with Clojure. Try it in a modern IDE, or from the command line, and see if the problem goes away. If it does, then you know the likely culprit and it isn't Clojure's transients. :) To be fair, John, you've had an issue or two with the REPL provided by your modern IDE too :) Raoul, are you using the server VM? Not sure how much of a difference it makes, and I haven't benchmarked this myself. -- Michael Wood esiot...@gmail.com -- 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
Re: Oh, yeah, transients are fast!
On Sun, Nov 22, 2009 at 6:05 AM, Michael Wood esiot...@gmail.com wrote: You've got some kind of system problem confounding your results, I'll bet. It got slower? One test actually hung? My suspicion, of course, lies with the emacs environment you've just confessed to using. Half the traffic on this list at times is from people having problems with configuring either emacs or vi to work well with Clojure. Try it in a modern IDE, or from the command line, and see if the problem goes away. If it does, then you know the likely culprit and it isn't Clojure's transients. :) To be fair, John, you've had an issue or two with the REPL provided by your modern IDE too :) Nothing so serious as a hang, though, and at least I can do basic things in my IDE without reaching for the frelling manual every two minutes to look up some key-combination :) -- 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
Re: Oh, yeah, transients are fast!
On Sun, Nov 22, 2009 at 11:45:46AM -0500, John Harrop wrote: Nothing so serious as a hang, though, and at least I can do basic things in my IDE without reaching for the frelling manual every two minutes to look up some key-combination :) I suspect both models are going to be important. I feel the same way you do when I'm running an IDE rather than vim. Oddly enough, I've found myself most productive with a repl in it's own window, and a short script (below) in user.clj to let me specify a namespace to test. I put this in a 'repl' directory that I only add to my classpath for the repl script. David (println Running user startup) (use 'clojure.contrib.repl-utils) (use 'clojure.contrib.str-utils) (use 'clojure.contrib.java-utils) (use 'clojure.contrib.def) (use 'clojure.contrib.ns-utils) (use 'clojure.contrib.test-is) (use 'clojure.contrib.sql) (use 'clojure.stacktrace) (set! *warn-on-reflection* true) ;;; The NS that we are testing/working on. ;;; This can be given through the java property work-ns, e.g. ;;; ./run.sh -Dwork-ns=org.davidb.chunk-file (def *work-ns* (symbol (or (System/getProperty work-ns) 'org.davidb.foo))) ;;; A testing NS. (def *test-ns* (symbol (or (System/getProperty test-ns) (symbol (re-sub #[^\.]+$ #(str test- %) (name *work-ns*)) ; (use *work-ns*) (defn l [] (binding [*warn-on-reflection* true] (use *work-ns* :reload))) (defn t [] (binding [*warn-on-reflection* true] (require *test-ns* :reload) (run-tests *test-ns*))) (defn i [] (in-ns *work-ns*)) ; Load the User's test-user.clj if present. (try (load /test-user) (catch java.io.FileNotFoundException e (println No test-user.clj found in classpath.))) -- 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
Re: Oh, yeah, transients are fast!
Try with a 1.6 JVM... wow. it actually got worse than when i was using 1.5. ... so much for hallowed write-once-run-the-same-anywhere-ish of the jvm, d'oh. Clojure 1.1.0-alpha-SNAPSHOT user= (load-file /tmp/test.clj) #'user/vrange2 user= (. System getProperty java.version) 1.6.0_15 user= (time (def v2 (vrange2 100))) Elapsed time: 329.906 msecs #'user/v2 user= (time (def v2 (vrange2 100))) Elapsed time: 888.734 msecs #'user/v2 user= (time (def v2 (vrange2 100))) Elapsed time: 546.982 msecs #'user/v2 user= (time (def v2 (vrange2 100))) Elapsed time: 517.969 msecs #'user/v2 user= (time (def v (vrange 100))) Elapsed time: 1649.449 msecs #'user/v user= (time (def v (vrange 100))) --- never came back, i had to ^C^C (in emacs buffer)!! -- 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
Re: Oh, yeah, transients are fast!
user= (time (def v (vrange 100))) --- never came back, i had to ^C^C (in emacs buffer)!! p.s. so at least the transient version never did that ;-) -- 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
Re: Oh, yeah, transients are fast!
On Fri, Nov 20, 2009 at 5:16 PM, Raoul Duke rao...@gmail.com wrote: Try with a 1.6 JVM... wow. it actually got worse than when i was using 1.5. ... so much for hallowed write-once-run-the-same-anywhere-ish of the jvm, d'oh. Clojure 1.1.0-alpha-SNAPSHOT user= (load-file /tmp/test.clj) #'user/vrange2 user= (. System getProperty java.version) 1.6.0_15 user= (time (def v2 (vrange2 100))) Elapsed time: 329.906 msecs #'user/v2 user= (time (def v2 (vrange2 100))) Elapsed time: 888.734 msecs #'user/v2 user= (time (def v2 (vrange2 100))) Elapsed time: 546.982 msecs #'user/v2 user= (time (def v2 (vrange2 100))) Elapsed time: 517.969 msecs #'user/v2 user= (time (def v (vrange 100))) Elapsed time: 1649.449 msecs #'user/v user= (time (def v (vrange 100))) --- never came back, i had to ^C^C (in emacs buffer)!! You've got some kind of system problem confounding your results, I'll bet. It got slower? One test actually hung? My suspicion, of course, lies with the emacs environment you've just confessed to using. Half the traffic on this list at times is from people having problems with configuring either emacs or vi to work well with Clojure. Try it in a modern IDE, or from the command line, and see if the problem goes away. If it does, then you know the likely culprit and it isn't Clojure's transients. :) -- 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
Oh, yeah, transients are fast!
hm. i copied and pasted the examples from the transient page into a local test.clj, and i don't really see a significant performance difference (it should be an order of magnitude according to the web page). macbook pro. just downloaded sources and built clojure-1.1.0-alpha-SNAPSHOT.jar for this. is there something about having to have the jit warm up or something over a zillion runs? help? :-) user= (load-file /tmp/test.clj) #'user/vrange2 user= (time (def v2 (vrange2 100))) Elapsed time: 502.842 msecs #'user/v2 user= (time (def v2 (vrange2 100))) Elapsed time: 407.568 msecs #'user/v2 user= (time (def v2 (vrange2 100))) Elapsed time: 364.29 msecs #'user/v2 user= (time (def v (vrange 100))) Elapsed time: 558.235 msecs user= (. System getProperty java.version) 1.5.0_20 -- 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
Re: Oh, yeah, transients are fast!
clojure-1.1.0-alpha-SNAPSHOT.jar for this. is there something about having to have the jit warm up or something over a zillion runs? I see an immediate speed improvement from transients with a recent Clojure, and after a few runs the difference is obvious. user= (time (def v (vrange 100))) Elapsed time: 342.455 msecs user= (time (def v2 (vrange2 100))) Elapsed time: 39.054 msecs Often vrange takes 800+msec, even after a number of runs. I'm using 1.6.0_15, -server. Try with a 1.6 JVM... -- 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
