This has been delayed because of a) the need for a real-world example to diffuse the possibility of critisism for using an artifical example; b) the difficulty in actually making the translated code work.

Starting with the full solution to Re^4: Restore the original order of an array after sort and performing some funchtion on array values, and using a dataset size per Re^6: Restore the original order of an array after sort and performing some funchtion on array values,

this code 4000: Took 1.369 seconds:

Standard Perl:

<Reveal this spoiler or all spoilers in this node or all in this thread>

Converted to use autobox::Core,

it requires 4000: Took 237.630 seconds:

(Partially) autoboxed code:

<Reveal this spoiler>

I get that to be a 99.5% slowdown.

Update: It was pointed out that the above figure is deceptive, in that what it shows is that the standard perl version took just 0.5% of the time that the autoboxed version took. Looked at the other way around, the autoboxed version took 17000% longer.

Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.
"Science is about questioning the status quo. Questioning authority".
In the absence of evidence, opinion is indistinguishable from prejudice.
"I'd rather go naked than blow up my ass"

Replies are listed 'Best First'.
Re: autobox performance:a real-world comparison
by chromatic (Archbishop) on Mar 05, 2010 at 05:02 UTC

    What an odd result. I get times around 1.87 seconds for the autobox version and 1.68 seconds for the non-autobox version (Perl 5.10.0 with ithreads). On a second machine, I get 4.82 seconds for autobox and 4.32 seconds for non-autobox (Perl 5.10.1 without ithreads). A 10% penalty seems much more reasonable.

[reply]

      Intriguing indeed. Especially as there is such close correspondance betwen our standard Perl runtimes, and such disparity with the autoboxed version. I'm at a loss ot explain it. (Could you try running with the command line options show below?).

      Perhaps a few others could run the code--with emphasis on AS & non-AS Perl's on Windows; 32-bit & 64-bit; threaded & non-threaded--though you seem to have ruled that out. Maybe they would shed some light.

      This is a direct, unedited c&p from my console running first the standard Perl version on 4000, and then the autoboxed version on 400-4000 showing the exponential growth in runtime: 

      C:\test>826456 -N=4000
4000: Took 1.367 seconds

C:\test>826456-a -N=400
400: Took 0.025 seconds

C:\test>826456-a -N=800
800: Took 0.988 seconds

C:\test>826456-a -N=1600
1600: Took 13.999 seconds

C:\test>826456-a -N=3200
3200: Took 119.228 seconds

C:\test>826456-a -N=4000
4000: Took 234.037 seconds

      [download]

      My system is 

      C:\test>perl -V
Summary of my perl5 (revision 5 version 10 subversion 1) configuration
+:

  Platform:
    osname=MSWin32, osvers=5.2, archname=MSWin32-x64-multi-thread
    uname=''
    config_args='undef'
    hint=recommended, useposix=true, d_sigaction=undef
    useithreads=define, usemultiplicity=define
    useperlio=define, d_sfio=undef, uselargefiles=define, usesocks=und
+ef
    use64bitint=define, use64bitall=undef, uselongdouble=undef
    usemymalloc=n, bincompat5005=undef
  Compiler:
    cc='cl', ccflags ='-nologo -GF -W3 -MD -Zi -DNDEBUG -Ox -GL -Wp64 
+-fp:precise -DWIN32 -D_CON
    optimize='-MD -Zi -DNDEBUG -Ox -GL -Wp64 -fp:precise',
    cppflags='-DWIN32'
    ccversion='15.0.21022', gccversion='', gccosandvers=''
    intsize=4, longsize=4, ptrsize=8, doublesize=8, byteorder=12345678
    d_longlong=undef, longlongsize=8, d_longdbl=define, longdblsize=8
    ivtype='__int64', ivsize=8, nvtype='double', nvsize=8, Off_t='__in
+t64', lseeksize=8
    alignbytes=8, prototype=define
  Linker and Libraries:
    ld='link', ldflags ='-nologo -nodefaultlib -debug -opt:ref,icf -lt
+cg  -libpath:"C:\Perl64\li
    libpth=\lib
    libs=  oldnames.lib kernel32.lib user32.lib gdi32.lib winspool.lib
+  comdlg32.lib advapi32.li
    perllibs=  oldnames.lib kernel32.lib user32.lib gdi32.lib winspool
+.lib  comdlg32.lib advapi3
    libc=msvcrt.lib, so=dll, useshrplib=true, libperl=perl510.lib
    gnulibc_version=''
  Dynamic Linking:
    dlsrc=dl_win32.xs, dlext=dll, d_dlsymun=undef, ccdlflags=' '
    cccdlflags=' ', lddlflags='-dll -nologo -nodefaultlib -debug -opt:
+ref,icf -ltcg  -libpath:"C


Characteristics of this binary (from libperl):
  Compile-time options: MULTIPLICITY PERL_DONT_CREATE_GVSV
                        PERL_IMPLICIT_CONTEXT PERL_IMPLICIT_SYS
                        PERL_MALLOC_WRAP PL_OP_SLAB_ALLOC USE_64_BIT_I
+NT
                        USE_ITHREADS USE_LARGE_FILES USE_PERLIO
                        USE_SITECUSTOMIZE
  Locally applied patches:
        ActivePerl Build 1006 [291086]
        32728 64-bit fix for Time::Local
  Built under MSWin32
  Compiled at Aug 24 2009 13:45:20
  @INC:
    C:/Perl64/site/lib
    C:/Perl64/lib
    .

C:\test>perl -Mautobox -E"say $autobox::VERSION"
2.55

C:\test>perl -Mautobox::Core -E"say $autobox::Core::VERSION"
0.6

      [download]

      Full console log including dumping sources to the screen:

      Read more... (3 kB)
      Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.
      "Science is about questioning the status quo. Questioning authority".
      In the absence of evidence, opinion is indistinguishable from prejudice.
[reply]
[d/l]
[select]
        My run of the test on perl-5.10.1 (linux-amd64) confirms cromatics' findings: 
        luben@karavelov:~/perl$ ./t -N=4000
4000: Took 2.062 seconds
luben@karavelov:~/perl$ ./tac -N=4000
4000: Took 2.538 seconds

        [download]
        And the perl version: 
        luben@karavelov:~/perl$ perl -V
Summary of my perl5 (revision 5 version 10 subversion 1) configuration
+:
   
  Platform:
    osname=linux, osvers=2.6.32-trunk-amd64, archname=x86_64-linux-gnu
+-thread-multi
    uname='linux madeleine 2.6.32-trunk-amd64 #1 smp sun jan 10 22:40:
+40 utc 2010 x86_64 gnulinux '
    config_args='-Dusethreads -Duselargefiles -Dccflags=-DDEBIAN -Dccc
+dlflags=-fPIC -Darchname=x86_64-linux-gnu -Dprefix=/usr -Dprivlib=/us
+r/share/perl/5.10 -Darchlib=/usr/lib/perl/5.10 -Dvendorprefix=/usr -D
+vendorlib=/usr/share/perl5 -Dvendorarch=/usr/lib/perl5 -Dsiteprefix=/
+usr/local -Dsitelib=/usr/local/share/perl/5.10.1 -Dsitearch=/usr/loca
+l/lib/perl/5.10.1 -Dman1dir=/usr/share/man/man1 -Dman3dir=/usr/share/
+man/man3 -Dsiteman1dir=/usr/local/man/man1 -Dsiteman3dir=/usr/local/m
+an/man3 -Dman1ext=1 -Dman3ext=3perl -Dpager=/usr/bin/sensible-pager -
+Uafs -Ud_csh -Ud_ualarm -Uusesfio -Uusenm -DDEBUGGING=-g -Doptimize=-
+O2 -Duseshrplib -Dlibperl=libperl.so.5.10.1 -Dd_dosuid -des'
    hint=recommended, useposix=true, d_sigaction=define
    useithreads=define, usemultiplicity=define
    useperlio=define, d_sfio=undef, uselargefiles=define, usesocks=und
+ef
    use64bitint=define, use64bitall=define, uselongdouble=undef
    usemymalloc=n, bincompat5005=undef
  ...
Characteristics of this binary (from libperl): 
  Compile-time options: MULTIPLICITY PERL_DONT_CREATE_GVSV
                        PERL_IMPLICIT_CONTEXT PERL_MALLOC_WRAP USE_64_
+BIT_ALL
                        USE_64_BIT_INT USE_ITHREADS USE_LARGE_FILES
                        USE_PERLIO USE_REENTRANT_API
...

        [download]
        The versions of autobox and autobox::Core are the same as yours. May be it is a problem with the windows version of perl.
[reply]
[d/l]
[select]

      The cause is memory growth! Despite that they both process identical datsets:

      • The perl version grabs 6.2MB of ram immediately, and then terminates.

        Memory without autobox

      • The autobixed version starts at about 40MB and then slowly grabs more and more memory until it finally finishes with over 300MB.

        Memory with autobox

      I'd be interested to hear if the autobox also grabs a heap of memory on your systems?

      I also instrumenting them both with NYTProf, but the output is so bewildering (and large) that I'm not able to make any sense of it. If there is anyone out there that wants to try and fix this, I'll make the NYTProf output, plus any other info I can supply, to them.

      Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.
      "Science is about questioning the status quo. Questioning authority".
      In the absence of evidence, opinion is indistinguishable from prejudice.
[reply]

      Hm. You weren't by any chance using version 0.7 when you ran your tests? With a few tweaks maybe?

      Because

      1. I posted my meditation Mar 04, 2010 at 20:33 GMT.
      2. Version 0.7 was apparently uploaded on 4th March.
      3. You posted your reply Mar 05, 2010 at 05:02 GMT

      My hat's off to Scott Walters for such a fast response. Especially on a module he hadn't had occasion to touch for two years.

      Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.
      "Science is about questioning the status quo. Questioning authority".
      In the absence of evidence, opinion is indistinguishable from prejudice.
[reply]
Re: autobox performance:a real-world comparison
by JavaFan (Canon) on Mar 05, 2010 at 11:41 UTC
    I get autobox to be slower as well, but not that much slower. Using a wrapper program to run your programs 10 times, and calculating a confidence interval, I get: 
    standard: mean = 3.881; 90% confidence interval: 3.809-3.952
 autobox: mean = 4.915; 90% confidence interval: 4.365-5.465

    [download]
    Another run, now with 95% confidence: 
    standard: mean = 4.100; 95% confidence interval: 3.866-4.335
 autobox: mean = 4.695; 95% confidence interval: 4.402-4.989

    [download]
[reply]
[d/l]
[select]
Re: autobox performance:a real-world comparison
by syphilis (Archbishop) on Mar 07, 2010 at 02:23 UTC
    For me, using MinGW-built perl-5.10.1 (32 bit):
    Without autobox::Core took 5.188 seconds
    With autobox::Core took 203.469 seconds

    Using MinGW-built perl-5.10.1 (64 bit):
    Without autobox::Core took 6.203 seconds
    With autobox::Core took 486.675 seconds

    Using ActivePerl-5.10.1 (64-bit) build 1007:
    Without autobox::Core took 6.563 seconds
    With autobox::Core took 489.123 seconds

    As you can see, it's a fairly slow box - which may be at least partly due to some dumb configuring of Vista64. Of course the main thing we're looking at is the relative speeds, and autobox::Core is just plain woeful on Windows - roughly twice as woeful on 64-bit perls.

    Cheers,
    Rob
[reply]

      Thankyou. At least I now know that it isn't something peculiar to just my machine.

      The fact that this is affecting all three of your builds 32&64 bit; MSC & MinGW (with presumably different C-runtimes) makes me think that this isn't something completely outside of autobox--like a slow memory allocator--but rather some particular assumption within autobox that is true under Linux and not under Win.

      Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.
      "Science is about questioning the status quo. Questioning authority".
      In the absence of evidence, opinion is indistinguishable from prejudice.
[reply]

Back to Meditations