While doing some benchmark testing on both Windows and Linux, the
results of the exact same code was reversed. A slight difference in
the percentages is understandable, but I fail to see why the results
would be reversed. Could someone shed some light on this issue?

First the benchmark results:

C:\TEMP>timestamp.pl
Rate Matt Ron
Matt 162840/s -- -37%
Ron 257003/s 58% --


[root [at] 099vicidial101 ~]# ./timestamp.pl
Rate Ron Matt
Ron 110132/s -- -29%
Matt 155763/s 41% --


The code:

#!/usr/bin/perl

use strict;
use warnings;
use POSIX qw(strftime);
use Benchmark qw(:all);

my $count = 1_000_000;

cmpthese($count, {
Matt => \&matt,
Ron => \&ron,
});


sub matt {
my $now_date_epoch = time();
my $BDtarget = ($now_date_epoch - 5);
my ($Bsec,$Bmin,$Bhour,$Bmday,$Bmon,$Byear,$Bwday,$Byday,$Bisds t)
= localtime($BDtarget);
$Byear = ($Byear + 1900);
$Bmon++;
if ($Bmon < 10) {$Bmon = "0$Bmon";}
if ($Bmday < 10) {$Bmday = "0$Bmday";}
if ($Bhour < 10) {$Bhour = "0$Bhour";}
if ($Bmin < 10) {$Bmin = "0$Bmin";}
if ($Bsec < 10) {$Bsec = "0$Bsec";}
my $BDtsSQLdate = "$Byear$Bmon$Bmday$Bhour$Bmin$Bsec";
}

sub ron {
my $BDtsSQLdate = strftime("%Y%m%d%H%M%S", localtime(time() -
5) );
}


--
To unsubscribe, e-mail: beginners-unsubscribe [at] perl.org
For additional commands, e-mail: beginners-help [at] perl.org
http://learn.perl.org/

On 10-08-10 11:43 AM, Ron Bergin wrote:
> While doing some benchmark testing on both Windows and Linux, the
> results of the exact same code was reversed. A slight difference in
> the percentages is understandable, but I fail to see why the results
> would be reversed. Could someone shed some light on this issue?

Maybe because strftime() is part of *NIX?


--
Just my 0.00000002 million dollars worth,
Shawn

Programming is as much about organization and communication
as it is about coding.

The secret to great software: Fail early & often.

Eliminate software piracy: use only FLOSS.

--
To unsubscribe, e-mail: beginners-unsubscribe [at] perl.org
For additional commands, e-mail: beginners-help [at] perl.org
http://learn.perl.org/

Shawn H Corey wrote:
> On 10-08-10 11:43 AM, Ron Bergin wrote:
>> While doing some benchmark testing on both Windows and
>> Linux, the
>> results of the exact same code was reversed. A slight
>> difference in
>> the percentages is understandable, but I fail to see why
>> the results
>> would be reversed. Could someone shed some light on
>> this issue?
>
> Maybe because strftime() is part of *NIX?
>

Can you expand on what you mean by that statement?





--
To unsubscribe, e-mail: beginners-unsubscribe [at] perl.org
For additional commands, e-mail: beginners-help [at] perl.org
http://learn.perl.org/

On Tue, Aug 10, 2010 at 11:43, Ron Bergin <rkb [at] i.frys.com> wrote:
> While doing some benchmark testing on both Windows and Linux, the
> results of the exact same code was reversed. =C2=A0A slight difference in
> the percentages is understandable, but I fail to see why the results
> would be reversed. =C2=A0Could someone shed some light on this issue?
snip
> =C2=A0 =C2=A0my $BDtsSQLdate =3D "$Byear$Bmon$Bmday$Bhour$Bmin$Bsec";
snip

I seem to recall reading something on p5p recently about string
concatenations being slower on Win32 machines for some reason. Since
that line is really six string concatenations, you may be better off
using join:

my $BDtsSQLdate =3D join "", $Byear, $Bmon, $Bmday, $Bhour, $Bmin, $Bsec;

You might also try profiling the slow subroutine with Devel::NYTProf
to find what is slower between Linux and Win32.



--
Chas. Owens
wonkden.net
The most important skill a programmer can have is the ability to read.

--
To unsubscribe, e-mail: beginners-unsubscribe [at] perl.org
For additional commands, e-mail: beginners-help [at] perl.org
http://learn.perl.org/

On Tue, Aug 10, 2010 at 11:53, Shawn H Corey <shawnhcorey [at] gmail.com> wrote:
> On 10-08-10 11:43 AM, Ron Bergin wrote:
>>
>> While doing some benchmark testing on both Windows and Linux, the
>> results of the exact same code was reversed. =C2=A0A slight difference i=
n
>> the percentages is understandable, but I fail to see why the results
>> would be reversed. =C2=A0Could someone shed some light on this issue?
>
> Maybe because strftime() is part of *NIX?
snip

Unlikely the problem. The strftime using function is faster on Win32
(it would really help if the names given were descriptive, not
people's names).


--
Chas. Owens
wonkden.net
The most important skill a programmer can have is the ability to read.

--
To unsubscribe, e-mail: beginners-unsubscribe [at] perl.org
For additional commands, e-mail: beginners-help [at] perl.org
http://learn.perl.org/

On Tue, Aug 10, 2010 at 12:12, <rkb [at] i.frys.com> wrote:
> Shawn H Corey wrote:
>> On 10-08-10 11:43 AM, Ron Bergin wrote:
>>> While doing some benchmark testing on both Windows and
>>> Linux, the
>>> results of the exact same code was reversed. =C2=A0A slight
>>> difference in
>>> the percentages is understandable, but I fail to see why
>>> the results
>>> would be reversed. =C2=A0Could someone shed some light on
>>> this issue?
>>
>> Maybe because strftime() is part of *NIX?
>>
>
> Can you expand on what you mean by that statement?
snip

I am assuming that his assumption was that since strftime is part of
POSIX/SUS (i.e. Unix specifications) Win32 might have a slower
implementation; however, Windows NT and OSes based on Windows NT
implemented most of the POSIX spec, so that is unlikely part of the
problem.

--
Chas. Owens
wonkden.net
The most important skill a programmer can have is the ability to read.

--
To unsubscribe, e-mail: beginners-unsubscribe [at] perl.org
For additional commands, e-mail: beginners-help [at] perl.org
http://learn.perl.org/

Chas. Owens wrote:
> On Tue, Aug 10, 2010 at 11:43, Ron Bergin <rkb [at] i.frys.com>
> wrote:
> snip
>>    my $BDtsSQLdate =
>> "$Byear$Bmon$Bmday$Bhour$Bmin$Bsec";
> snip
>
> I seem to recall reading something on p5p recently about
> string
> concatenations being slower on Win32 machines for some
> reason. Since
> that line is really six string concatenations, you may be
> better off
> using join:
>
> my $BDtsSQLdate = join "", $Byear, $Bmon, $Bmday, $Bhour,
> $Bmin, $Bsec;

That didn't have any appreciable difference.

>
> You might also try profiling the slow subroutine with
> Devel::NYTProf
> to find what is slower between Linux and Win32.
>

That sounds like a good idea.


--
To unsubscribe, e-mail: beginners-unsubscribe [at] perl.org
For additional commands, e-mail: beginners-help [at] perl.org
http://learn.perl.org/

On Aug 10, 8:43=A0am, r... [at] i.frys.com (Ron Bergin) wrote:
> While doing some benchmark testing on both Windows and Linux, the
> results of the exact same code was reversed. =A0A slight difference in
> the percentages is understandable, but I fail to see why the results
> would be reversed. =A0Could someone shed some light on this issue?
>
> First the benchmark results:
>
> C:\TEMP>timestamp.pl
> =A0 =A0 =A0 =A0 =A0Rate Matt =A0Ron
> Matt 162840/s =A0 -- -37%
> Ron =A0257003/s =A058% =A0 --
>
> [root [at] 099vicidial101 ~]# ./timestamp.pl
> =A0 =A0 =A0 =A0 =A0Rate =A0Ron Matt
> Ron =A0110132/s =A0 -- -29%
> Matt 155763/s =A041% =A0 --
>
> The code:
>
> #!/usr/bin/perl
>
> use strict;
> use warnings;
> use POSIX qw(strftime);
> use Benchmark qw(:all);
>
> my $count =3D 1_000_000;
>
> cmpthese($count, {
> =A0 =A0 Matt =3D> \&matt,
> =A0 =A0 Ron =A0=3D> \&ron,
>
> });
>
> sub matt {
> =A0 =A0 my $now_date_epoch =3D time();
> =A0 =A0 my $BDtarget =3D ($now_date_epoch - 5);
> =A0 =A0 my ($Bsec,$Bmin,$Bhour,$Bmday,$Bmon,$Byear,$Bwday,$Byday,$Bisds t)
> =3D localtime($BDtarget);
> =A0 =A0 $Byear =3D ($Byear + 1900);
> =A0 =A0 $Bmon++;
> =A0 =A0 if ($Bmon < 10) {$Bmon =3D "0$Bmon";}
> =A0 =A0 if ($Bmday < 10) {$Bmday =3D "0$Bmday";}
> =A0 =A0 if ($Bhour < 10) {$Bhour =3D "0$Bhour";}
> =A0 =A0 if ($Bmin < 10) {$Bmin =3D "0$Bmin";}
> =A0 =A0 if ($Bsec < 10) {$Bsec =3D "0$Bsec";}
> =A0 =A0 my $BDtsSQLdate =3D "$Byear$Bmon$Bmday$Bhour$Bmin$Bsec";
>
> }
>
> sub ron {
> =A0 =A0 my $BDtsSQLdate =3D strftime("%Y%m%d%H%M%S", localtime(time() -
> 5) );
>
> }

I think there was also a thread about string concatenation
being horrifically slow on comp.lang.perl.misc.

sprintf provides a substantial speedup...maybe less
malloc'ing on Win32. Fewer perl op's probably help
too:

sub matt {
my $now_date_epoch =3D time();
my $BDtarget =3D ($now_date_epoch - 5);
my ($Bsec,$Bmin,$Bhour,$Bmday,$Bmon,$Byear) =3D
localtime($BDtarget);
$Byear =3D ($Byear + 1900);
$Bmon++;

my $BDtsSQLdate =3D sprintf "%4d%02d%02d%02d%02d%02d",
$Byear,$Bmon,$Bmday, $Bhour,$Bmin, $Bsec;
}

Rate Matt Ron
Matt 323729/s -- -13%
Ron 372717/s 15% --

Matt 323834/s -- -13%
Ron 370508/s 14% --

Matt 323834/s -- -13%
Ron 370508/s 14% --

Matt 328731/s -- -11%
Ron 370645/s 13% --

--
Charles DeRykus


--
To unsubscribe, e-mail: beginners-unsubscribe [at] perl.org
For additional commands, e-mail: beginners-help [at] perl.org
http://learn.perl.org/

C.DeRykus wrote:
> On Aug 10, 8:43 am, r... [at] i.frys.com (Ron Bergin) wrote:
[snip]
>>
>> sub matt {
>>     my $now_date_epoch = time();
>>     my $BDtarget = ($now_date_epoch - 5);
>>     my
>> ($Bsec,$Bmin,$Bhour,$Bmday,$Bmon,$Byear,$Bwday,$Byday,$Bisds t)
>> = localtime($BDtarget);
>>     $Byear = ($Byear + 1900);
>>     $Bmon++;
>>     if ($Bmon < 10) {$Bmon = "0$Bmon";}
>>     if ($Bmday < 10) {$Bmday = "0$Bmday";}
>>     if ($Bhour < 10) {$Bhour = "0$Bhour";}
>>     if ($Bmin < 10) {$Bmin = "0$Bmin";}
>>     if ($Bsec < 10) {$Bsec = "0$Bsec";}
>>     my $BDtsSQLdate =
>> "$Byear$Bmon$Bmday$Bhour$Bmin$Bsec";
>>
>> }
>>
>> sub ron {
>>     my $BDtsSQLdate = strftime("%Y%m%d%H%M%S",
>> localtime(time() -
>> 5) );
>>
>> }
>
> I think there was also a thread about string concatenation
> being horrifically slow on comp.lang.perl.misc.
>
> sprintf provides a substantial speedup...maybe less
> malloc'ing on Win32. Fewer perl op's probably help
> too:
>
> sub matt {
> my $now_date_epoch = time();
> my $BDtarget = ($now_date_epoch - 5);
> my ($Bsec,$Bmin,$Bhour,$Bmday,$Bmon,$Byear) =
> localtime($BDtarget);
> $Byear = ($Byear + 1900);
> $Bmon++;
>
> my $BDtsSQLdate = sprintf "%4d%02d%02d%02d%02d%02d",
> $Byear,$Bmon,$Bmday, $Bhour,$Bmin, $Bsec;
> }
>
> Rate Matt Ron
> Matt 323729/s -- -13%
> Ron 372717/s 15% --
>
> Matt 323834/s -- -13%
> Ron 370508/s 14% --
>
> Matt 323834/s -- -13%
> Ron 370508/s 14% --
>
> Matt 328731/s -- -11%
> Ron 370645/s 13% --
>
> --
> Charles DeRykus
>

Did you run that on Windows or Linux? I tested it on both
and my results would indicate that the sprintf version is
slower than concatenation and the results were still
flipped between OS's.

I probably should have mentioned that the "Matt" code is
what is currently being used in production and I need to
profile/benchmark it against different approaches.

I haven't profiled it as Chas suggested, but I suspect the
issue is due to the difference in perl versions.

5.8.8 on Linux
5.10.0 on Windows.

My theory is that 5.10.x implemented some optimizations
that improved the speed of strftime.

Ron


--
To unsubscribe, e-mail: beginners-unsubscribe [at] perl.org
For additional commands, e-mail: beginners-help [at] perl.org
http://learn.perl.org/

On Wed, Aug 11, 2010 at 08:39, <rkb [at] i.frys.com> wrote:
snip
> I haven't profiled it as Chas suggested, but I suspect the
> issue is due to the difference in perl versions.
>
> 5.8.8 on Linux
> 5.10.0 on Windows.
>
> My theory is that 5.10.x implemented some optimizations
> that improved the speed of strftime.
snip

Unlikely, I don't remember any big changes to strftime (it is
implemented by Perl_my_strftime in [util.c][0] if you want to check).
I ran your code against 5.12.1, 5.10.0, and 5.8.9 on OS X and got

5.12.1
Rate strftime manual
strftime 69591/s -- -67%
manual 212384/s 205% --

5.10.0
Rate strftime manual
strftime 63621/s -- -69%
manual 203558/s 220% --

5.8.9
Rate strftime manual
strftime 62421/s -- -65%
manual 178641/s 186% --

[0]: http://perl5.git.perl.org/perl.git/history/HEAD:/util.c

--
Chas. Owens
wonkden.net
The most important skill a programmer can have is the ability to read.

--
To unsubscribe, e-mail: beginners-unsubscribe [at] perl.org
For additional commands, e-mail: beginners-help [at] perl.org
http://learn.perl.org/

On Wed, Aug 11, 2010 at 08:39, =C2=A0<rkb [at] i.frys.com> wrote:
snip
> I probably should have mentioned that the "Matt" code is
> what is currently being used in production and I need to
> profile/benchmark it against different approaches.
snip

Just getting rid of the stupid in your production code increases the
speed by between forty and fifty percent (at least on OS X). It also
drops the line count from twelve to three.

#!/usr/bin/perl

use strict;
use warnings;
use POSIX qw(strftime);
use Benchmark qw(:all);

cmpthese(-1, {
production =3D> \&production,
experimental =3D> \&experimental,
});


sub production {
my $now_date_epoch =3D time();
my $BDtarget =3D ($now_date_epoch - 5);
my ($Bsec,$Bmin,$Bhour,$Bmday,$Bmon,$Byear,$Bwday,$Byday,$Bisds t)
=3D localtime($BDtarget);
$Byear =3D ($Byear + 1900);
$Bmon++;
if ($Bmon < 10) {$Bmon =3D "0$Bmon";}
if ($Bmday < 10) {$Bmday =3D "0$Bmday";}
if ($Bhour < 10) {$Bhour =3D "0$Bhour";}
if ($Bmin < 10) {$Bmin =3D "0$Bmin";}
if ($Bsec < 10) {$Bsec =3D "0$Bsec";}
my $BDtsSQLdate =3D "$Byear$Bmon$Bmday$Bhour$Bmin$Bsec";
}

sub experimental {
my ($s, $min, $h, $d, $mon, $y) =3D localtime(time - 5);
my $BDtsSQLdate =3D sprintf "%d%02d%02d%02d%02d%02d", $y + 1900,
$mon + 1, $d, $h, $min, $s;
}



--
Chas. Owens
wonkden.net
The most important skill a programmer can have is the ability to read.

--
To unsubscribe, e-mail: beginners-unsubscribe [at] perl.org
For additional commands, e-mail: beginners-help [at] perl.org
http://learn.perl.org/

Chas. Owens wrote:
> On Wed, Aug 11, 2010 at 08:39, <rkb [at] i.frys.com> wrote:
> snip
>> I haven't profiled it as Chas suggested, but I suspect
>> the
>> issue is due to the difference in perl versions.
>>
>> 5.8.8 on Linux
>> 5.10.0 on Windows.
>>
>> My theory is that 5.10.x implemented some optimizations
>> that improved the speed of strftime.
> snip
>
> Unlikely, I don't remember any big changes to strftime (it
> is
> implemented by Perl_my_strftime in [util.c][0] if you want
> to check).
> I ran your code against 5.12.1, 5.10.0, and 5.8.9 on OS X
> and got
>
> 5.12.1
> Rate strftime manual
> strftime 69591/s -- -67%
> manual 212384/s 205% --
>
> 5.10.0
> Rate strftime manual
> strftime 63621/s -- -69%
> manual 203558/s 220% --
>
> 5.8.9
> Rate strftime manual
> strftime 62421/s -- -65%
> manual 178641/s 186% --
>
> [0]:
> http://perl5.git.perl.org/perl.git/history/HEAD:/util.c
>
> --
> Chas. Owens

That's interesting, and quite a bit of difference from my
runs.


--
To unsubscribe, e-mail: beginners-unsubscribe [at] perl.org
For additional commands, e-mail: beginners-help [at] perl.org
http://learn.perl.org/

On Wed, Aug 11, 2010 at 09:28, Chas. Owens <chas.owens [at] gmail.com> wrote:
> On Wed, Aug 11, 2010 at 08:39, =C2=A0<rkb [at] i.frys.com> wrote:
> snip
>> I haven't profiled it as Chas suggested, but I suspect the
>> issue is due to the difference in perl versions.
>>
>> 5.8.8 on Linux
>> 5.10.0 on Windows.
>>
>> My theory is that 5.10.x implemented some optimizations
>> that improved the speed of strftime.
> snip
>
> Unlikely, I don't remember any big changes to strftime (it is
> implemented by Perl_my_strftime in [util.c][0] if you want to check).
snip

I think what you are seeing is the penalty that strftime takes for
normalizing the data handed to it. For instance, this code prints
today's date and the date 60 days in the future.

#!/usr/bin/perl

use strict;
use warnings;

use POSIX qw/strftime/;

my [at] t =3D (0) x 3;
my ($d, $m, $y) =3D (localtime)[3..5];
print
"now: ", strftime("%Y-%m-%d", [at] t, $d, $m, $y), "\n",
"60 days from now: ", strftime("%Y-%m-%d", [at] t, $d+60, $m, $y), "\n";

We can also see this penalty in the C version of the function:

#!/usr/bin/perl

use strict;
use warnings;
use Benchmark qw(:all);
use Inline "C";

print map { "$_\n" } production(), experimental(), ftime(), manual();
cmpthese(-1, {
production =3D> \&production,
experimental =3D> \&experimental,
inline_strftime =3D> \&ftime,
inline_manual =3D> \&manual,
});

sub production {
my $now_date_epoch =3D time();
my $BDtarget =3D ($now_date_epoch - 5);
my ($Bsec,$Bmin,$Bhour,$Bmday,$Bmon,$Byear,$Bwday,$Byday,$Bisds t)
=3D localtime($BDtarget);
$Byear =3D ($Byear + 1900);
$Bmon++;
if ($Bmon < 10) {$Bmon =3D "0$Bmon";}
if ($Bmday < 10) {$Bmday =3D "0$Bmday";}
if ($Bhour < 10) {$Bhour =3D "0$Bhour";}
if ($Bmin < 10) {$Bmin =3D "0$Bmin";}
if ($Bsec < 10) {$Bsec =3D "0$Bsec";}
my $BDtsSQLdate =3D "$Byear$Bmon$Bmday$Bhour$Bmin$Bsec";
}

sub experimental {
my ($s, $min, $h, $d, $mon, $y) =3D localtime(time - 5);
my $BDtsSQLdate =3D sprintf "%d%02d%02d%02d%02d%02d", $y + 1900,
$mon + 1, $d, $h, $min, $s;
}

__DATA__
__C__

#include <time.h>
#include <stdio.h>

char* ftime() {
static char ret[15];
time_t now =3D time(NULL) - 5;
strftime(ret, 14, "%Y%m%d%H%M%S", localtime(&now));
return ret;
}

char* manual() {
static char ret[15];
time_t now =3D time(NULL) - 5;
struct tm *d =3D localtime(&now);

snprintf(ret, 15, "%d%02d%02d%02d%02d%02d", d->tm_year + 1900,
d->tm_mon + 1, d->tm_mday, d->tm_hour, d->tm_min, d->tm_sec);

return ret;
}



--
Chas. Owens
wonkden.net
The most important skill a programmer can have is the ability to read.

--
To unsubscribe, e-mail: beginners-unsubscribe [at] perl.org
For additional commands, e-mail: beginners-help [at] perl.org
http://learn.perl.org/

Chas. Owens wrote:
> On Wed, Aug 11, 2010 at 08:39,  <rkb [at] i.frys.com> wrote:
> snip
>> I probably should have mentioned that the "Matt" code is
>> what is currently being used in production and I need to
>> profile/benchmark it against different approaches.
> snip
>
> Just getting rid of the stupid in your production code
> increases the
> speed by between forty and fifty percent (at least on OS
> X). It also
> drops the line count from twelve to three.
>
> #!/usr/bin/perl
>
> use strict;
> use warnings;
> use POSIX qw(strftime);
> use Benchmark qw(:all);
>
> cmpthese(-1, {
> production => \&production,
> experimental => \&experimental,
> });
>
>
> sub production {
> my $now_date_epoch = time();
> my $BDtarget = ($now_date_epoch - 5);
> my
> ($Bsec,$Bmin,$Bhour,$Bmday,$Bmon,$Byear,$Bwday,$Byday,$Bisds t)
> = localtime($BDtarget);
> $Byear = ($Byear + 1900);
> $Bmon++;
> if ($Bmon < 10) {$Bmon = "0$Bmon";}
> if ($Bmday < 10) {$Bmday = "0$Bmday";}
> if ($Bhour < 10) {$Bhour = "0$Bhour";}
> if ($Bmin < 10) {$Bmin = "0$Bmin";}
> if ($Bsec < 10) {$Bsec = "0$Bsec";}
> my $BDtsSQLdate = "$Byear$Bmon$Bmday$Bhour$Bmin$Bsec";
> }
>
> sub experimental {
> my ($s, $min, $h, $d, $mon, $y) = localtime(time - 5);
> my $BDtsSQLdate = sprintf "%d%02d%02d%02d%02d%02d", $y
> + 1900,
> $mon + 1, $d, $h, $min, $s;
> }
>
>
>
> --
> Chas. Owens

That looks like it might be the best correction and will
most likely be what I'll use, but I'd like to test 1 or 2
other possibilities.

Thanks,

Ron


--
To unsubscribe, e-mail: beginners-unsubscribe [at] perl.org
For additional commands, e-mail: beginners-help [at] perl.org
http://learn.perl.org/

2010/8/11 Chas. Owens <chas.owens [at] gmail.com>:
> On Wed, Aug 11, 2010 at 08:39, =A0<rkb [at] i.frys.com> wrote:

>>
>> My theory is that 5.10.x implemented some optimizations
>> that improved the speed of strftime.
> snip
>
> Unlikely, I don't remember any big changes to strftime (it is
> implemented by Perl_my_strftime in [util.c][0] if you want to check).
> I ran your code against 5.12.1, 5.10.0, and 5.8.9 on OS X and got
>

Me also. Didn't get big difference among the three versions for that functi=
on.


--
Jeff Peng
Web: http://home.arcor.de/pangj/
Blog: http://squidcn.spaces.live.com/

--
To unsubscribe, e-mail: beginners-unsubscribe [at] perl.org
For additional commands, e-mail: beginners-help [at] perl.org
http://learn.perl.org/