Asp Forum - Re: String performance

Gavin Kistner

1/9/2007 9:08:00 PM

From: Vincent Fourmond
> user system total real
> 0.010000 0.000000 0.010000 ( 0.009087)
> 0.010000 0.000000 0.010000 ( 0.008774)
> 0.000000 0.000000 0.000000 ( 0.004621)

Perhaps your machine is more deterministic than mine, but successive
runs of that benchmark (and using #bmbm to be safer about the
measurement) sometimes show 'a' faster than "a", sometimes slower.

Even with benchmarking, I wouldn't trust that answers that are within a
few percent of each other. And I certainly wouldn't rush off to refactor
code because of it.

6 Answers

khaines

1/9/2007 9:17:00 PM

Vincent Fourmond

1/9/2007 9:17:00 PM

Gavin Kistner wrote:
> From: Vincent Fourmond
>> user system total real
>> 0.010000 0.000000 0.010000 ( 0.009087)
>> 0.010000 0.000000 0.010000 ( 0.008774)
>> 0.000000 0.000000 0.000000 ( 0.004621)
>
> Perhaps your machine is more deterministic than mine, but successive
> runs of that benchmark (and using #bmbm to be safer about the
> measurement) sometimes show 'a' faster than "a", sometimes slower.

The thing is they are rigourosly equivalents. As soon as the program
is parsed, they are represented exactly as the same objects, a String.
So they are the same, that's why you get around the same processing times.

Moreover, it is normal that interpolation is faster, because it
involves only evaluation of x as a String and string copy, whereas
addition involves two method calls (+ and +), which are rather expensive
(at least more than a string copy for such small strings).

Cheers,

Vince

--
Vincent Fourmond, PhD student
http://vincent.fourmon...

Robert Klemme

1/10/2007 8:11:00 AM

On 09.01.2007 22:17, khaines@enigo.com wrote:
> On Wed, 10 Jan 2007, Gavin Kistner wrote:
>
>> From: Vincent Fourmond
>>> user system total real
>>> 0.010000 0.000000 0.010000 ( 0.009087)
>>> 0.010000 0.000000 0.010000 ( 0.008774)
>>> 0.000000 0.000000 0.000000 ( 0.004621)
>>
>> Perhaps your machine is more deterministic than mine, but successive
>> runs of that benchmark (and using #bmbm to be safer about the
>> measurement) sometimes show 'a' faster than "a", sometimes slower.
>>
>> Even with benchmarking, I wouldn't trust that answers that are within a
>> few percent of each other. And I certainly wouldn't rush off to refactor
>> code because of it.
>
> Increase n from 5000 to 500000 or 5000000.
>
> To understand the difference, just think about how many strings are
> being created with each.
>
> 'a' creates a new string, as does 'b'.
> The + operation creates a new string, as well.
>
> So, there's a lot of new string creation happening with either of the +
> examples.
>
> Change the +'s to << and you will see a difference.
>
> 'a' << x << 'b'
>
> << just changes the old String.
>
> The "a#{x}b" example does the least work.

I have added some alternatives - string interpolation still wins

robert@fussel /cygdrive/c/temp
$ ruby str_bench.rb
Rehearsal -------------------------------------------
'a' + 2.860000 0.000000 2.860000 ( 2.859000)
"a" + 2.890000 0.000000 2.890000 ( 2.891000)
a#{ 1.860000 0.000000 1.860000 ( 1.859000)
"" << 3.734000 0.000000 3.734000 ( 3.734000)
"a" << 2.328000 0.000000 2.328000 ( 2.328000)
A + 2.625000 0.000000 2.625000 ( 2.625000)
"" << A 3.453000 0.000000 3.453000 ( 3.453000)
--------------------------------- total: 19.750000sec

user system total real
'a' + 2.906000 0.000000 2.906000 ( 2.907000)
"a" + 2.891000 0.000000 2.891000 ( 2.890000)
a#{ 1.859000 0.000000 1.859000 ( 1.860000)
"" << 3.766000 0.000000 3.766000 ( 3.765000)
"a" << 2.344000 0.000000 2.344000 ( 2.344000)
A + 2.640000 0.000000 2.640000 ( 2.641000)
"" << A 3.469000 0.000000 3.469000 ( 3.468000)

robert@fussel /cygdrive/c/temp
$ cat str_bench.rb
require 'benchmark'

n = 1_000_000
c = "stuff"

A = "a"
B = "b"

Benchmark.bmbm do |x|
x.report('\'a\' +') { n.times {'a' + c + 'b'}}
x.report('"a" +') { n.times {"a" + c + "b"}}
x.report('a#{') { n.times {"a#{c}b"}}
x.report('"" <<') { n.times {"" << "a" << c << "b"}}
x.report('"a" <<') { n.times {"a" << c << "b"}}

x.report('A +') { n.times {A + c + B}}
x.report('"" << A') { n.times {"" << A << c << B}}
end

robert

Jano Svitok

1/10/2007 12:42:00 PM

On 1/10/07, Jason Mayer <slamboy@gmail.com> wrote:
> On 1/10/07, Robert Klemme <shortcutter@googlemail.com> wrote:
> >
> > On 09.01.2007 22:17, khaines@enigo.com wrote:
> > > On Wed, 10 Jan 2007, Gavin Kistner wrote:
> > >
> > >> From: Vincent Fourmond
> > >>> user system total real
> > >>> 0.010000 0.000000 0.010000 ( 0.009087)
> > >>> 0.010000 0.000000 0.010000 ( 0.008774)
> > >>> 0.000000 0.000000 0.000000 ( 0.004621)
> > >>
>
> What do the different columns actually mean?

http://ruby-doc.org/stdlib/libdoc/benchmark/rdoc/classes/Benc...

This report shows the user CPU time, system CPU time, the sum of the
user and system CPU times, and the elapsed real time. The unit of time
is seconds.

i.e. time spent in user mode, kernel mode, user+kernel (for these only
time spent by this particular program is counted) and elapsed real
("wallclock" time)

Pit Capitain

1/10/2007 2:31:00 PM

Jason Mayer schrieb:
> (...)
> Can someone help me understand if this is a problem with benchmark or with
> my code?

Jason, simply run the same loops without the benchmark library. Since
benchmark is just recording some times, I guess you'll see the same
behaviour.

Regards,
Pit

Jean-François

1/10/2007 4:24:00 PM

Robert :> robert@fussel /cygdrive/c/temp> $ cat str_bench.rb> require 'benchmark'>> n = 1_000_000> c = "stuff">> A = "a"> B = "b">> Benchmark.bmbm do |x|> x.report('\'a\' +') { n.times {'a' + c + 'b'}}> x.report('"a" +') { n.times {"a" + c + "b"}}> x.report('a#{') { n.times {"a#{c}b"}}> x.report('"" <<') { n.times {"" << "a" << c << "b"}}> x.report('"a" <<') { n.times {"a" << c << "b"}}>> x.report('A +') { n.times {A + c + B}}> x.report('"" << A') { n.times {"" << A << c << B}}> endHi,You can add the format way also : "a%sb" % cx.report('a%sb') { n.times { "a%sb" % c }}slower than string interpolation. -- Jean-François.-- À la renverse.

comp.lang.ruby

Re: String performance