Here’s a real Array performance mystery. I can’t help my friend figure
this one out. He writes,
========
Is there a known bug in Ruby array performance? I spent a lot of time
yesterday boiling down the following example
http://blogs.codehaus.org/people/geir/archives/001768_baffled_with_ruby.html
only because I still can’t believe I’m not doing something incredibly
stupid. (I’m a newbie, “Reluctant Rubyist”)
I’m just not used to arrays behaving this way 
Anyone have any
insight?
========
He’d love a response in his blog, but with permission I’ll copy
answers here over there. He’s not a regular ruby-talk reader…yet.
Jim
2009/2/27 Jim M. [email protected]:
only because I still can’t believe I’m not doing something incredibly
stupid. Â (I’m a newbie, “Reluctant Rubyist”)
 I’m just not used to arrays behaving this way  Anyone have any
insight?
========
He’d love a response in his blog, but with permission I’ll copy
answers here over there. He’s not a regular ruby-talk reader…yet.
Slightly modified test (attached):
15:44:06 Temp$ ruby bm.rb bm; ruby bm.rb bmbm; ruby19 bm.rb bm; ruby19
bm.rb bmbm
1.8.7
user system total
real
size = 10000 10000 0.125000 0.000000 0.125000 (
0.114000)
size = 100000 100000 0.859000 0.000000 0.859000 (
0.863000)
size = 1000000 1000000 24.969000 0.000000 24.969000 (
24.961000)
1.8.7
Rehearsal
size = 10000 10000 24.235000 0.032000 24.267000 (
24.308000)
size = 100000 100000 18.078000 0.000000 18.078000 (
18.150630)
size = 1000000 1000000 14.297000 0.000000 14.297000 (
14.289000)
------------------------------------------------------- total:
56.642000sec
user system total
real
size = 10000 10000 12.031000 0.000000 12.031000 (
12.039000)
size = 100000 100000 10.250000 0.000000 10.250000 (
10.290950)
size = 1000000 1000000 8.985000 0.000000 8.985000 (
8.989000)
1.9.1
user system total
real
size = 10000 10000 0.016000 0.000000 0.016000 (
0.027000)
size = 100000 100000 0.031000 0.000000 0.031000 (
0.028000)
size = 1000000 1000000 0.047000 0.000000 0.047000 (
0.043000)
1.9.1
Rehearsal
size = 10000 10000 0.032000 0.000000 0.032000 (
0.043000)
size = 100000 100000 0.032000 0.000000 0.032000 (
0.043000)
size = 1000000 1000000 0.031000 0.000000 0.031000 (
0.043000)
-------------------------------------------------------- total:
0.095000sec
user system total
real
size = 10000 10000 0.031000 0.000000 0.031000 (
0.039000)
size = 100000 100000 0.047000 0.000000 0.047000 (
0.043000)
size = 1000000 1000000 0.032000 0.000000 0.032000 (
0.043000)
Strange that the rehearsal has this dramatic effect. Normally it
should simply execute the block twice to warm up the system.
Kind regards
robert
2009/2/27 Robert K. [email protected]:
Strange that the rehearsal has this dramatic effect. Â Normally it
should simply execute the block twice to warm up the system.
Here’s a better test
16:16:32 Temp$ ruby bm.rb; ruby19 bm.rb
1.8.7
user system total
real
size = 10000 10000 0.110000 0.000000 0.110000 (
0.118000)
size = 100000 100000 0.781000 0.000000 0.781000 (
0.769000)
size = 1000000 1000000 25.172000 0.000000 25.172000 (
25.185000)
size = 10000 10000 0.328000 0.000000 0.328000 (
0.329000)
size = 100000 100000 0.547000 0.000000 0.547000 (
0.542000)
size = 1000000 1000000 18.156000 0.000000 18.156000 (
18.151000)
1.9.1
user system total
real
size = 10000 10000 0.031000 0.000000 0.031000 (
0.027000)
size = 100000 100000 0.031000 0.000000 0.031000 (
0.028000)
size = 1000000 1000000 0.031000 0.000000 0.031000 (
0.043000)
size = 10000 10000 0.031000 0.000000 0.031000 (
0.031000)
size = 100000 100000 0.032000 0.000000 0.032000 (
0.023000)
size = 1000000 1000000 0.031000 0.000000 0.031000 (
0.043000)
“better” in the sense that we’re not suffering from bmbm’s magic.
Kind regards
robert
Jim M. wrote:
Jim
This variation exhibits the same behavior.
require ‘benchmark’
class BufferContainer
def initialize( initial_data )
@buf = initial_data
end
def put_array( array )
@buf[0, array.size] = array
end
end
size = 999
the_array = (0…4).to_a
(1…3).each{|i|
size *= 10
a = [nil] * size
RUN WITH THIS COMMENTED OUT FIRST
a << 9
puts “size = #{ size } #{ a.size }”
puts Benchmark.measure {
10_000.times {
buf = BufferContainer.new( a )
buf.put_array( the_array )
}
}
}
On Feb 27, 2009, at 05:16 , Jim M. wrote:
Is there a known bug in Ruby array performance? I spent a lot of time
yesterday boiling down the following example
http://blogs.codehaus.org/people/geir/archives/001768_baffled_with_ruby.html
I don’t have time to dig into this, but I believe he’s hitting a GC
threshold by going over N objects.
On Feb 28, 2009, at 21:32 , Ryan D. wrote:
threshold by going over N objects.
Poke around gc.c and see what you can figure out. Also, look at the
1.9 version or through the changelog to find hints…
#ifndef GC_MALLOC_LIMIT
#if defined(MSDOS) || defined(human68k)
#define GC_MALLOC_LIMIT 200000
#else
#define GC_MALLOC_LIMIT 8000000
#endif
#endif
Jim M. wrote:
Jim
I am betting Ryan is right about GC issues. JRuby does not fall down
doing this either (which probably also backs up the GC theory):
ruby ~/jruby/scripts/arr_ben.rb
size = 10000 10000
0.050000 0.000000 0.050000 ( 0.057190)
size = 100000 100000
0.660000 0.000000 0.660000 ( 0.672178)
size = 1000000 1000000
34.430000 0.340000 34.770000 ( 35.562454)
jruby --server ~/jruby/scripts/arr_ben.rb
size = 10000 10000
0.372000 0.000000 0.372000 ( 0.266000)
size = 100000 100000
0.099000 0.000000 0.099000 ( 0.099000)
size = 1000000 1000000
0.154000 0.000000 0.154000 ( 0.154000)
-Tom