$ uname -a
Linux kubo39 3.2.0-51-generic-pae #77-Ubuntu SMP Wed Jul 24 20:40:32 UTC 2013 i686 i686 i386 GNU/Linux
$ cat /proc/cpuinfo | grep "model name"
model name : Intel(R) Core(TM) i7-3517U CPU @ 1.90GHz
model name : Intel(R) Core(TM) i7-3517U CPU @ 1.90GHz
model name : Intel(R) Core(TM) i7-3517U CPU @ 1.90GHz
model name : Intel(R) Core(TM) i7-3517U CPU @ 1.90GHz
$ cat /proc/meminfo | grep MemTotal
MemTotal: 4011464 kB
Version of each language
python 2.7.3
ruby 2.0.0-p247
--Python code
def test_call_next(n=100001):
iter = range(0, n).__iter__()
while True:
try:
iter.next()
except StopIteration:
break
$ time python iter.py
real 0m0.042s
user 0m0.028s
sys 0m0.012s
$ time python iter.py
real 0m0.046s
user 0m0.044s
sys 0m0.004s
$ time python iter.py
real 0m0.036s
user 0m0.028s
sys 0m0.004s
--Ruby code
def test_call_next n=100000
iter = [*0..n].each
loop do
iter.next
end
end
$ time ruby iter.rb
real 0m0.138s
user 0m0.096s
sys 0m0.040s
$ time ruby iter.rb
real 0m0.145s
user 0m0.116s
sys 0m0.028s
$ time ruby iter.rb
real 0m0.147s
user 0m0.124s
sys 0m0.020s
It's about 3.5 times faster than Python, but it may not be a good bench because it takes into account the cost of iterator generation.
The reason why sys is large in Ruby
$ strace -c ruby iter.rb
% time seconds usecs/call calls errors syscall
------ ----------- ----------- --------- --------- ----------------
98.14 0.006114 0 200005 sigprocmask
1.86 0.000116 1 85 read
0.00 0.000000 0 1 write
0.00 0.000000 0 200 146 open
0.00 0.000000 0 55 close
0.00 0.000000 0 1 execve
0.00 0.000000 0 1 time
0.00 0.000000 0 8 8 access
0.00 0.000000 0 27 brk
0.00 0.000000 0 25 22 ioctl
0.00 0.000000 0 1 gettimeofday
0.00 0.000000 0 7 munmap
0.00 0.000000 0 1 clone
0.00 0.000000 0 1 uname
0.00 0.000000 0 12 mprotect
0.00 0.000000 0 9 _llseek
0.00 0.000000 0 1 mremap
0.00 0.000000 0 16 rt_sigaction
0.00 0.000000 0 23 rt_sigprocmask
0.00 0.000000 0 1 getcwd
0.00 0.000000 0 1 sigaltstack
0.00 0.000000 0 6 getrlimit
0.00 0.000000 0 37 mmap2
0.00 0.000000 0 37 15 stat64
0.00 0.000000 0 96 lstat64
0.00 0.000000 0 117 fstat64
0.00 0.000000 0 14 getuid32
0.00 0.000000 0 14 getgid32
0.00 0.000000 0 15 geteuid32
0.00 0.000000 0 15 getegid32
0.00 0.000000 0 2 getdents64
0.00 0.000000 0 46 fcntl64
0.00 0.000000 0 2 1 futex
0.00 0.000000 0 5 sched_getaffinity
0.00 0.000000 0 1 set_thread_area
0.00 0.000000 0 1 set_tid_address
0.00 0.000000 0 2 clock_gettime
0.00 0.000000 0 1 openat
0.00 0.000000 0 1 set_robust_list
0.00 0.000000 0 2 pipe2
------ ----------- ----------- --------- --------- ----------------
100.00 0.006230 200895 192 total
It seems that it is because it calls sigprocmask (2) every time.
--Python code
def test_create_iterator(n=10001):
[range(0, 1001).__iter__ for _ in xrange(n)]
$ time python iter.py
real 0m0.328s
user 0m0.280s
sys 0m0.044s
$ time python iter.py
real 0m0.342s
user 0m0.276s
sys 0m0.064s
$ time python iter.py
real 0m0.324s
user 0m0.268s
sys 0m0.052s
--Ruby code
def test_create_enum n=10000
n.times{ [*0..1001].to_enum }
end
$ time ruby iter.rb
real 0m0.554s
user 0m0.548s
sys 0m0.004s
$ time ruby iter.rb
real 0m0.558s
user 0m0.552s
sys 0m0.004s
$ time ruby iter.rb
real 0m0.566s
user 0m0.560s
sys 0m0.000s
Again, Python is about 1.7 times faster.
I'm curious that Python has a large sys time.
$ strace -c python iter.py
% time seconds usecs/call calls errors syscall
------ ----------- ----------- --------- --------- ----------------
53.85 0.000049 0 337 250 open
46.15 0.000042 0 1292 brk
0.00 0.000000 0 183 read
0.00 0.000000 0 89 close
0.00 0.000000 0 1 execve
0.00 0.000000 0 11 11 access
0.00 0.000000 0 5 1 ioctl
0.00 0.000000 0 4 2 readlink
0.00 0.000000 0 55 munmap
0.00 0.000000 0 1 uname
0.00 0.000000 0 11 mprotect
0.00 0.000000 0 3 _llseek
0.00 0.000000 0 68 rt_sigaction
0.00 0.000000 0 1 rt_sigprocmask
0.00 0.000000 0 1 getcwd
0.00 0.000000 0 1 getrlimit
0.00 0.000000 0 86 mmap2
0.00 0.000000 0 172 96 stat64
0.00 0.000000 0 9 lstat64
0.00 0.000000 0 141 fstat64
0.00 0.000000 0 1 getuid32
0.00 0.000000 0 1 getgid32
0.00 0.000000 0 1 geteuid32
0.00 0.000000 0 1 getegid32
0.00 0.000000 0 4 getdents64
0.00 0.000000 0 1 1 futex
0.00 0.000000 0 1 set_thread_area
0.00 0.000000 0 1 set_tid_address
0.00 0.000000 0 2 openat
0.00 0.000000 0 1 set_robust_list
------ ----------- ----------- --------- --------- ----------------
100.00 0.000091 2485 361 total
ʻOpen (2)andbrk (2)are spending a lot of time, especially the number of calls tobrk (2)`.
By the way, brk (2) is a system call to change the amount of memory allocated to a process's data segment.
The heap size is not enough when mallocing && The process is called when there is enough memory available, so it is actually I think many people use it without knowing it.
Compare with an abstraction of the code you really needed
--Python code
def test_for_generate_enumerator(n=50001):
arr = range(0, 11)
for i in xrange(0, n):
iter = arr.__iter__()
while True:
try:
iter.next()
except StopIteration:
break
$ time python iter.py
real 0m0.134s
user 0m0.128s
sys 0m0.004s
$ time python iter.py
real 0m0.134s
user 0m0.128s
sys 0m0.004s
$ time python iter.py
real 0m0.142s
user 0m0.132s
sys 0m0.008s
--Ruby code
def test_for_iter_with_generate_enumerator n=50000
arr = [*0..10]
n.times {
iter = arr.to_enum
loop do
iter.next
end
}
end
$ time ruby iter.rb
real 0m1.370s
user 0m1.080s
sys 0m0.288s
$ time ruby iter.rb
real 0m1.377s
user 0m0.992s
sys 0m0.380s
$ time ruby iter.rb
real 0m1.362s
user 0m1.060s
sys 0m0.296s
Ruby is extremely slow like this ...
But it's strange that Python's sys time is smaller in this code than when it generated a lot of iterators.
Apparently, the Python iterator is faster for both generation and next element call.
Next time (if any), I'd like to actually follow the processing code.
Recommended Posts