Re^6: Question about curious performance of if...elsif block

[markseger]

> No. Because it would not run here. ahh, so I guess you're not running linux.

no problems. here's a run with the subroutine call commented out:

[root@poker collectl]# time ./test.pl
real    0m14.464s
user    0m5.089s
sys     0m9.087s

now with the call executing BUT the return statement executing first thing in the subroutine:

[root@poker collectl]# time ./test.pl
real    0m19.508s
user    0m10.461s
sys     0m8.853s

so clearly the overhead in calling the subroutine 1M times is about 5-1/2 seconds. Now I'll comment out the early return in the subroutine in the subroutine and run it again:

[root@poker collectl]# time ./test.pl
real    0m25.648s
user    0m16.350s
sys     0m9.123s

This says to me the if...elsif... takes about 6 seconds and I can't fathom why! Could there be some weird interaction going on between perl/linux and the /proc filesystem?

As I said in my base note I'm not an innards kind of a guy but it feels like something that is not going to show up in every perl script. Rather if feels like there is some kind of weird interraction.

and speaking of environment this is rhel5.3 and perl 5.8.8. I just tried the code again on rhel6.2 in a VM with perl 5.10, and the differences between the last 2 runs are:

real    0m7.359s
user    0m2.840s
sys     0m4.481s

real    0m9.782s
user    0m5.082s
sys     0m4.615s

real    0m11.927s
user    0m7.266s
sys     0m4.604s

so in this case both the subroutine calls and the if...elsif... block are only taking a little over 2 seconds each. This says to me there is something more efficient going on with 5.10.

Perhaps this is less of a big deal as 2 extra seconds is getting less of a concern, but I still an curious about what exactly is going on.

-mark

[bulk88]

Have you tried using B::Concise "perl -MO=Concise,-exec,-src" on your scripts?

[jethro]

Remember that a processor is a complex beast. For example your bigger script might just have exceeded the small L1 instruction cache of your processsor, which is just 32k on a core2 CPU.