From: Brian Candler <b.candler@...>
Date: 2009-04-30T05:21:18+09:00
Subject: Re: Weird performance issue

court3nay wrote:
> It's not just requiring
> libraries -- I can reproduce with pure math.  On the "fast" machine:
> 
> irb(main):007:0> Benchmark.realtime { 1000.times { 10 ** rand
> (1000) } }
> => 0.00720095634460449
> 
> On the "slow" machine....
> 
> Benchmark.realtime { 1000.times { 10 ** rand(1000) } }
> => 0.142554998397827

Then that rules out rubygems, since under 1.8.6, unless you require 
'rubygems' (or have -rubygems in your RUBYOPT) it won't be touched at 
all.

Make sure you don't compare just real time, but also CPU time used on 
the two machines. If the CPU time is the same but the real time is 
vastly different, it could mean that one VM is running a bunch of other 
processes (compare load average using 'uptime'), or that one VM is 
deprioritised with respect to the other.

I've never heard of "arch linux" (is that a distribution?), but assuming 
it comes with the usual bits of code, I'd also suggest running a 
comparable benchmark using say Perl. e.g.

$ time perl -e 'for ($i=0;$i<1000000;$i++) {10*10}'

real  0m0.737s
user  0m0.720s
sys  0m0.008s

If the times are vastly different between the two machines, then you 
know the problem is nothing to do with ruby and is something to do with 
the VM platform.

If the times are identical for Perl but different for Ruby, then 
something odd is going on. You can try copying the Ruby from one over to 
the other, but that's unlikely to explain the difference if Ruby came 
from the distro as a binary, or was the same binary unpacked from a 
tarball.

Regards,

Brian.
-- 
Posted via http://www.ruby-forum.com/.