From: Daniel Berger <Daniel.Berger@...>
Date: 2005-10-14T01:50:05+09:00
Subject: Re: ANN:  Crypt::ISAAC 0.9.1 released

Kirk Haines wrote:
> This release is primarily a reorganization of the old package.  It has a much 
> better installer, a basic unit test suite, better README and licensing info, 
> and a gem has been built for Crypt::ISAAC as well.  The only functionality 
> change is to add an option when creating a new generator that will force it 
> to try to use /dev/random to seed the generator, instead of /dev/urandom 
> (falling back to rand() if neither is available).  This is an experimental 
> feature -- seeding from /dev/random appears to be almost too slow to be 
> usable since so much entropy is needed, but to use it, create pass false when 
> creating the object, as follows:
> 
> prng = Crypt::ISAAC.new(false)

I got some improvement when I replaced "read" with "sysread" in ISAAC.rb. 
Here's a little benchmark script I wrote:

require "crypt/ISAAC"
require "benchmark"
include Benchmark

MAX = ARGV[0] || 100

bm do |x|
    x.report("rand"){
       MAX.times{ Crypt::ISAAC.new(true) }
    }

    x.report("/dev"){
       MAX.times{ Crypt::ISAAC.new(false) }
    }
end

Results:

# Using read
 >ruby bench_isaac.rb
       user     system      total        real
rand  6.290000   0.410000   6.700000 (  6.928163)
/dev  6.490000   9.640000  16.130000 ( 23.197032)

# Using sysread
 >ruby bench_isaac.rb
       user     system      total        real
rand  6.340000   0.550000   6.890000 (  7.148228)
/dev  6.560000   3.050000   9.610000 (  9.976435)

This was on a stock Sunblade 150 running Solaris 10.

Regards,

Dan