Keep up the good work.

Some applications need to generate millions of random numbers as quickly as possible. When you benchmark SecureRandom in this context it is many times slower than the alternatives.

The RNGs in Uncommons Maths were originally written for use in evolutionary computation. When I was running evolutionary simulations for my masters degree (without these RNGs), a single experiment would take about 12 hours to complete, and the time spent generating random values was a significant chunk of this.

The following code took 60 ms to execute on my Pentium 4, 3.0 GHZ 5 year old PC.

long t1 = System.currentTimeMillis();
byte[] holder = new byte[1024/8];
SecureRandom secureRandom = SecureRandom.getInstance(”SHA1PRNG”);
secureRandom.nextBytes(holder);
System.out.println(”new BASE64Encoder().encode(holder) = ” + new BASE64Encoder().encode(holder));
long t2 = System.currentTimeMillis();
System.out.println(t2-t1);

Which makes me wonder why you call SecureRandom slow?

CellularAutomatonRNG sounds intriguing. I’ll have to take a closer look.

I wrote a lengthy article about randomness in games (shuffling in particular). Eventually it’s of some use for the follow up article.

http://kaioa.com/node/53