I would remove the last sentence or just talk about the benefit of having it in ALv2 without claiming any firsts. (Also, the code was adapted from another source and I think that should be acknowledged. I don't know the source. It is not exactly the algorithm published in the Wichmann-Hill paper I have.)
- Dennis Concerning performance, etc. The improved Wichmann-Hill pseudo-random generator paper that is referenced is behind a pay wall. There is a non-pay-wall version available at <http://www.eurometros.org/component_search.php?component_type=algorithms>. Finally, the new Wichmann-Hill algorithm is slower than the original version and also Mersenne Twister. It is simple to implement. The implementation works on any 32-bit-capable (x86 but also including x64) system by running 4 nearly-31-bit pseudo-random generators in parallel and combining their results. The 4 generators have different maximum periods such that it takes an extremely long period before those maximum periods simultaneously return to their starting values. In the addition of the improved Wichmann-Hill algorithm for ScRandom() in Calc by Pedro, <https://issues.apache.org/ooo/show_bug.cgi?id=121421#c28>, the initial seed is essentially a randomly-generated, roughly 124-bit sequence distributed into seeds for the 4 internal generators. (128 bits are generated, but since the generators work modulo 31-bit primes, the unique seeds have fewer than 2^124 combinations.) I'm not aware of any other analysis that has been done. The only unexplored area, I think, is whether further initial-seed conditioning is needed to ensure maximum periods and also avoid a few improbable but awful pathological cases. It would be useful to run one of the well-known pseudo-random-generator stress tests to confirm that the reported qualities of Wichmann-Hill #2 are being delivered. -----Original Message----- From: Rob Weir [mailto:rabas...@gmail.com] Sent: Tuesday, December 18, 2012 09:29 To: dev@openoffice.apache.org Subject: Re: Blog post On Dec 18, 2012, at 10:12 AM, Pedro Giffuni <p...@apache.org> wrote: > Hello; > > Just to get the general public to know some of the things there are going on > in > the AOO code, Andrea and I have been preparing a blog post about the new > random number generator: > > https://blogs.apache.org/preview/OOo/?previewEntry=random_numbers_in_calc_small > Nice. Is it worth describing the testing? In particular, do we have any tests that show clear improvements? Anything that can be shown in a chart? -Rob > > Just thought we should give you a chance to complain about it before it goes > live ;). > > cheers, > > Pedro.
