Selected Files From

The Marsaglia Random Number CDROM

including the

DIEHARD Battery of Tests of Randomness

Professor Marsaglia's CDROM, based on research supported by the National Science Foundation under grants DMS-8807976 and DMS-9206972, contains approximately 600 Mbytes of files of random bytes that pass all known tests for randomness. Those files are not included on this CDROM. Here is the top level readme file from Marsaglia's full CDROM.

The files included on this CDROM relate to Marsaglia's rightly famous, if whimsically named, "Diehard battery of tests of randomness". These have come be viewed as something of an industry standard for testing random number generators. Over the years, several generators thought to be good have been shown to be inadequate by these tests.

Here is a directory of source files (or, Mac alternate listing) for the Diehard tests. The principal program needed is diehard.c. Note that this is a C program automatically translated from the original Fortran by the program "f2c" (included elsewhere on this CDROM. To compile Diehard successfully, you will need to compile and build both of the f2c libraries, and then compile with a command line like

cc -o diehard diehard.c -lf2c -lm

For further information, look at the readme file. (Sorry, neither we, nor Professor Marsaglia, can provide any support on this. We've succeeded in doing it, so we know it can be done.)

You may find it easier to modify and compile the original Fortran version (diehard.f), also included by permission of Professor Marsaglia, but with the strict understanding that it be treated as a well-tested "working composite code," with 30 years of accumulated patches; it is not polished code intended for public release.

Much easier than recompiling is to use one of the precompiled versions of the executable files that are supplied for the following machine types:

A word of caution about the files canada.bit, germany.bit, and calif.bit, included in this directory. As Marsaglia explains, these files do not pass all tests for randomness, but are included for their usefulness in demonstrating Diehard, or in assessing other routines for finding nonrandomness.