The benchmark is 5323 fortran-lines long (eg. the computational kernel, without any comment line) while the original application with the comments and the I/O parts -- pre- and postprocessing, dump and restart features, etc ... -- is approximatively 18000 lines long. The executable size is about 350 kbytes or 100 kwords. The test has to be run in 64-bits-precision mode eg. in single precision mode on 60- or 64-bits machines or in appropriate-precision mode on smaller-word, even 16bits, machines.
The program is best described as an heavy computationnal profile (CP bound) with negligible I/O code. Almost all the explicit processings are done with floating point values in 64-bits precision. However the program doesn't carry much array processing that can be vectorized, this later part being estimated at about 2%. Instead it has an iterative structure with much abundance of short branches and small loops that could be best served by features like cache memory ....
It is important to note that, while minor modification can be targeted toward a particular hardware, or even software, this program is run as is on every machine, and since 1978. Lastly, it is worth noting that this program, being a simulation code, has an observed behavior much similar to another group of simulation families: Monte-Carlo code.