Path: utzoo!attcan!uunet!wuarchive!zaphod.mps.ohio-state.edu!swrinde!ucsd!ucbvax!bloom-beacon!oliver
From: oliver@athena.mit.edu (James D. Oliver III)
Newsgroups: comp.lang.fortran
Subject: Random Number Functions--Generate versus Input
Message-ID: <1990Nov8.170334.24579@athena.mit.edu>
Date: 8 Nov 90 17:03:34 GMT
Sender: daemon@athena.mit.edu (Mr Background)
Organization: Massachusetts Institute of Technology
Lines: 21

I have a number (order of magnitude 10**6) of random number function values
to generate as part of a Monte Carlo simulation using Cray Fortran.  The
actual variables of interest are independent of the random numbers, so I
can use the same series for each simulation.  My question is, is it faster
in general to regenerate the random function values every time (this part
of the code can't be vectorized), or to just read the values in from a
file?

Obviously, the solution is to run benchmark programs, which I intend to do,
but just as a rule of thumb, is I/O considerably slower than calling
functions?

Also, could someone repost that pointer to a supercomputer
optimization/vectorization reference that was made earlier?


--
____________________________
	Jim Oliver   
	oliver@athena.mit.edu /	joliver@hstbme.mit.edu
	oliver%mitwccf.BITNET@MITVMA.MIT.EDU