fast-pseudorandom

This software is owned by The university of California, Irvine, and is not distributed under any version of the GPL. GPL is a fine series of licenses, but the owners of the software need it to be distributed under these terms.
The fast-pseudorandom program just generates lots of data, fast. It is in a sense, especially well suited to benchmarking, because while it is outputting a pseudo-random number sequence, that sequence is always the same, for a given byte stream length, eliminating a class of variable from your tests.

Usage is like:

esmf04m-root> fast-pseudorandom  -h
Usage: fast-pseudorandom [-h] [-f] [-u]
-h      Show this help message
-f      Go fast (faster than the default, but far less random)
-u      Use unbuffered I/O (IE, do not use stdio)

This program just writes a stream of pseudorandom characters very rapidly
If you need good entropy, you want /dev/random or prngd or something instead.
This program is for when you need something that bounces around a lot,
just for benchmarking, because you're thinking a stream of zeros might
be optimized away somehow, but true randomness isn't at all important - while speed is

BTW, the -u option doesn't seem to change things much, at least not on an AIX 5.1 system I tested with.

Note that generally from fastest to slowest, I'm guessing you'll find (and there may be some large drops):

Data source	Availability	Suitable for testing with compression?	Bandwidth (as measured by reblock) on an 8 CPU AIX 5.1 system with single-CPU nbench numbers of: MEMORY INDEX: 5.965 INTEGER INDEX: 6.386 FLOATING-POINT INDEX: 18.523 ...using only one CPU in the tests, and transmitting data for 4 minutes for each test (far from a large sample size, but for a thumbnail comparison...)
reading from /dev/zero	nearly all *ix have this	no	2.9 Gigabits/second
reading from /dev/zero and intentionally piping through an unnecessary cat	nearly all *ix have this	no	2.3 Gigabits/second
reading from fast-pseudorandom -f (character increment mode)	build the program	no	1.6 Gigabits/second
reading from fast-pseudorandom (linear congruential PRNG)	build the program	yes	212.5 Megabits/second with gcc -O3, and 348.8 Megabits/second with xlc -O3
reading from /dev/urandom or prngd	many *ix have /dev/urandom, including linux and solaris - or install prngd	yes
reading from /dev/random or prngd	many *ix have /dev/random, including linux and solaris - or install prngd	yes
reading from disk	very nearly all *ix have this	usually, unless you're reading from a device file, and the filesystem hasn't been filled	205.1 Megabits/second
reading from network (unless you have a gigabit net or 10 gigabit net or something)	nearly all *ix have this	yes

Hits: 4261
Timestamp: 2024-12-27 07:38:24 PST

Back to Dan's tech tidbits

You can e-mail the author with questions or comments: