Bring in FreeBSD's stress2 stress testing suite.

[dragonfly.git] / test / stress / stress2 / README

$FreeBSD$

This is the FreeBSD kernel stress test suite. The purpose is to crash the
computer, by stressing selected parts of the kernel, thus exposing inadequate
error handling.

Do not run the syscall test as root.

To build and use:

make
sh ./run.sh

The "run.sh" script accepts an optional configuration file in order to test specific areas.
For example:

./run.sh vfs.sh


To run all of the different test scenarios type:

./run.sh -a

You may have to tune the stress test to make sure that your test box run low
on resources.  For example:

INCARNATIONS=125 ./run.sh

The following sub-directories exists:

./doc		Documentation
./include	Include file for building the test programs in ./testcases
./lib		Common support files for the test programs
./misc		Various test scenarios. Mostly regression tests
./testcases	Test programs
./tools		Supplementary tools used in stress testing

The name stress2 indicates that this is the second generation of the Kernel Stress Test Suite.
The first version (stress) was based mostly on scripts.


20090120: des@ pointed out it would be nice to be able to set the
	random seed in order to be able to reproduce errors in time.
	A test of this show that the occurrence of panics are still
	non-deterministic even with a fixed seed. The known broken
	unionfs was used to test this:
	http://people.freebsd.org/~pho/stress/log/marcus006.txt.
	First panic occurred after 29 loops and the second after 8
	with a seed of "123".

	The randomness that is provided by lack of synchronization
	with timer interrupt, disk block layout and disk access
	latencies caused by platter and heads speed changes, disk
	firmware internal operations, and so on are _much_ bigger
	then the undeterminism caused by the non-repeatable random
	seed.

	Anyway, I have always view the this issue as only a slight
	inconvenience.