stress is not a benchmark, but is rather a tool designed to put given subsytems under a specified load. Instances in which this is useful include those in which a system administrator wishes to perform tuning activities, a kernel or libc programmer wishes to evaluate denial of service possibilities, etc.
The format for running the stress program is:
stress [option [args]] ...
stress supports the following options:
% stress --vm 2 --vm-bytes 128M --vm-hang
Note: Suffixes may be s,m,h,d,y (time) or k,m,g (size).
The simple case is that you just want to bring the system load average up to an arbitrary value. The following forks 13 processes, each of which spins in a tight loop calculating the sqrt() of a random number acquired with rand().
% ‘stress -c 13’
Long options are supported, as well as is making the output less verbose. The following forks 1024 processes, and only reports error messages if any.
% ‘stress --quiet --cpu 1k’
To see how your system performs when it is I/O bound, use the -i switch. The following forks 4 processes, each of which spins in a tight loop calling sync(), which is a system call that flushes memory buffers to disk.
% ‘stress -i 4’
Multiple hogs may be combined on the same command line. The following does everything the preceding examples did in one command, but also turns up the verbosity level as well as showing how to cause the command to self-terminate after 1 minute.
% ‘stress -c 13 -i 4 --verbose --timeout 1m’
You can write a file of arbitrary length to disk. The file is created with mkstemp() in the current directory, the default is to unlink it, but unlinking can be overridden with the –hdd-noclean flag.
% ‘stress -d 1 --hdd-noclean --hdd-bytes 13’
Large file support is enabled.
% ‘stress -d 1 --hdd-noclean --hdd-bytes 3G’
You may copy this manual under the terms of the GNU General Public License.