This tool generates an I/O workload which stresses the filesystem cache. It can be configred for reading or writing, and in either a sequential or random pattern.

arcstress [-r] [-w rand | seq] [-s <file size>] [-n <number of files>]
    [-d <dir>]

-r: Read file(s) instead of writing them.  Default is writing
-w: Specify type of workload.  Default is sequential
-s: Size of each file to read/write.  Default 10MB.
-n: Number of files to read/write.  Default is 10.
-d: Directory in which to read/write files.  Default is $PWD/ab.$PID.

If the tool is configured for reading, it's not necessary to specify the file size, as the tool will stat each file to determine its size.

In addition, when the random workload is used, the tool will run forever. With the sequential workload, the tool will iterate through each file in numerical order, and exit once it has read/written each file.

This tool is a work in progress, and there is some duplicated code which can be refactored. In addition, more tunables are needed to tune the file size, I/O size, workload type, and files/directories used in this test.

This tool was derived from iopattern (http://github.com/pijewski/iopattern). At some point, I may recombine the funcionality of both tools into one single tool.