Testing direct IO with FIO

[Rodney Tortorich]

I'm trying to test block layer IO performance using FIO (v1.58 on rhel 5 and 6) and I'm not able to verify that the tool is actually performing non-buffered I/O.

In my job file I'm turning on the "direct" option on by setting it to 1 and I've also tried the "buffered" option as well.

According to the man page:

"direct=bool
              If true, use non-buffered I/O (usually O_DIRECT).  Default: false."

I do see a difference in the performance when I run the job with this option enabled vs disabled.

However when I strace it, the "O_DIRECT" flag is not set on the open() sys call as I would expect.

The open() in both non-buffered I/O and buffered I/O tests look identical, so why the difference in performance between the 2 tests?


Anyone have any thoughts or experience using this tool?

-Rodney
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.gnhlug.org/mailman/private/gnhlug-discuss/attachments/20120123/8e9015a5/attachment.html