Re: fi massive testing

On Wed, Mar 05 2008, Zhang, Yanmin wrote:
> Jens,
>
> Thanks for the wonderful tool. I plan to use fio to replace most disk
> I/O benchmakrs I am using.
>
> Currently, I create more than 40 job files to test on JBOD.
> Unfortunately, with the massive disk I/O, some disks on my 2 old JBOD
> crashed after running for a couple of days.
>
> Mostly, I run aio/vsync/sync with all rand/read/write/rw and
> less4k/4k/64k/128k block size combinations. I also simulate some disk
> I/O like media play server and some database disk I/O benchmarks.

Glad to hear it's useful to you! If you have interesting workload
descriptions, please do pass them on and I'll add them to the example
section. Be sure to comment the workload so that others can figure out
what it does and why.

> Do you have any recommendations about the job file which could
> simulate application models?

Well, you could run your application and capture io actions with
blktrace, then later replay those traces with read_iolog=trace.bin. Or
look into the stuff that Joshua is playing with, there's lots of
possibilities.

-- 
Jens Axboe