I/O bound, what is it, and what can be done about it?

Joshua Murphy poisonbl at gmail.com
Tue Sep 13 21:43:54 PDT 2011


On Sun, Sep 11, 2011 at 8:51 PM, Nathan Coulson <conathan at gmail.com> wrote:
> Something I have been thinking about,  My computer can be near
> unusable if I am doing too much I/O tasks at once (let's say writing
> 0's to one drive, while running a backup on another [using two hard
> drives]).  Programs that do not use hard drives at all are almost
> unusable.
>
> [and if you are curious why I was doing that,  Was recovering data
> from a dying harddrive for a friend, and wanted to ensure the "new"
> ancient harddrive (40GB) I salvaged was stable].
>
> Compared to earlier systems,  my machine feels like a powerhouse (AMD
> 890GX w/ Integrated Graphics, Phenom ][ 945 processor [X4, 3GHz], 4GB
> 1333mhz Ram, 60GB OCZ Vertex 2 SSD).  Was surprised when I realized I
> can slow it down without even using the processor.
>
> although, i recall my AMD Duron 800Mhz system felt like a powerhouse
> as well...  but mycommon  workload seems to consist of openbox,
> firefox, urxvt, vim, and compiling software packages.
>
>
> What am I seeing?  What defines the limit of I/O Traffic on a system?
>
> --
> Nathan Coulson (conathan)
> ------
> Location: British Columbia, Canada
> Timezone: PST (-8)
> Webpage: http://www.nathancoulson.com
> --
> http://linuxfromscratch.org/mailman/listinfo/lfs-chat
> FAQ: http://www.linuxfromscratch.org/faq/
> Unsubscribe: See the above information page
>

For what it is, it's the I/O scheduler not sharing its toys with other
processes, from what I've found. As for what to do about it, you might
experiment with the different schedulers available in the kernel, see
which works best for your everyday workload, but something I've found
a fair bit more handy in practice... is mixing nice and ionice for
things that bring the system to a crawl. While both are likely to
increase the time to completion of whatever tasks it is you're running
with them, they do quite wonderfully when it comes to keeping a system
responsive, and if you want shortest time to completion more than
responsiveness, you will probably want to avoid interrupting it with
other processes anyhow.

-- 
Poison [BLX]
Joshua M. Murphy



More information about the lfs-chat mailing list