On 01/02/2011 01:45, Phillip Susi wrote:
On 01/31/2011 06:07 PM, Piergiorgio Sartor wrote:
That's interesting information. I suspected something
like async access could have happened.
Is this confirmed? Are the requests from md to the
device layer asynchronous?
Yes, it sends down requests to all drives simultaneously to achieve
maximum performance. This is the main purpose of raid-5.
How about libata? Does this serialize wake ups?
It doesn't. It will wake them all up at once.
If you want to force serial wakeup, you would need to do some hacking in
libata or the scsi layer to only allow one wakeup request at once. Of
course, with a 6 disk raid where each disk takes 6 seconds to spin up,
you will be waiting 36 seconds to get your data. Not exactly desirable.
That'd be an excessive amount of time to wait. A quarter of a second is
more than enough, a tenth of a second would probably be enough. It's
just the motor inrush current you're trying to avoid having simultaneously.
So waiting another second for your array to wake up would mean you could
use a sensibly-sized PSU operating in its 80%+ efficiency range, rather
than a huge PSU operating inefficiently.
Cheers,
John.
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
