Power Consumption

Mark Hahn hahn at physics.mcmaster.ca
Fri Sep 14 08:14:25 PDT 2001

> Should, one set an idle power down timeout for all your IDE 
> drives in the cluster?

I don't think so, though it might make sense in certain special cases.

> Does spinning down an IDE drive reduce MTBF?

MTBF is usually quoted for power-on hours; merely spinning down the platters
probably doesn't effect this at all, since all the electronics are powerd-on.
obviously, MTBF really should be given for the whole range of activities:
you'd certainly expect a disk to fail sooner if you keep seeking between 
the same two tracks, for instance.

> How much stress is really incured during spinup and/or spindown?

wherever I've seen start/stop cycles rated, the specs offer O(50K) cycles
presumably during the warranty period (3-5 years).  that's only 45 per day
for a 3 year lifespan!

> The logistics of this becomes important, especially when you have
> hundreds of IDE drives deployed in a large cluster :)

most ide drives idle at around 5W; I'm not sure power is a serious
argument here.

in short, I'd say that for reliability reasons, you wouldn't want to cycle
drives more than a few times a day.  saving ~4W doesn't seem like a big deal
to me, even if you have hundreds of drives.  otoh, there are lots of windows
machines out there that default to spinning down after an hour or so...

regards, mark hahn.

More information about the Beowulf mailing list