On Sat, 8 Apr 2000, Mike Bilow wrote:
> Why on earth would you even bother with RAID on a machine which can
> experience 30 minutes of idle time? We only use RAID for machines which
> run web servers, mail servers, name servers, and things like this. Such
> machines here never exeperience even a solid minute of idle time.
>
> -- Mike
Look at what "problem" RAID was designed to solve and you'll see the
answer to your question. RAID exists to provide higher performance,
higher reliability, and greater scalability than single large disks.
I can think of a very good example of a system with 30 minutes of idle
time: a fileserver serving up a data disk which is used frequently
9-5 while the staff is in the office, but only sporadically during the
night.
Now, the question really is "should my RAID disks spin down"? Well,
why do disks spin down? To increase longevity? To decrease power
consumption? I can't see either of those being valid arguments in a
RAID environment because 1) individual drive longevity is not important in
RAID- when a disk fails you replace it and 2) if you're doing something
important enough for RAID, you can probably afford the extra power draw
:-)
Technically, I could see a setup where a RAID controller could monitor
activity, and if it hadn't been active for quite some time, spin the whole
cluster down. Since the machine sees the disk "through" the cache, and
blocks while waiting for data (don't know the timeout) there is no reason
you can't just have the cache waiting for the drives to spin up. Spin up
on modern drives is about a second, and if nobody has used the machine in
two hours it's not unreasonable for a 1-second latency for the system to
come back up, provided that it provides full response after that.
