On Mon, Aug 16, 2010 at 12:29 AM, Florin Iucha <florin at iucha.net> wrote:

> On Sun, Aug 15, 2010 at 02:26:06PM -0500, Robert Nesius wrote:
>
> What?  If I ask "get this on the platter", it better be on a platter,
> not in the cache of some smart-less drive, waiting for the power
> failure.
>

I think you missed my point here. I was talking about writes, not reads,
btw.


>
> >                                     That's assuming you didn't match
> cache
> > sizes, which most people likely would have the sense to not do even if
> they
> > were mixing/matching drives/vendors.
>
> I'm not sure what you mean here.
>

Yep, you missed it.  But my example here is... not a big deal.  The larger
point has already been said better by someone else - drives from different
manufacturers are not guaranteed to behave precisely the same way.  That
matters more for some RAID types than others I guess.



>
> > My experience in large-enterprise installs is that everyone uses the same
> > drives within their arrays and disk cabinets.
>
> My experience with large-scale restaurant operations is that the food
> is crap.


I'm talking about large-enterprise computing.  Not restaurants.


> What would happen to an admin if 3 out of the 6 drives in a
> cabinet fail?


The cabinets I'm used to have 350+ spindles so ... probably not much.

What would happen to the manager who signed the
> purchase order for 12 identical drives?


Why should anything happen to him?


> Nothing -- failure has no
> consequences.  (I am ignoring the 'find the tape and restore the
> stuff'.  Downtime is real money, it must come from somewhere.)
>

If a drive fails in the woods and no one sees it, is data actually lost?


> > Lastly, diversifying drive as a risk-mitigation strategy can have the
> > opposite effect - one of the mix-in products may be far worse than the
> > others.
>
> So you'd rather have 6 identical crappy drives, or a prayer's chance
> that at least half are good?


I'd rather have six good drives versus a stronger assurance of 3 bad
drives.  If my quality metrics show a statistically significantly higher
failure rate for the brand I'm using, I'll try a different one.

Btw, let's say you have 10,000 spindles going... are you going to make sure
every cabinet, every drawer, and every array is going to have the same
mix/distribution of drives from different vendors?

But even talking about just apples (a raid array with six drives) versus
oranges (10,000 spindles) I'd still be inclined to keep the hardware the
same as much as possible.  But maybe in a domain with that few spindles it
makes sense to diversify.  The thought had never crossed my mind before -
that's why I asked what people thought.

-Rob
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.mn-linux.org/pipermail/tclug-list/attachments/20100816/3d648a4b/attachment.htm