> -----Original Message-----
> From: Frank Joerdens [mailto:[EMAIL PROTECTED]]
> Sent: Monday, March 13, 2000 9:02 AM
> To: Chris Mauritz
> Subject: Re: IDE hardware RAID
>
>
> On Mon, Mar 13, 2000 at 08:40:57AM -0500, Chris Mauritz wrote:
> > I'm going to start using them for applications where the
> data is expendible
> > (mp3 jukebox comes to mind) where I just want a cheap RAID
> 0 array to hold a
> > big chunk of bits. Another good application might be large
> web farms where
> > the data on any one web server can go poof without any
> problems and you can
> > then stripe a pair of inexpensive IDE disks for better
> throughput on a
> > budget. When you're talking 10's or 100's of servers, the
> savings adds up.
> > With 40gig ATA66 drives going for under $300 these days
> (and 10gigs in the
> > $100 range), it seems like a no brainer in that kind of
> situation. I'd
> > never use it on a busy database box, but the economics are
> quite compelling
> > in certain situations.
>
> You're only referring to the RAID 0 option you have with
> these devices.
> What about RAID 1 and 10? I think particularly the latter is a good
> option for low-end low-cost servers that still require some
> redundancy.
It depends on what you're looking for. RAID 10 is generally the best
solution if you need redundancy AND speed. If you need redundancy and
maximum storage capacity, RAID 5 wins out. If you need just speed, RAID0,
and if you need maxuimum redundancy, then RAID1. Here's an example, using
9GB disks. 4 disks for all solutions. RAID 10 gives you ~18GB of storage,
a mirror of two stripes (or is that a stripe of two mirrors?). RAID 5 gives
you ~27GB of storage. RAID 0 gives you 36GB of storage, and RAID 1 gives
you 9GB of storage (unless you assume a true mirror, in which case it gives
you 9GB of storage with 2 hot spares). You can invent layouts of other RAID
combinations for speed and or redundancy, but those are the biggest ones.
> If you have a RAID 10 array comprised of 4 disks (i.e. a mirrored
> striped pair), the likelihood of losing the entire array (i.e. either
> one from each mirrored pair, or both from either plus one from the
> other, or all 4 failing simultaneously) - if my probability theory is
> not entirely rusty - is (2n+1)squared divided by n to the
> power of 4 if
> the probability of losing a single disk is the inverse of n
> (how do you
> write this stuff properly using just ascii?)
You don't. :) How this:
[(2n+1)^2]/(n^4)
> which approaches the
> inverse of n squared if n is large enough; which means that since n is
> pretty big (the probability of one of today's IDE drivers failing is
> pretty low), RAID 10 is not significantly worse than RAID 1
> in terms of
> inherent redundancy, whilst being significantly faster.
Greg
