On Thursday 07 June 2007 16:08:54 Greg Freemyer wrote:
> On 6/6/07, Matthew Stringer <[EMAIL PROTECTED]> wrote:
> > Hi,
> >
> > I've a server with 24 500GB SATA 2 hard drives with RAID5 running 10.2.
>
> *** Warning, the hp trained storage engineer part of me is coming out.****
>
> A 24 disk raid5 is not normally recommended.  It is much more likely
> to fail than a raid50 where you are striping across several smaller
> raid5's.
>
> Even more reliable is raid6.  Personally I like raid10 for most
> situations.  raid10 is the most expensive, but the most reliable one
> highest performing.
>
> As an example, assume you have a raid50 made of 6 4-disk raid5 sets
> striped together.  Assume the chance of a first disk failure in a year
> is X.
> Assume the chance of a second disk failure during the window of
> vulnerability is Y.
>
> FYI: The window of vulnerability is however long it takes to replace a
> failed drive and rebuild the array.  On big arrays, just the rebuild
> time can be a couple days.  And if the array is both very large and in
> extremely heavy use 24/7 it can be a couple weeks.
>
> Back to the example:
> The overall failure likely hood is [(4 * X) * ((4-1) * Y)] * 6  or 72 * X *
> Y.
>
> Hopefully you can figure out where the numbers came from.
>
> Now for a single 24 disk array the failure likelihood is:
> (24 * X) * ((24 -1) * Y) or 552 * X * Y.
>
> That is the big raid5 array is roughly 7.5 times more likely to have a
> total failure than a raid50 with the same number of drives would be.
>
> (Admittedly the raid50 holds only 75% of raw disk capacity and the
> raid5 is going to hold a little more than 95%.)
>
> Given the price of drives, you might want to consider replacing all of
> those 500GB drives with 1000GB drives and doing raid10.  (about $500
> each I think).
>
> Then you have the same capacity and the failure likelyhood would be:
> 24 * X * 1 * Y   (ie any of the 24 can initially fail, but only its
> partner failing causes a data loss).
>
> So your online data would be about 20 times safer for $12K.  If I had
> that much data, I would seriously consider what raid strategy I would
> use, but I really doubt I would end up choosing a 24-disk raid5.
>
> Greg
> --
> Greg Freemyer
> The Norcross Group
> Forensics for the 21st Century

All you say is true however, budget dictates, I needed >10TB of storage for 
£6K.
You're suggesting spending another £12K on new drives alone! If I'd bought 
larger disks and used RAID50 I'd have half the disk I/O too which isn't 
desirable.

In my case the data isn't that critical hence the solution I put in place, 
which incidentally with OpenSUSE 10.2 and XFS has been spot on!

Matthew




 
--
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]

Reply via email to