On Tue, Dec 27, 2005 at 02:57:13PM -0500, Ron wrote:
Your quoted position is "there isn't a 'RAID 5 penalty' assuming
you've got a reasonably fast controller and you're doing large
sequential writes (or have enough cache that random writes can be
batched as large sequential writes)."
And you said that RAID 5 writes are slower than reads. That's a
completely different statement. The traditional meaning of "RAID 5
penalty" is the cost of reading a stripe to calculate parity if only a
small part of the stripe changes. It has a special name because it can
result in a condition that the performance is catastrophically worse
than an optimal workload, or even the single-disk non-RAID case. It's
still an issue, but might not be relevant for a particular workload.
(Hence the recommendation to benchmark.)
My experience across a wide range of HW, OSs, DBMS, and applications
says you are wrong. Given enough IO, RAID 5 takes a bigger
performance hit for writes than RAID 10 does.
I don't understand why you keep using the pejorative term "performance
hit". Try describing the "performance characteristics" instead. Also,
claims about performance claims based on experience are fairly useless.
Either you have data to provide (in which case claiming vast experience
is unnecessary) or you don't.
Said RAID 10 array will also be more robust than a RAID 5 built using
the same number of HDs.
And a RAID 6 will be more robust than either. Basing reliability on
"hopefully you wont have both disks in a mirror fail" is just silly.
Either you need double disk failure protection or you don't.
Mike Stone
