Disk Failure on Raid 5

Can a disk failure on Raid 5 cause torn pages? It actually did cause
torn pages on several databases for us, so I am wondering if we have
some kind of configuration problem.
ThanksA torn page basically reads the last two bits of a page that is written to
disk. If you have a hardware issue as the page is written to disk you always
have the potential to have part of the data written and part not. I think
anytime you have a disk failure you run the risk of a torn page.
--
Andrew J. Kelly SQL MVP
"Dave" <daveg.01@.gmail.com> wrote in message
news:1136847162.322252.52170@.o13g2000cwo.googlegroups.com...
> Can a disk failure on Raid 5 cause torn pages? It actually did cause
> torn pages on several databases for us, so I am wondering if we have
> some kind of configuration problem.
> Thanks
>|||Is there anything we could do to prevent this or is it just a fact of
life?
I guess I don't understand why the data page would not have been
written correctly to the redundant drive.|||I'm no storage expert, but I would ask the storage vendor whether the RAID system fulfils Write
ordering and other aspects mentioned in
http://www.microsoft.com/technet/prodtechnol/sql/2000/maintain/sqlIObasics.mspx.in case of a drive
failure.
--
Tibor Karaszi, SQL Server MVP
http://www.karaszi.com/sqlserver/default.asp
http://www.solidqualitylearning.com/
Blog: http://solidqualitylearning.com/blogs/tibor/
"Dave" <daveg.01@.gmail.com> wrote in message
news:1136912372.389395.67820@.g47g2000cwa.googlegroups.com...
> Is there anything we could do to prevent this or is it just a fact of
> life?
> I guess I don't understand why the data page would not have been
> written correctly to the redundant drive.
>|||Well a RAID 5 does not actually have a redundant drive. All the drives are
written to with a small piece of the data. One of the drives holds the
parity while the others each get a piece of the actual data. So if any one
piece is missing it can rebuild the data with the parity. But that does not
mean you can not get corruption on the write especially during a hardware
failure. I won't claim to know how the drive controllers work internally and
how they each do their stuff. So I am not sure other than ensuring you have
good name equipment and all in proper working order. Especially the UPS.
--
Andrew J. Kelly SQL MVP
"Dave" <daveg.01@.gmail.com> wrote in message
news:1136912372.389395.67820@.g47g2000cwa.googlegroups.com...
> Is there anything we could do to prevent this or is it just a fact of
> life?
> I guess I don't understand why the data page would not have been
> written correctly to the redundant drive.
>|||Guys, I guess I am just not getting it. I thought that RAID 5 was
redundant, meaning it should not be affected by a disk failure. The
data should be stored on 2 drives right? So how can you get torn page
errors when a disk fails?
Can someone please step me though a scenario of how data can be
corrupted with a RAID 5 disk failure?|||No that is not how a RAID 5 works. If you have 4 drives in a RAID 5 you
will essentially split the data into 3 pieces. One of each of the pieces
will go onto 3 of the drives and a parity is calculated and placed on the
fourth drive. Each time you write to the drive array this is repeated but
the parity moves around so it is not always on the same drive. Under normal
conditions when you read the data the parity is not used and the whole data
block is created by piecing the three pieces back together. In the event of
a single disk failure the controller can read the two remaining good pieces
and using the parity recreate the third to get the data back. A Raid 5 does
not store the data twice. But even if it did that still does not prevent
torn pages. As I mentioned a torn page occurs when for some reason (usually
hardware related) the last two bits on a page did not get written properly
or at all. This can happen when the driver thinks it wrote the page
properly but the hardware didn't. A Raid 5 array does not claim to stop this
from occurring. That is why backups are so important. You can not protect
your data 100% with a Raid array.
--
Andrew J. Kelly SQL MVP
"Dave" <daveg.01@.gmail.com> wrote in message
news:1137002846.251710.290370@.g44g2000cwa.googlegroups.com...
> Guys, I guess I am just not getting it. I thought that RAID 5 was
> redundant, meaning it should not be affected by a disk failure. The
> data should be stored on 2 drives right? So how can you get torn page
> errors when a disk fails?
> Can someone please step me though a scenario of how data can be
> corrupted with a RAID 5 disk failure?
>|||Thanks, that helps a little.
I am still having a hard time grasping parity.
I will use Raid 3 for simplicity.
Disk 1: 00000000
Disk 2: 11111111
Disk 3: ''
On Raid 3, if Disk 3 stores parity data, what would it store? I
don't understand how one drive could store enough data to rebuild
Disk 1 or Disk 2.|||This should explain Parity:
http://www.pcguide.com/ref/hdd/perf/raid/concepts/genParity-c.html
This shows how raid 3 (and others) use parity.
http://www.storagereview.com/guide2000/ref/hdd/perf/raid/levels/singleLevel3.html
This tells you why you may want to consider something other than RAID 5.
http://www.baarf.com/
Andrew J. Kelly SQL MVP
"Dave" <daveg.01@.gmail.com> wrote in message
news:1137006115.234760.166690@.g44g2000cwa.googlegroups.com...
> Thanks, that helps a little.
> I am still having a hard time grasping parity.
> I will use Raid 3 for simplicity.
> Disk 1: 00000000
> Disk 2: 11111111
> Disk 3: ''
> On Raid 3, if Disk 3 stores parity data, what would it store? I
> don't understand how one drive could store enough data to rebuild
> Disk 1 or Disk 2.
>|||How would RAID 10 affect this scenerio?|||I don't think it would matter what raid level it was. While Raid 10 does not
use parity it still writes to the disk and any write has the potential to
have a torn page.
--
Andrew J. Kelly SQL MVP
"JLA" <info@.jlaenterprises-dot-com.no-spam.invalid> wrote in message
news:43c5acee$0$17777$c3e8da3@.news.astraweb.com...
> How would RAID 10 affect this scenerio?
>|||Thanks! that was some good reading!
I think I am going to push for Raid 10. :-)