[mythtv-users] Hard Drive reliability esp. RAID issues
stef.coene at docum.org
Mon Mar 8 18:52:40 UTC 2010
On Monday 08 March 2010, John Drescher wrote:
> >> I would say from experience that this is very unlikely that a second
> >> drive will fail in the 8 hours it takes to rebuild a 6TB array. And
> >> even if it did with linux software raid you could recover. With HW
> >> raid you will probably need specialized software to force the array to
> >> except drives that are marked bad.
> > The problem is not that a second drive will fail. The problem is,
> > statistical speaking, that 1 bit of 10 TB is unreadable, the so called
> > UER (uncorrectable error rate).
> > See this document from 2005 http://arxiv.org/pdf/cs/0701166
> > This means that if you have to rebuild a 10 TB raid, you have to read 10
> > TB to calculate the parity bit. But 1 bit will not be readable, so bye
> > bye raid 5.
> > What if you have a 10 TB disk? Then a simple mirror will not be enough
> > because you can never read the disk without missing 1 bit.
> I have over 30TB of software raid 5/6 and I have never had a single
> unreadable bit bringing down a raid.
I manage over a few petabyte of raid 5 and I hever had a unreadable bit
bringing down a raid.
This is enterprise stuff. The raid controllers are constantly reading and
checking the raid sets. When a URE is found, the controllers will recalculate
the missing bit. Only on a rebuild, an URE can bring down a raid.
The fact is that the URE of a disk is 1 in a few TB. So you have a unreadable
bit somewhere on your raid. I just hope that you will never need it when
doing a rebuild.
More information about the mythtv-users