RSYNC Fedora

[Douglas Furlong]

On Wed, 2003-11-26 at 06:11, Scott Burns wrote:
> Eric J. Feldhusen wrote:
> 
> > On Nov 25, 2003, at 9:40 PM, Scott Burns wrote:
> >
> >> Eric J. Feldhusen wrote:
> >
> >
> >>> A computer with 4 - 100GB hard drives in a RAID 5 (parity) set up, 3 
> >>> drives are active in the RAID 5, 1 hot spare drive, Total storage 
> >>> 200GB, actually a little less due to overhead. In this 
> >>> configuration, one of the drives in the raid fails, the hot spare 
> >>> becomes active, the raid rebuilds, and you replace the dead drive 
> >>> with a fresh drive and this drive becomes the hot spare.  In this 
> >>> case, you can have 2 drives fail and the system can keep going.
> >>>
> >> Same deal as above, what happens if a second drive fails during rebuild?
> >
> >
> > I haven't had it happen yet, but "in theory", with the setup above, it 
> > "should" continue to work, but you're really want to get another 
> > couple drives in that machine fast.
> >
> You might want to give this a try if you have a spare machine available. 
>  Fedora + software raid to make a 5 odd partition RAID5 set and then 
> kill one partition.  Midway through the rebuild with a hot spare, kill 
> another partition.  I'm certain that if RAID5 loses two of it's disks at 
> the same time, it's gone.
Morning all.

I just thought I would add my two pence worth.

Just to clear up the definitions (for my own benefit as much as any one
else).
RAID0 = Two (or more) disks with data spread over them, there is no
redundancy in this configuration
RAID1 = Two (or more) disks, one a mirror image of the other.
RAID4 = Parity data stored on 1 drive, data stored on the rest.
RAID5 = Parity data shared across multiple disks, allowing any one drive
to fail, and the raid to continue working.
RAID10 = RAID1 pairs, merged in to RAID0 (raid 0 has no redundancy)
RAID14 = RAID1 pairs, merged in to a RAID4 setup.
RAID15 = RAID1 pairs, merged in to a RAID5 setup.

In a RAID10 set up, if you have any two drives in a single RAID1
configuration, fail then the entire raid array is lost, as RAID0
provides no redundancy (but can provide performance improvements).

I can't be bothered to work out the chances of that happening, but it is
high enough to be a pain in the arse.

If however you have two drives fail that are in separate RAID1's fail
then the RAID0 will carry on working with out knowing any better.

I believe in a RAID 5 configuration, depending on the number of disks
present in the actual array, you can have multiple disk failures and for
the system to still be functional, so long as said disk failures are not
next to each other. If the disks are next to each other, then there will
always be a total failure, which is in my opinion is very similar to the
multiple disk failures in a RAID1 causing the entire RAID0 to fail.

I have not read the article, however I am afraid that I would be very
quick to dismiss the article if some one is comparing a 12 disk RAID10
set up with a 4 disk RAID5 configuration. That really is just plain
unfair, it is like saying this 12 man team can paint your house faster
then the 4 man team. Hardly a valid comparison.

If we had a 12 disk RAID15 configuration, then we would be in a position
where we could have,
Multiple failures in one RAID1 array (that array disappears) and the
RAID5 would be up and running (just slowed), In this cenario the RAID10
would fail.
Failures in multiple seporate RAID1's and the RAID5 should be
unaffected, just as with the RAID10.

I believe comparing RAID10, with RAID15, with the same number of disks
is considerably more logical then a RAID10 with 15 disks, to a RAID5
with 4 disks.

May I also point out that when we start dealing with these levels of
raid, it is not just reliability that is a factor, we start heavily
considering performance, in which case RAID14 with a battery backup on a
large ramdrive to hold the parity data may also be worth considering.

Dougie