Re: SCSI Disk/Controller advice please
On Sun, 31 Oct 2004 15:31:03 +0000, Joao Clemente <jpcl@rnl.ist.utl.pt> wrote:
> Hi Paolo, Alvin, Pigeon, Ron & Tim, thanks for all the replies...
>
> Paolo Alexis Falcone wrote:
>
> [snip]
> >>Which are the tradeoffs of hard vs software raid1? What happens/How do
> >>we proceed if 1 disk fails (how do we know it, how do we replace/resync
> >>them?)
> >
> [snip]
> >
> > Do note though that RAID 1 won't help you that much - it's better if
> > you could try higher RAID levels (RAID 5) for data integrity. RAID 1
> > will only mirror disks - and that would also mean should there be
> > errors in one disk it gets propagated to the mirror as well.
>
> Alvin, first off all I'm aware of high availability solutions (I've done
> my master thesis on those setups), but together with HA solutions we can
> use RAID anyway...
> Alvin and Paolo, I'm quite stunned with these claims that "errors on one
> disk will be propagated to the other when using RAID1". It still makes
> no sense to me that something like that could happen. Quoting Tim:
> "
> >>problem with raid1 ( aka mirror )
> >> - if one disk goes bad, the other disk will copy that bad info
> >> onto the good disk .... the whole point of mirror, both disk
> >> is identical
>
> Completely false. Physical disk errors mirrored by raid? No, No,
> No. Fat fingered deletes? Yes.
My bad. I made a glaring misconception here.
Here's the case here - RAID 1 indeed does not mirror physical disk
errors (else there's no real point in using RAID at all). However,
should there be errors in the disks during reconstruction of the RAID
array, RAID 1 won't save you as the errors would propagate anyway.
RAID 5 alleviates this by using parity information stored across the
disks - now it takes more than 1 disk failure for RAID 5 to fail.
> Paolo, as far as I understand your statements, you state this behaviour
> (suposing that it does happen) does not happen with RAID5. Why? With
> RAID5 you "checksum" data and in RAID1 you mirror sectors?
> I've googled for these problems you claim in RAID1 and haven't found
> nothing stating that these things could happen!
See above. The problem with RAID 5 is this - the benefits doesn't
really match the costs. You get additional checking but at a very high
cost (as additional space are used to store parity information, and it
takes more than two disks to implement RAID 5). Some alleviate this
problem by combining RAID 1 with RAID 0 as this is somehow an
acceptable trade off between economics and performance.
> Altough I'll not be going to SCA (as it appears to add somewhat
> significant $$$ to my environment where I don't neet 24/7 availability),
> just confirm this: There is no SCA controllers. The controllers have
> 68 pins wich connect to the hot-swap rack (wich will also receive power
> from a regular power cable) and the hot-swap rack will have the sca
> connector to connect to a sca disk. Is that it?
SCA is quite useful if you need hotswapping of SCSI disks. If that
isn't the case - there's not much economic incentive in purchasing
them.
--
Paolo Alexis Falcone
pfalcone@gmail.com
Reply to: