C
Curious George
Don't bother. Every troll says they aren't one. Don't worry. I
think you sound like a newbie instead, otherwise I wouldn't be
explaining so much. Of course I do admit making such assumptions is a
character flaw of mine.
Basically it's a waste of time, money, and effort _unless_ there is a
critical need to keep the data online and uninterrupted through sudden
catastrophic disk failure or upgrades. It is an "insurance policy"
for continuous service/uptime. It is not necessarily "more reliable"
in the sense of say MTBF, read/write verification, etc.
You haven't expounded upon why you think you need raid. So I'm trying
not to say explicitly yes or no to it. If you can make an argument
for it then FWIW, as I said before, I do think that generally one is
better off with a quality software raid than cheap PCI or ROMB RAID.
Furthermore if you really want to throw money at the reliability issue
(a la Arno) IMHO it makes more sense to look at a better controller or
better backup device rather than buy 2 cheap RAID HBAs.
That is often the case, esp with (s)ATA host-based software-assisted
hardware. But sometimes the data _can_ actually be easily retrieved
with a generic HBA. I don't know if that is the case for your SI
array. Since you have it, that's easy for you to test.
Frankly if one knows what one is doing, I don't see why you would need
anything more than a disk editor to make a RAID1 disk member readable
by a generic HBA that otherwise could not.
No, there are indeed benefits to raid. I use and implement it
regularly. But the benefits come with prices and the devil is in the
details.
For one thing, contrary to the theoretical ideal that ppl use to sell
raid, your data isn't always 100% safe on a mirror. True a drive can
often fail and be removed gracefully, leaving the mirrored copy up.
But occasionally a failing drive or controller can crap on the good
drive and you loose good data on both the failing and the healthy
mirror.
Another factor is the mathematical reliability of an array is
significantly less than a single drive. This occurs essentially
because it is more complex and there are more points of failure in an
array. Even though this is typically described via MTBF, which is a
problematic measurement, most ppl do not dispute the basic logic that
more complex storage has potential downsides regarding reliability.
Arrays are however sometimes described as _more_ reliable. That is
because many of these hardware problems do not affect uptime. However
it is NOT because array problems occur with lower frequency. The
array seems more reliable to the end-user because there are less
outages. However what the Administrator, the array, the wallet behind
the machine goes through behind the scenes is not factored in. Since
you are the owner and administrator, it bears looking into.
RAID protection is intended only for premature failure of redundant
parts. Lets say, for example, you buy 2 identical model drives, from
the same store and lot. You install them at the same time and due to
RAID 1 they get identical use. It would not be unreasonable to expect
that they would both die at similar times. This would not only result
in array failure if you couldn't get replacements in that small
window, but the array service life ended up being no greater than a
single drive. The devil is in the details.
So I ask the question, what about backups? Backups provide protection
from common types of loss that RAID simply can't. Backups allow you
to go back to an earlier, trusted point in time. RAID can't. Instead
Raid's advantage is graceful availability through certain hardware
problems. But this advantage is superfluous if you have ample time
windows when work or data access isn't being done or the work or data
has limited value.
Finally if we accept Arno's point that these types of low-end
controllers are prone to premature failure and are not very
compatible, the result is greater trouble, manual effort, and less
uptime compared to a single disk and a good backup. IMHO throwing a
second controller at the problem is misguided. It's poor planning and
a misallocation to buy 2 unreliable pieces of crap, because you don't
trust them, they are not intelligent enough to alert you BEFORE
failure, and you need that kind of a way out from problems because you
have bad computing practices and are not comfortable with low-level
recovery/repair.
As long as they are good racks, that works. But if you are serious
about media rotation and retention, media transportation, media
reliability, and shelf life, it actually end up not really cheaper
than some tape technologies, for example, even for smaller data sets.
