First, that is an excellent differentiation to make Eric -- although I
do sometimes use SOFTWARE RAID to MIRROR a hardware RAID drive set
(essentially creating a 3rd mirror) as a form of backup... not
preferred, but a valid (if expensive) way to get a point-in-time copy of
the filesystem. But to me, *the MOST important thing about backups *(to
tape, to hard drive, or to the cloud) *is that they *(or at least one
copy, if there are many) *MUST be located AWAY from the primary data
storage location! *
Storing your backup media in the same building as your primary data is
just begging for trouble! (Think fire, flood, meteor strike, etc!)
I also found it interesting, Eric, how I'm willing to pay for the HW
controller, but do NOT use RAID on my backup location, while you prefer
to save on the HW controller but spend on the redundancy in the backup
storage! Different histories create different priorities! :-)
If my backups fail, they get restored when the backup runs again)...
the only time I could get hosed would be if my backup drive(s)
failed and I had a catastrophic (RAID) failure at the same time --
and since they're a continent apart (geographically), I think the
risk is acceptable (given the added cost my hosting company applies
to using multiple drives!
As for my experience with HW cards, the Adaptec issue dates to the
mid-1990's & SCSI (I guess I should've made that more clear, perhaps --
yes, I've been in this business a long LONG time!)
My later issue with Promise was 2-fold: a) I wasn't aware it was
"Fake-RAID" until I'd already installed about a dozen of them, and b) in
spite of the "Fake-RAID" implementation, it was a solid performer and
did well when a drive failed... the only real "problem" was Promise's
failure to support it in a 2.6 kernel. (By the time it was natively
supported in the 2.6 kernel, I had moved on to 3ware, and as I've said
before, I've had nothing but EXCELLENT results there.) FWIW - I went to
Promise RAID for IDE drives when the Linux software raid (kernels 2.2
and 2.4) botched a few client's systems by continually trying to boot
off a broken array with the wrong drive (a known issue at the time --
caused by BIOS re-arranging of the device mapping when the drive went
away completely). I guess you could say I was gun-shy of Linux RAID
until I was forced to use it for a non-profit client who literally had
no money, but whose data (they are, after all, a MUSEUM) just could not
be risked!
But enough history - I need to suck up to the boss (er, agree with Eric)
more:
*Assuming a server-based SATA controller (with AHCI and NCQ enabled), I
would fully agree that the performance differences between my 3ware and
Eric's md (Linux) RAID would be /minimal /*-- at least on live (both
drives are healthy) systems. However, the performance I look at when
evaluating RAID (both technology and implementation) is the /degraded
/performance -- as that's likely to be the only measurement with
significant differentiation. For me, that's where the HW RAID excels...
and I absolutely LOVE it when I walk into a client's office and explain
to them that a drive failed in the server.... AND THEY NEVER EVEN NOTICED!
To be fair, I have never had a problem with Linux RAID in the 2.6
kernel, but I also only have a small handful of installs running that
solution.
Interesting discussion - I love to hear other people's views on these
topics, as it helps me to re-evaluate my own decisions and examine my
own reasons for doing certain things (its far too easy to fall into the
"I've always done it that way" trap!)
Dan McAllister
IT4SOHO
On 5/16/2012 12:31 PM, Eric Shubert wrote:
Let's be clear, that raid is *not* a backup solution. The purpose of
raid is to keep the system running in the event of a hard drive
failure. Sort of like a UPS does for power. You still need a backup /
disaster recovery solution above and beyond raid. I recommend off-site
backups (or archives if you prefer) with redundant drives there as well.
I do agree with Dan for the most part as he wrote before, however I
prefer a software raid implementation over hardware (as we've
discussed here recently). I think SW provides best reliability while
HW provides good reliability, and I'm apparently more frugal regarding
expense. While HW provides slightly better performance, I don't think
the improvement is worth the cost.
I'm a little confused why Dan rates HW as best reliability (previous
email) given the experiences he's had. ;) To be honest, I have very
little experience with HW raid, but I don't have any horror stories to
go along with it either.
On 05/16/2012 08:34 AM, Dan McAllister wrote:
I used to be a fan of ADAPTEC RAID, but TWICE I had an issue where a
RAID1 mirror recovered from the wrong disk, thus erasing TONS of data
that had to be recovered from backups (archives, to be more
language-correct). In both cases, ADAPTEC admitted that it was a flaw in
their firmware that re-numbered the disks without warning. (Admittedly,
both were issues where the hot-swap capability was not available, and so
the RAID card was booting cold with 2 drives and had to determine which
was the "good" drive -- not the most optimal method of recovering a
RAID-1 array!)
Back in the early 2000's, I had good success with Promise TX-1000's (for
IDE-based RAID1) -- but that is a "Fake-RAID" product, and while stable
in RedHat 5-9, it became less so in the RHEL era. I dumped Promise when
they reported that they had no plans to write drivers for the 2.6
kernel.
So, I switched to 3ware (now LSI) -- and I couldn't be happier! They
have given me a product (both the 9550 [PCI-X] and 9650se [PCI-E] lines]
that has performed FLAWLESSLY for me for both Windows AND Linux servers.
When (not IF) a drive fails, I can hot-swap in a new one -- even from a
different manufacturer -- and it rebuilds the RAID 1 (or RAID-10) array
without interference or input at all -- just plug and play!
I have also started to use the 9750 [also PCI-E, but 6Gbps SATA/SAS],
but I honestly don't have any experience with this card in a failure
mode, so I'm not in a position to review it.
My lowest-end servers use Linux md-RAID, and as with the 9750, I just
don't have enough failure experience to judge it....
Just my thoughts -- you know I like to share them! :-)
Dan
--
IT4SOHO, LLC
PO Box 507
St. Petersburg, FL 33731-0507
CALL TOLL FREE:
877-IT4SOHO
We have support plans for QMail!
