Searching \ for '[EE]:: Whr R.A.I.D. is a terrible idea' in subject line. ()
Make payments with PayPal - it's fast, free and secure! Help us get a faster server
FAQ page: www.piclist.com/techref/index.htm?key=whr+raid+terrible
Search entire site for: ': Whr R.A.I.D. is a terrible idea'.

Exact match. Not showing close matches.
PICList Thread
'[EE]:: Whr R.A.I.D. is a terrible idea'
2008\02\24@025455 by Apptech

face
flavicon
face
He argues that RAID typically reduces reliability.
ie done well it can be an improvement, but ... .

Makes some sense.
What do others think

   www.pugetsystems.com/articles?&id=29

2008\02\24@033955 by William \Chops\ Westfield

face picon face

On Feb 23, 2008, at 11:53 PM, Apptech wrote:

> He argues that RAID typically reduces reliability.
> ie done well it can be an improvement, but ... .

For Raid0 (striping), he's clearly right.  Double
the potential for failure.

For Raid1 (mirroring), I'm less sure.  He seems to be comparing  
"real" raid experience with "ideal" backup plans, and I'm not sure  
about his claim that a disk failure in a raid1 "means a long  
frustrating hassle."  (It's not supposed to, is it?  Swap out the bad  
drive, add in the new drive, let the mirror re-build?)  And I suspect  
we're all aware of just how difficult it is to maintain that 'ideal  
backup plan.'

And then there's the disclaimer of "servers are different."  When  
does my desktop become a server, anyway?  Our household has at least  
five computers, of which three are moderately "serious", and I tend  
to back them up each to another one.  Does that make them all  
servers?  I'm starting to think about a new desktop for myself, and  
was thinking of a raid 5 array (it's looking less likely; I was  
hoping for a software raid5, and it doesn't look like that's actually  
supported.  Sigh.)

BillW

2008\02\24@040037 by Peter Todd

picon face
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On Sun, Feb 24, 2008 at 08:53:34PM +1300, Apptech wrote:
> He argues that RAID typically reduces reliability.
> ie done well it can be an improvement, but ... .
>  
> Makes some sense.
> What do others think
>
>     http://www.pugetsystems.com/articles?&id=29

I've had much better experiences than that using software raid on Linux.
Saved many hours of restoring from backups twice now. But then again,
that particular client has 750gb of data, stored on a RAID1 array, with
another two 750gb drives for live backups, one of which is stored
off-site. Not to mention the yet another automated off-site backup for
the 15gb of "really critical" data we have. Restores are testing every
once in awhile, like when I accidentally delete a file...

The point being, we know what we're doing, raid is just to save me a few
hours of downtime and annoying backup restoration when the inevitable
failure happens.

- --
http://petertodd.org
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)

iD8DBQFHwTDb3bMhDbI9xWQRAvxXAKCo1Ew9/alii3KTIoVuOJCwNW6SQgCfcSmf
MN/O6gVwa6EuN5zogS/glDw=
=tNCK
-----END PGP SIGNATURE-----

2008\02\24@053500 by Jake Anderson

flavicon
face
Apptech wrote:
> He argues that RAID typically reduces reliability.
> ie done well it can be an improvement, but ... .
>  
> Makes some sense.
> What do others think
>
>     www.pugetsystems.com/articles?&id=29
>  
From his perspective i can see the downsides.
Trying to talk some monkey through rebuilding his raid array when what
he really wants to do is play world of warcraft could be a frustrating
experience.
On the other hand for an it pro to rebuild an array when it has gone
splat just after he has unraveled some arcane looping bug in some
software would be an almost relaxing task.

I know i was setting up a system for a client once, Linux Raid 1
software with mdadm. I literally was doing a final reboot just to make
sure everything came up the way i wanted before taking it to the clients
site when smoke came pouring out of one of the drive bays. I happened to
have a spare drive of the same capacity and about 10 minutes later the
system was back up and running rebuilding the array.

The array saved be about 8 hours work that no sane backup scheme would
have saved. (That array has been running with 0 errors,reboots or
downtime for 445 days so far for an office of 15). They also have a
simple backup system in place that just dumps copies of all their
important files onto an external usb drive each night, which the
secretary then hurls down a flight of concrete steps on her way home
before microwaving it (well thats what i assume happens) each week. She
seems surprised when sticky taping the housing back together didn't make
it all work again.

Eventually i hope to upgrade them a nightly offsite backup of an image
of their systems so they can be back up and running much faster in the
event of a failure.

2008\02\24@084340 by Gerhard Fiedler

picon face
William "Chops" Westfield wrote:

>> He argues that RAID typically reduces reliability. ie done well it can
>> be an improvement, but ... .
>
> For Raid0 (striping), he's clearly right.  Double the potential for
> failure.

Agreed.

> For Raid1 (mirroring), I'm less sure.  He seems to be comparing "real"
> raid experience with "ideal" backup plans, and I'm not sure about his
> claim that a disk failure in a raid1 "means a long frustrating hassle."
> (It's not supposed to, is it?  Swap out the bad drive, add in the new
> drive, let the mirror re-build?)  And I suspect we're all aware of just
> how difficult it is to maintain that 'ideal backup plan.'

Also agreed.

I don't think it makes sense to talk about "RAID" in the generality that
he's using.

First, there are many different RAID setups, with different objectives.
It's somewhat comparable to saying "paralleling MOSFETs is not good".
(Perhaps... you know, analogies... :)

Then, there is hardware RAID and software RAID. For hardware RAID, you
definitely need think carefully (and up front) about how to replace your
controller once it's broken. It's now part of the necessary infrastructure
to access your data. Software RAID doesn't suffer from that drawback.

And third, RAID is no replacement for a backup strategy, it's a complement.
It doesn't help you get the file back that you accidentally deleted or
changed.

The harddrive failure rates he cites are bogus (or at least incomplete).
What does "%" mean? Harddrives do fail, and you never know when it's time
for yours. May be today, or in twenty years. Or later still -- but you
don't know until it fails. And a RAID1 (or similar redundant config) array
lets you continue your work without (immediate) interruption -- and only a
short delay at a convenient time for replacing the defective drive and
initiating the rebuild of the array. In the meantime (while one drive is
broken), it's no worse than a normal drive.

I'm a fan of software RAID1 (integrated into a decent backup strategy, of
course). Very little extra cost for a huge increase in peace of mind.

> I'm starting to think about a new desktop for myself, and was thinking of
> a raid 5 array (it's looking less likely; I was hoping for a software
> raid5, and it doesn't look like that's actually supported.  

I'm not sure software RAID5 makes sense. There seems to be a lot of
overhead involved with RAID5. I also think that the advantages of RAID5
compared to RAID1 for normal desktop/home server use are not significant.

Gerhard

2008\02\24@171324 by PICLIST

flavicon
face
What I think he is missing here is: "What is the cost of lost data?"

He sees a cost of installing RAID systems for customers as increased support
calls. Without RAID, he has only a fraction of the support calls and if a HD
actually breaks, he replaces the drive (no cost as it is warranted by the mfg),
and does re-install of the OS and apps he originally installed and he's done. So
what if the customer just lost a bunch of valuable data? The occasional failure
costs him maybe 3-4 hours of tech time.

No cost is assigned to the data that is lost, or additional software that has
been installed and configured since the customer bought the machine. The cost of
the data lost can be from nearly zero (high scores on a couple of games) to HUGE
(a company's complete accounting system, the novel you just wrote, the images
from a professional photo-shoot, the PCB design and code for a product, all of
the above, etc).

In my mind (I'm obviously an old IT guy whose experience goes way back to
mainframe days), the data on the system is the only thing of real value and
anything that reduces the chance of data loss is a good thing.

Additional complexity? A RAID system is more complex than no RAID, but it is
automated and always backing up your data. A backup plan is also complex, with
many more human based things that can go wrong (forgot to do it for the last
month, staff changed a year ago and didn't know they were supposed to do it, "I
thought that message was normal - its been happening for over a year.", etc).

I think RAID-1 systems (mirroring) are pretty good for protecting against
hardware failure in the drives. RAID-0 (striping or volume concatenation) is a
dangerous thing that must be considered carefully for actively modified or
valuable data. RAID-5 can be good if you have a high-end RAID controller with
non-volatile memory - otherwise the write performance can be dismal, and the
risk of data loss significant.

-----Original Message-----
From: spam_OUTpiclist-bouncesTakeThisOuTspamMIT.EDU [.....piclist-bouncesKILLspamspam@spam@MIT.EDU] On Behalf Of
Apptech
Sent: Sunday, February 24, 2008 12:54 AM
To: PIC List
Subject: [EE]:: Whr R.A.I.D. is a terrible idea

He argues that RAID typically reduces reliability.
ie done well it can be an improvement, but ... .

Makes some sense.
What do others think

   www.pugetsystems.com/articles?&id=29

2008\02\24@172754 by David VanHorn

picon face
But what happens when the OS goes nuts, and tells the RAID controller
to write crap all over your data?  Don't you just end up with a
redundantly imaged disk full of crap?

For myself, I copy data from the working machine to a server, and then
to an offsite backup server.   I'm also partial to putting critical
things onto thumb drives which can be easily unplugged.  Pretty hard
for a crashing OS to hose up data sitting on an unplugged drive.

2008\02\24@180124 by peter green

flavicon
face

> What do others think
>
>  
Raid done right by someone aware of the limitations makes a LOT of
sense. Raid done wrong can reduce your reliability while at the same
time lulling you into a false sense of security.

I agree with him that fakeraid controllers are a bad idea, there is the
risk of some OS not detecting your drives are an array and doing bad
things because of that and there is the problem that products at that
end of the market typically do not come with good quality diagnostic and
repair tools.

Real hardware raid controllers can be a good choice but you MUST keep an
identical spare and get familiar with the recovery procedures to the
point that you are confidant you can get the array up on the spare
controller without losing data before putting them into service. This is
especially true if you go for striping/parity based raid (with raid1 if
worst comes to worst you can probablly find software that will search
for the raw data on the drive and just ignore the raid headers).

Linux software raid can also be a good choice, it's well engineered and
hardware independent, the downsides are you can't boot off it, you can't
use it for other operating systems and it probablly won't perform quite
as well as a real hardware raid controller (though for raid 1 there
probablly won't be much difference unless your system is IO bus
bandwidth starved).

Also as drives get bigger the proabability of secondry failure during
rebuild is going up. So if you are going for a parity based raid it may
be more sensible to go for a more advanced parity system like raid6. You
might want to consider a hot spare as well if you can't gaurantee fast
drive replacements.

Finally raid must be considered as an extra to not a substitute for
backup. While hard drive failure is a threat to your data there are many
other threats which raid does not protect against.

I remember a story of a server with a moderate sized raid setup which
the owners thought meant they didn't have to backup. A psu fault put
mains on the DC rails and fried the controller boards on all the drives
at once. They got thier data back by using a data recovery firm but iirc
it cost them something like £10000 .

2008\02\24@230410 by Peter Todd

picon face
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On Sun, Feb 24, 2008 at 11:01:00PM +0000, peter green wrote:
> Linux software raid can also be a good choice, it's well engineered and
> hardware independent, the downsides are you can't boot off it, you can't

Actually that's not true anymore. I don't know all the details of how it
works, but RAID1 is bootable on Linux now, even your /boot partition can
be RAID1. Since both drives are essentially a simple mirror the code
is probably not too difficult. I've got two machines setup that way with
Debian etch.

http://somedec.com/downloads/howto-bootable-linux-raid1.html

- --
http://petertodd.org
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)

iD8DBQFHwj2a3bMhDbI9xWQRAsUVAJoCUkLQP6Ex8C71VZ0De1sEIpwngwCfVH2M
y/MAjNKomhWbJot4msXm6+I=
=IZ4b
-----END PGP SIGNATURE-----

2008\02\24@233645 by Carl Denk

flavicon
face
For raid check out the Startech PCIIDE100R PCI raid card. It's bootable,
and has 2 IDE connectors. I use it for 2 DVD's where motherboard (Intel
D965-LT) has one IDE (2 hard drives) and 4 SATA unused at this time.
Startech tech support supplied a custom driver for me since I'm not
using the Raid at this time.

Peter Todd wrote:
{Quote hidden}

2008\02\24@234434 by peter green

flavicon
face

> Actually that's not true anymore. I don't know all the details of how it
> works, but RAID1 is bootable on Linux now, even your /boot partition can
> be RAID1. Since both drives are essentially a simple mirror the code
> is probably not too difficult. I've got two machines setup that way with
> Debian etch.
>  
It's possible to get a system that boots with either one of it's drives
unplugged certainly but what about the different types of failure that
can happen? will the bios correctly boot when a drive is still present
but is failing to identify correcly or will it hang up at some kind of
error? You can't really test those kinds of things so you just have to
hope the bios vendor did what would be sensible for a raid setup.

2008\02\25@023711 by Jake Anderson

flavicon
face
peter green wrote:
>> Actually that's not true anymore. I don't know all the details of how it
>> works, but RAID1 is bootable on Linux now, even your /boot partition can
>> be RAID1. Since both drives are essentially a simple mirror the code
>> is probably not too difficult. I've got two machines setup that way with
>> Debian etch.
>>  
>>    
> It's possible to get a system that boots with either one of it's drives
> unplugged certainly but what about the different types of failure that
> can happen? will the bios correctly boot when a drive is still present
> but is failing to identify correcly or will it hang up at some kind of
> error? You can't really test those kinds of things so you just have to
> hope the bios vendor did what would be sensible for a raid setup.
>
>  
With linux software raid i just set the bios to boot off one then the other.
Works fine. The other thing i did on a different system a while ago was
to put /boot onto a usb stick.
That was what i had just finished testing when the smoke came out.

2008\02\25@053645 by Ariel Rocholl

flavicon
face
I'm not sure either on the numbers he gives for Raid1 with embedded
RAID controller. I have experience with 20+ Dell and HP desktop
computers using RAID1 embedded controller with SATA disks (maxtor and
seagate) for about 3 years and, even though some HD have failed, I've
still to see the first problem with the RAID controller in any of the
desktops. Certainly they are simple ones, the software for controlling
and configuring it is just a subset of what a RAID offers for a server
setup, but you don't really need more than that for a desktop.
Coincidentally, a couple of months ago we had a critical failure in a
RAID controller in a high profile (and high cost) server that let us
naked on a RAID5 configuration.

{Quote hidden}

--
Ariel Rocholl
Madrid, Spain

2008\02\25@064413 by Carl Denk

flavicon
face
And it hasn't been mentioned that Windows can and does go in and change
the Bios settings to make it difficult to boot another system. On this
new Intel motherboard I have installed Kubuntu dual boot, went back to
XP Pro to check the E-mail, and went to boot from a CD finding the Bios
boot order changed to put the CD on the bottom of the list, or ignore it
with it on top. Or the Bios only allows on CD to boot, and not the 2nd CD.

peter green wrote:
{Quote hidden}

2008\02\25@070506 by Gerhard Fiedler

picon face
Jake Anderson wrote:

> peter green wrote:
>>> Actually that's not true anymore. I don't know all the details of how
>>> it works, but RAID1 is bootable on Linux now, even your /boot
>>> partition can be RAID1. Since both drives are essentially a simple
>>> mirror the code is probably not too difficult. I've got two machines
>>> setup that way with Debian etch.
>>>
>> It's possible to get a system that boots with either one of it's drives
>> unplugged certainly but what about the different types of failure that
>> can happen? will the bios correctly boot when a drive is still present
>> but is failing to identify correcly or will it hang up at some kind of
>> error? You can't really test those kinds of things so you just have to
>> hope the bios vendor did what would be sensible for a raid setup.
>>
> With linux software raid i just set the bios to boot off one then the
> other. Works fine. The other thing i did on a different system a while
> ago was to put /boot onto a usb stick. That was what i had just finished
> testing when the smoke came out.

FWIW, it is possible to have a software RAID1 system on Windows, too. Works
well, also on the boot drive. It is possible to boot from either of the
drives that make up the set. See
<http://www.techsoftpl.com/backup/index.php>.

Gerhard

2008\02\26@144157 by M. Adam Davis

face picon face
His main points are correct:

* RAID increases the complexity of the system
* Many people don't understand the handicaps RAID introduces, and only
focus on the benefits.
* While there are exceptions most consumer RAID setups produce more
complexity and associated lower reliability than is returned in
benefits. (though some of his cost/benefit ratios are based on
subjective measures)
* RAID 0 (striping for speed) is only benefitial when 1) HD throughput
is the bottleneck and 2) Access time is not an issue (ie, seeking
doesn't speed up, but throughput does)
* RAID 1 (Mirroring) is no replacement for a good backup
* RAID 5 and 10 are really not interesting as they either have the
same problems, or are simply not applicable to the customers.

Regarding specific bits of it:
His estimate of 20-30% of RAID tech support issues is in no way
comparable to the hard drive failure rate.  All he's really saying is,
"RAID requires more configuration than a hard drive, and we have to
support it more than hard drives"

It's not a complete article - it really should go through the benefits
in addition to the problems RAID introduces.  He doesn't point out,
for instance, that while RAID 1 isn't a replacement for backup, it
does keep the system up and running in the face of failure, which no
other backup system will do - only redundant systems/disks (ie, RAID)
will do that.

He suggests that it's a hassle to re-sync a RAID 1 drive, but the
reality is that it depends on the failure and what you plan on doing
about it.  It's certainly more of a hassle to rebuild your system from
scratch with a backup and a new hard drive, especially if it's just
the RAID card that failed - buy a new one and install it (you'd have
to buy a new HD in the other scenario anyway).

He also gives interesting statistics about hard drive failures and
RAID support issues, but he never mentions RAID card failures, and HD
failures in RAID arrays.  Sounds like either they don't track it (ie,
it's not frequent enough to track) or the numbers weren't conducive to
his argument.

Still, his points are valid - most people, even those that think they
need one, don't need a RAID array.

But I'll tell you what, my system was never faster than when I had two
10k Raptors striped for my main windows drive.

Hard drives are still significantly slower than the CPU and main
memory, and even in systems with 4GB of ram a significant amount of
paging occurs.

-Adam

On 2/24/08, Apptech <apptechspamKILLspamparadise.net.nz> wrote:
> He argues that RAID typically reduces reliability.
> ie done well it can be an improvement, but ... .
>
> Makes some sense.
> What do others think
>
>    www.pugetsystems.com/articles?&id=29
> -

2008\02\27@074317 by Gerhard Fiedler

picon face
M. Adam Davis wrote:

> * RAID 1 (Mirroring) is no replacement for a good backup

I read that often. It's true, but in most contexts it's a strawman. RAID1
is no replacement for a good backup strategy, and a good backup strategy is
no replacement for RAID1. They are in different categories, and complement
each other. So mentioning the obvious when nobody suggested that RAID1 was
a replacement for a good backup strategy is definitely what classifies as a
strawman.

> He also gives interesting statistics about hard drive failures and RAID
> support issues, but he never mentions RAID card failures, and HD
> failures in RAID arrays.  Sounds like either they don't track it (ie,
> it's not frequent enough to track) or the numbers weren't conducive to
> his argument.

I think that maybe the probabilities of RAID controller failures are low,
but if it happens and you're not in a pro shop that has dozens of systems
with the same controller and have a few replacement controllers on hand,
you're pretty much in a bad spot. I think it's a bad idea for a consumer or
small shop to be so dependent on a single piece of hardware for access to
the data. (That would of course be different if there was a standard how to
mark RAID arrays, so that I could take the disks and connect them to any
RAID controller. But that's just wishful thinking...)

> Still, his points are valid - most people, even those that think they
> need one, don't need a RAID array.

How do you (or anybody, for that matter) know what "most people" need? In
that same vein, I could say that you have a hard time figuring out what
/you/ need (just as me, FWIW :)

I think RAID1 is something that many people could need and could (maybe
even should) have. Even though it's more difficult to talk someone through
rebuilding a RAID1 array after a hardisk failure than talk him through
boiling a pot of tea, it's definitely less difficult than reinstalling the
system (not to the state it was in when it left the dealer, but to the
state it was before the crash) and bringing the data back. Even with a
backup.

The major difficulty in rebuilding a RAID1 array is replacing the disk --
which needs to be done anyway, RAID or not. Most people who can do that and
are able to restore a backup onto it are also capable of performing the few
steps of rebuilding a RAID1 array. In that sense, I think the article was
mainly crap, written exclusively from the perspective of a service provider
(as someone else already stated) and completely forgetting the user of the
system and the data and configurations on it.

Gerhard

More... (looser matching)
- Last day of these posts
- In 2008 , 2009 only
- Today
- New search...