Migrating to RAID 0 without losing data?

  • Thread starter Thread starter Ed Jay
  • Start date Start date
E

Ed Jay

The disk performance numbers I'm reading about have convinced me I
want to experience RAID 0 with my P4P800 and two Maxtor 120 Gb SATA
drives. Already having tons of software and data I'm reticent to
'start from scratch' to improve performance; however, I'm presuming
I'll lose all data on the drives when I migrate to RAID. What is the
best strategy to employ in moving to RAID 0 and retaining all my data?

Also, am I correct in thinking that the end result of striping two 120
Gb drives yields a single 120 Gb psuedo drive?

Ed Jay (remove M to respond)
 
Striping, raid 0, yields 2x the drive(2x120=240), back up the data
somewhere else. If the data can't be backedup and software reinstalled
stay away from stripes, that performance comes at
a penalty, one glitch and you lose everything!
 
Ed said:
The disk performance numbers I'm reading about have convinced me I
want to experience RAID 0 with my P4P800 and two Maxtor 120 Gb SATA
drives. Already having tons of software and data I'm reticent to
'start from scratch' to improve performance; however, I'm presuming
I'll lose all data on the drives when I migrate to RAID. What is the
best strategy to employ in moving to RAID 0 and retaining all my data?

You must create a new partition (that spans two drives), and then format it,
which will destroy the data. I very much doubt that partition magic will
enable you to do non-destructive partition exansion across multiple drives
in this manner, but anything is possible these days... :-)

The best strategy to no loss of data is to back it up.

I think that you can usually do non destructive RAID when you are mirroring,
but not when striping as the partition has to change. I hope I'm wrong and
that somebody can tell you otherwise, I have not done RAID personally.
Also, am I correct in thinking that the end result of striping two 120
Gb drives yields a single 120 Gb psuedo drive?

Striping the drives would result in a single 240Gig partition with a higher
chance of data loss (since either drive dying would kill all of the data, or
render it useless anyway)

Mirroring would result in a single 120Gig partition (on the face of it) with
redundancy and thus much less chance of data loss (since both drives would
have to die simultaneously)

Ben
 
RAID5 takes care of that doesn't it? striped mirroring? requires three
drives, best of both worlds..

But in real world terms, any penalties in speed and performance?
 
RAID5 is a different kettle of fish all together... It spreads redundant
data across all drives in the RAID set. A minimum of 3 drives is needed with
the equivalent of 1 drive for recovery data. So if you have 3 drive of
100GB, you end up with 200GB disc space. If you have 4 drives, you end up
with 300 GB disc space. The more spindles in a RAID 5 array, the slower
writes become since to do a write to a sector, all other matching sectors
have to be read, the redundant data (its an XOR apparently) recalculated and
re-written along with the sector that needs to be written. So with 3 discs,
1 write = 3 reads + XOR calc + rewrite of recovery data + write of sector
you wanted. Slow is another word for it. Read speeds are good.

Few IDE RAID (or SATA) offer RAID5 - only the top end, and at the price of
these you may as well get SCSI RAID.

With the price of drives these days, I would always go for RAID 1 (mirror)
for resilience, RAID 0 for speed, or RAID 10 ( RAID 1 + RAID 0) for both.

- Tim
 
I used RAID0 for several years on a pair of PATA Barracuda's. True, it does
improve performance...but as BoB said, one hiccup kills the lot.

OTOH...mirroring means that any and every itsy bitsy error/hiccup/bad code
etc. that comes along is instantaneously written to both drives. So nearly
anything - except mechanical failure - corrupts 100% of your drive capacity
just as surely and efficiently as if you only had one "main" drive.

So. What I did, and what I advise others to do, is set up a striped
array...but have a THIRD drive in place (non-RAIDed) for imaging the array
[when it is pristine] and/or back-ups and/or archiving large downloaded
files.

Please advise of your decision/progress.
Ron
 
Ron said:
I used RAID0 for several years on a pair of PATA Barracuda's. True, it
does improve performance...but as BoB said, one hiccup kills the lot.

I said that too. :-P
OTOH...mirroring means that any and every itsy bitsy error/hiccup/bad code
etc. that comes along is instantaneously written to both drives. So
nearly anything - except mechanical failure - corrupts 100% of your drive
capacity just as surely and efficiently as if you only had one "main"
drive.

Yeah, but you have to assume the data being passed to the drive is correct.
Mirroring will protect you from drive failure, thats it's purpose.
So. What I did, and what I advise others to do, is set up a striped
array...but have a THIRD drive in place (non-RAIDed) for imaging the array
[when it is pristine] and/or back-ups and/or archiving large downloaded
files.

How do you know it's pristine unless you assume that all data written is
correct? Unless you assume that, you may as well not use a PC!

Ben
 
Oh sure, Ben! NOW you respond to me! ;-)

Yes. Quite right - you did say the same thing as BoB...and I neglected to
include you in my observance. In any case, as far as the "pristine" thing
goes...

A user presumably chooses to create a back-up at an "ideal" time. i.e. when
the system is working smoothly. Unless under extreme circumstances,
wouldn't that be the best time to take a snapshot? Esp. if the back-up
is/includes the 'System State'. So when I said that a user might image the
array "when it is pristine", I meant doing it when it appears to be working
well. This presupposes that it is *that* stage to which the user would
choose to restore the machine subsequent to a meltdown of some sort!

Unfortunately, as we all know, these darned machines can be tempermental.
And it is prudent to have some sort of fall-back in the [inevitable] event
of a failure of some kind. The trick is in finding a reasonable compromise
between smugness and paranoia. Sure, you can create a RAID0 for speed...if
you don't mind sacrificing safety. Or you can mirror and sacrifice speed.
Or, if you have FOUR similar HDD's, a 0+1 will deliver the best of both.

But do you stop there? Many say no. So they add a stand-alone drive for
back-ups (or put stuff on remvble disc). The point is, speed is the brass
ring in this adventure. At least for most of us. And as always, you have
to remember that s/w can be re-installed. It's only the *created stuff that
must be preserved. So the "risk" [of striping] outweighs the danger of mech
failure. For many people, anyway.

So Ed? Go for it! Bite the bullet and re-format/install your stuff on a
striped array...but pause early on and take a snapshot - j.i.c. - at a
point to which you'd like to restore in the event of a problem.

<he continued muttering to himself even as he wandered off...>
Ron
 
Ron said:
Oh sure, Ben! NOW you respond to me! ;-)

It's early, I don't understand... Did I miss something?
Yes. Quite right - you did say the same thing as BoB...and I neglected to
include you in my observance. In any case, as far as the "pristine" thing
goes...

A user presumably chooses to create a back-up at an "ideal" time. i.e.
when the system is working smoothly. Unless under extreme circumstances,
wouldn't that be the best time to take a snapshot? Esp. if the back-up
is/includes the 'System State'. So when I said that a user might image
the array "when it is pristine", I meant doing it when it appears to be
working well. This presupposes that it is *that* stage to which the user
would choose to restore the machine subsequent to a meltdown of some sort!

Well it depends what you want to back up... I don't really do them, but
then I don't really create much valuable content locally. I see little
point in backing up the entire system in case it goes down, 'cos if it does,
then thats a good reason to reinstalll windows. I've had a hard drive die
on me, so I know what it feels like.
Unfortunately, as we all know, these darned machines can be tempermental.
And it is prudent to have some sort of fall-back in the [inevitable] event
of a failure of some kind. The trick is in finding a reasonable
compromise between smugness and paranoia. Sure, you can create a RAID0
for speed...if you don't mind sacrificing safety. Or you can mirror and
sacrifice speed. Or, if you have FOUR similar HDD's, a 0+1 will deliver
the best of both.

Well yeah.. I decided to go with a single Raptor for speed.
But do you stop there? Many say no. So they add a stand-alone drive for
back-ups (or put stuff on remvble disc). The point is, speed is the brass
ring in this adventure. At least for most of us. And as always, you have
to remember that s/w can be re-installed. It's only the *created stuff
that must be preserved. So the "risk" [of striping] outweighs the danger
of mech failure. For many people, anyway.

And indeed it should... if you've ever had two hard drives die on you,
according to the manufacturers of these things, that would probably make you
the least lucky person in the world.
So Ed? Go for it! Bite the bullet and re-format/install your stuff on a
striped array...but pause early on and take a snapshot - j.i.c. - at a
point to which you'd like to restore in the event of a problem.

Agreed. If you want your data, don't risk losing it on something as drastic
as changing a partition.

Ben
 
Oh, I'm ribbing you a bit, Ben. There was a time - still is, actually -
when I really wanted your feedback on some issues. So I posted a message
with the prerequisite "A7N8X Deluxe" in the subject line...hoping that you'd
reply. Nothing. So I copied 'n' pasted into a fresh message with "Ben?
AJ?" in the subject line. Still nada. So I threw in the facecloth.

Regardless...clearly (and not surprisingly) we agree [on the originally
presented issues].

Warmly,
Ron S.
 
Ron said:
Oh, I'm ribbing you a bit, Ben. There was a time - still is, actually -
when I really wanted your feedback on some issues. So I posted a message
with the prerequisite "A7N8X Deluxe" in the subject line...hoping that
you'd reply. Nothing. So I copied 'n' pasted into a fresh message with
"Ben? AJ?" in the subject line. Still nada. So I threw in the facecloth.

I don;t know if it's my news server or OE, but after the first couple of
days of being signed up to a group it starts dropping messages. I wasn't
ignoring you, I just didn't have the message!

Try again... now, I'll see if I get the message this time.

Ben
 
BoB said:
OE defaults to erasing them after 5 days, tools/options/maintenance

Thats turned off. It's not that they're being deleted, they never get read
from the news server. (Well they never appear in the list)

Ben
 
You´re right.
But the *ultimate* RAID solution would be a RAID 50 (up to now only an
expensive controller from LSI-Logic offers this for UDMA-drives).
It´s even faster than RAID 10 but offers the same level of redundancy
without wasting that much HDD-capacity.
But you need a *minimum* of 6 (six!) drives...

Cya -

Joachim
 
RAID5 is a different kettle of fish all together... It spreads redundant
data across all drives in the RAID set. A minimum of 3 drives is needed with
the equivalent of 1 drive for recovery data. So if you have 3 drive of
100GB, you end up with 200GB disc space. If you have 4 drives, you end up
with 300 GB disc space.

Okay so far.
The more spindles in a RAID 5 array, the slower
writes become since to do a write to a sector, all other matching sectors
have to be read, the redundant data (its an XOR apparently) recalculated and
re-written along with the sector that needs to be written. So with 3 discs,

Now you're completely off-target, incorrect, or just plan crazy.
Yes, in RAID-5 you do pay a penalty for writes, so if you have a heavy
write component in the READ:WRITE ratio, then it'll slow things down, but
not to the extent presented here.

RAID controllers (most anyway) will only require 2 reads, XOR, and two
writes for every write, regardless of the number of spindles in the
RAIDset. All that must be read is the XOR block for the block being
written, and the current "data" on that block. The current data is
"subtracted" from the XOR, and then the new data is "added" to create the
new XOR, and then the new data and the XOR are written out.

Many RAID subsystems hide this penalty with cache, but that comes with it's
own set of problems (what you *think* may be written may not actually be on
disk).

In general, if I/O is multi-streamed enough in the application(s), RAID-1+0
is the fastest and most resilient.
With the price of drives these days, I would always go for RAID 1 (mirror)
for resilience, RAID 0 for speed, or RAID 10 ( RAID 1 + RAID 0) for both.

Agreed, however....

The original message author (Ed) didn't say anything about his workload.
Frankly, for home desktop/side workstation systems, with mostly one single
really active application, I sincerely doubt that you'd see much speed-up
from a complex RAID configuration... and the added cost and complexity may
not be worth the effort.
 
Ok...thanks, Ben. I will need to re-run the benchmark test(s) and re-post
the results. I had wanted some help interpreting the numbers, as I have no
basis for comparison.

Will re-post with fresh info.

Regards,
Ron
 
Back
Top