PartitionMagic

I agree - RAID on the motherboard with no cache is a bad thing. In most
cases they implemented a cheap RAID chipset and without the cache it's
not much better than an OS RAID, but it does offer performance benefits.

Most people don't understand RAID, they think that RAID is always faster
than non-RAID, and that's just not always true. RAID 1 is almost as fast
as a single drive for writes, but, can be faster than a single drive for
reads since the data can be read from either drive. RAID 5 has poor
write times, but great read times....

In most cases, for home users, and for people doing PS type work, a
simple MIRROR will save their butts in case of a hardware failure, but
that's about the only benefit - most of these users set the mirror up as
one large mirror, then partition it, and still think it's two drives for
purposes of swap/temp space... They should have installed a single drive
as a drive for swap/temp space and then they would have seen a
difference.

 

RAID 1 is a mirror - two drives hold EXACTLY the same information. While
you have twice the chance for a drive to fail, you only have half the
chance for data loss due to failure of any drive. In the real world that
means that loosing a single drive does not impact the user.

RAID 0 is a stripe across two disks - this provides twice the chance for
mechanical failure and twice the chance to loose everything in the event
of a mechanical failure. Loosing either drive results in the complete
loss of everything.

RAID 5 is a stripe across 3+ disks with a parity stripe used to rebuild
the other drives in the event of a failure. In this situation you can
tipple your chance for a mechanical failure, but you benefit from being
able to replace the defective drive and rebuild the entire array without
the loss of data. During the time a drive is lost, you will notice a
significant performance hit.

Mechanical failure does not always mean data loss - only RAID 0 in the
example above suffers from data loss.

 

I always mix up the two, that is the designation of RAID 0 and 1.

Thanks for the lesson but I run RAID 0, 1, 5 across my network in both
S/W and H/W implementations using Mylex, IBM, Adaptec and Promise cards.

Rob

 

Rob - out of curiosity, what are you using RAID 0 for?

 

Really, what happens if you blow two drives in a raid 5 setup ? Dont say it
never happens, it did.

 

A mirror file is fine for storage but totally useless for a swap file. I
dont think that PS is a special example of RAID useage for storage - I think
somewhere back at the top someone was asking for a RAID that would improve
speed not storage.

The only setup that can improve swap file speed is a RAID 0 combo.

Silicon Graphics has the worlds benchmark on RAID performance. Why do you
think that is ?

 

My power work-station OS and apps drive, it's backed up after each
install and my profiles and all other data reside on a server which has
RAID 1 and 5 arrays.

Rob

 

I'm still confused about those RAID answers. Why the hell cant you use a
simple raid setup to have yet another HD attached to a computer and that HD
simply for backups? Eg, similar to having a Master on primary and a Master
on Secondary off the motherboard. Thus, if C screws up and as you have been
good people and done backups all the time, just attach raid drive to primary
or secondary port, boot off Drive Image 7 CD (just for example, not that I
am pushing the program) and restore. Buggered if I know why a suddenly
scrambled C drive that wasn't scrambled because of a hardware fault would
affect a raid mounted HD as suggested above.

 

You replace the drives in question and restore from last nights backup.

 

I once spoke to a very sad IT manager. He had a mirror setup for his boot
drive. When the main drive started to wig out and write garbage the RAID
controller obliged by writing garbage to the mirror. Thus killing both boot
disks.

The magical assumption that a hard drive just dies isnt valid. It can do
some very strange things in the read/write cycle and cause no end of bother
on the data bus. Having said that, some software, such as databases also
likes to write pre-and post- cache files which require a stable read-write
cycle to marry up data changes. You can join the dots on the rest.

 

You have a means of backing up 2 terrabytes ? The answer is to use a "cold"
standby in a spare drive bay. A drive that isnt running, spins up and takes
the place of the first dead one. The second failure, if it is more than 30
minutes after the first one, wont affect the raid stripe.

RAID is supposed to be N-1 ( for every 5 drives you use, you only get the
benefit of 4 drives storage in total.) but it is really N-2 or N-3

I agree with the earlier post .... the Raptor is a drive to keep an eye on.

 

The same thing that happens if you blow one drive in a single drive
system - you loose it all.

What's your point?

The likelyhood of blowing more than one drive at a time is slim, unless
you don't have a UPS, but you wouldn't be all that money into a R5 array
and then not buy a UPS would you?

In almost 10 years of configuring R5 arrays on servers, RAID 0+1, RAID
1, and RAID 1, I've only had one instance where I've ever lost more than
1 drive at a time. I should help you a little, I did loose everything on
a Mirror (RAID 1) array set once, but that was because I didn't break
the mirror and wiped the drive 0 boot partition.

 

Your very sad IT manager was ill informed - the RAID controller controls
what is WRITTEN to the drive, not the drive itself. Your statement that
the drive started wigging out and corrupted the other drive is bogus.
What happens is that the RAID controller goes bad and takes out both
drives. I have seen RAID controllers go bad and take out all drives
data, not the drives, just the data, but I've yet to see a bad drive
take out the other drives.

I think if you talk with IT managers around the country you'll find that
they would rather have a MIRRORED OS Boot drive than not have one. Think
about it, without a redundant drive you loose everything if the single
drive dies - not the data on the other drives/arrays, but it takes time
to rebuild the OS, reapply security settings, etc... With a mirror you
just swap out the bad drive and you are still up and running, not even
any down time in the case of SCSI or SATA based systems.

As for your database comments, I've managed and designed MS SQL
databases with sizes from 20gb to 750gb, with a current one expected to
hit 1.2tb later this year. Write caching on the controller is not a
problem for the drives, that why a good card will have a battery and
NVRAM, the drives don't know anything, they are dumb devices, it's the
controller that does all the work. Even the database doesn't know it's a
RAID array, it only knows that there is a bus to pull/feed data from/to.

 

As someone else posted further down, in a case where you (not in this
post) mentioned 1.2tb of data, you would ALWAYS have a hot spare that
auto-inserts itself in place of any suspected bad drive.

I think your questions have exceeded the scope of people using Photoshop
on workstations.

On a workstation, a single mirrored set of drives will provide increased
read performance and a redundant element in the case of hardware
failure.

A second drive set (or a third single drive) used for the SWAP, TEMP,
and scratch disk would be an ideal setup and provide the most
performance with Photoshop.

 

If you don't have the means of backing up 2 terabytes, then you shouldn't be
storing it. Unless of course your lovely system can withstand a fire or a flood,
or some other disaster. Or the data really isn't that important anyway. If it
is, then anyone not backing up such an amount of data is an idiot.

 

It's pretty damned obvious why a real time 100% of the time backup would
stuff up if you mirror C elsewhere and expect elsewhere to work continuously
every single time when C fails.

Rot. I have had that happen even to my own computer. Back when 20gig was a
medium sized drive, I had 2 and C died in 6 weeks. By then 40gigs were about
and I bought a 40gig for C drive and restored from backup. No problems. What
I do is have the other drive as a backup drive only where mirror images are
stored but as my data doesnt have to be correct to the nanosecond, I only do
mirrors 2 times a week (auto). I have compressed images on that drive. HDs
DO just die and saying they dont is just plain wrong.

Sure. That goes back a long way. I remember old 5.25 drives where you write
something to it and wait. The drive light had to go out then let 30 seconds
pass and it would come on again for a few seconds and go out again. THEN you
could remove the disk. Doing it before that ended in dead data. However,
what you said doesnt have any relevance to the last sentence of what I
posted above.

 

Even that isnt necessarily true. You would have to know whether the drive
had come into a situation that it shouldnt have, causing damage elsewhere.
Eg, throw a wrench in moving gears and strange thigns happen!

It's called "The FZZZST Principle". Sometimes a drive dies in ways most
people never ever get to personally see. However, it still doesnt really
explain that sentence I posed in that other letter.

Yeah that may well be true. I still keep thinking of the "virus that got in"
being mirrored though!

No, not necessarily. Even Powerquest sell Drive Image 7 in a version that
allows hourly incremental backup. This is, in fact, my personal preference.
Eg, 9AM tomorrow morning the incremental gets done and at 9:30AM a virus
gets in and wipes all the much needed data. What the company has lost is 30
minutes of work. Now depending on the company, that may not be a great loss.
Even in companies that WOULD consider that a great loss in some areas, there
are others where it wouldnt be.

No, with a proper mirrored array, your C goes down and the mirror takes
over. You fix C and put it back online and the other drive then becomes the
backup again, all done without interrupting work flow. Of course this DOES
depend on the nature of what caused C to die.

Admittedly more than I have done!

 

There is nothing in the world more bogus than a theory which flies in the
face of experience. Please learn some manners before you insult your own
self. You are far too stupid to be taken seriously by me.