Install on raid volume

[Andy]

Hi

I'm planning to set up a new workstation that I will use at home for
development (attending a computer science course), some games and to store
all my family's digital photos.
I have 2 options.

1) Install to a single raid-5 volume powered by an Adaptec raid controller.
2) Install Windows XP on a 200gb hardrive, and install all games and
programs on this partition, and store all data on a single raid5 volume.

How will performance be if I go with option 1? Or is it advisable to go with
option 2?

/A.

 

[Kerry Brown]

What is the reason for using RAID? This determines the answer. Normally
RAID is used for redundancy on a server that needs to be up 24/7. RAID 5
will cause a slight slowdown with writes being worse than reads. Here is a
link that has a more detailed analysis.

http://www.storagereview.com/guide2000/ref/hdd/perf/raid/concepts/perfReadWrite.html

If you are looking for performance then RAID (other than RAID 0 which isn't
really RAID) may not be the way to go. If you are looking for a easy backup
then RAID is definitely not the way to go.

 

[Andy]

What is the reason for using RAID? This determines the answer. Normally
RAID is used for redundancy on a server that needs to be up 24/7. RAID 5
will cause a slight slowdown with writes being worse than reads. Here is a
link that has a more detailed analysis.

The reason for using raid is to secure all our family photos, documents etc.
The idea is to use the raid volume only for datastorage, and all
applications will be installed to the root volume that is located on a
single 200gb disk. Windows will also be installed to this disk. In my mind
this should be a good solution. Don't you agree?

A/.

 

[Kerry Brown]

No I don't agree. RAID is not a substitute for backups. You will still need
to do backups. RAID will add cost, slow down drive access, and depending on
what type cut down on the amount of storage available. You would be much
better off to spend the money you were going to spend on the RAID setup on a
couple of external drives and develop a backup plan. You will probably have
some left over money that could be used to add more ram or get a faster CPU.

 

[HeyBub]

What is the reason for using RAID? This determines the answer.
Normally RAID is used for redundancy on a server that needs to be up 24/7.
RAID 5 will cause a slight slowdown with writes being worse than reads.
Here
is a link that has a more detailed analysis.

http://www.storagereview.com/guide2000/ref/hdd/perf/raid/concepts/perfReadWrite.html

If you are looking for performance then RAID (other than RAID 0 which
isn't really RAID) may not be the way to go. If you are looking for a easy
backup then RAID is definitely not the way to go.

RAID 1 can improve read performance dramatically because the closest head
(of the two drives) gets the call.

RAID 0 (and others) can improve file copying dramatically when the source
file is on a different drive than the target.

There is no easier backup than a RAID 1 -- the second drive is an exact
copy, in real time, of the primary.

In no case will RAID degrade performance over a single-drive system.

 

[Andy]

Hi
The raid will not substitute any backups, and this will not present any cost
for me. I've gotten the raid controller for free with 3 disks. The original
plan was to set up a windows 2003 server with raid for filestorage, but
since I don't need a fully fledge server install I'm considering to use
Windows Xp instead. And thats why I wonder the impact of a raid-5 on
performance on a XP installation used for development, gaming and
datastorage. All family photos are regulary backed up to DVDs ( I know, its
not optimal, but that is what I got)

/A.

 

[Kerry Brown]

If you have the hardware lying around then go for it. Personally if it's a
really fast controller and supports the drives in a standard configuration
without RAID that would be my choice. You'll get more storage space. Using
RAID 5 with three dives you only get the space of two of the drives.

 

[Kerry Brown]

RAID 1 can improve read performance dramatically because the closest
head (of the two drives) gets the call.

RAID 0 (and others) can improve file copying dramatically when the
source file is on a different drive than the target.

There is no easier backup than a RAID 1 -- the second drive is an
exact copy, in real time, of the primary.

In no case will RAID degrade performance over a single-drive system.

RAID other than RAID 0 always slows down reads or writes or both. With a
very fast controller with it's own CPU a RAID setup may be faster than an
onboard controller but it is slower over all than the same controller would
be without the overhead of RAID. Some types of RAID are faster at either
reads or writes but not both. A lot depends on the controller and if the
drives are designed for RAID. Striping in particular involves calculations
that have overhead causing slower writes. RAID is never a substitute for
doing backups. Your post said it all RAID 1 is a copy. If you don't know the
difference between a copy and a backup then you need to do some studying

 

[Ken Blake, MVP]

There is no easier backup than a RAID 1 -- the second drive is an
exact copy, in real time, of the primary.

Easiest, yes. It takes no effort and you get a second copy of everything you
write.

But despite it's being the easiest, it's also among the poorest, weakest
form of backup there is. Raid 1 is meant for redundancy--to let the mirror
drive seamlessly take over if the original fails. It's used in situations
where down-time.can't be tolerated. But invariably, companies who use RAID 1
*also* create real backups and store them externally.

Agin, Raid 1 is a reduncancy technique, not a backup technique. The problem
with thinking of RAID 1 a s adequate backup is that it leaves you
susceptible to simultaneous loss of the original and "backup" to many of the
most common dangers: severe power glitches, nearby lightning strikes, virus
attacks, even theft of the computer.

In my view, secure backup needs to be on removable media, and not kept in
the computer. For really secure backup (needed, for example, if the life of
your business depends on your data) you should have multiple generations of
backup, and at least one of those generations should be stored off-site.

 

[HeyBub]

Easiest, yes. It takes no effort and you get a second copy of
everything you write.

But despite it's being the easiest, it's also among the poorest,
weakest form of backup there is. Raid 1 is meant for redundancy--to
let the mirror drive seamlessly take over if the original fails. It's
used in situations where down-time.can't be tolerated. But
invariably, companies who use RAID 1 *also* create real backups and
store them externally.
Agin, Raid 1 is a reduncancy technique, not a backup technique. The
problem with thinking of RAID 1 a s adequate backup is that it leaves
you susceptible to simultaneous loss of the original and "backup" to
many of the most common dangers: severe power glitches, nearby
lightning strikes, virus attacks, even theft of the computer.

In my view, secure backup needs to be on removable media, and not
kept in the computer. For really secure backup (needed, for example,
if the life of your business depends on your data) you should have
multiple generations of backup, and at least one of those generations
should be stored off-site.

Your points are well taken. In our case, the mirrored drive is removable. We
swap it with another daily.

In the event of a catastrophe such as you listed (fire, theft, pizza without
anchovies, etc.), a competent technician is looking at at least a day to
restore a machine. Install the operating system, re-install all the
application programs, and restore the backups.

 

[HeyBub]

RAID other than RAID 0 always slows down reads or writes or both.
With a very fast controller with it's own CPU a RAID setup may be faster
than an onboard controller but it is slower over all than the same
controller
would be without the overhead of RAID. Some types of RAID are faster at
either reads or writes but not both. A lot depends on the controller and
if
the drives are designed for RAID. Striping in particular involves
calculations that have overhead causing slower writes. RAID is never a
substitute
for doing backups. Your post said it all RAID 1 is a copy. If you don't
know the difference between a copy and a backup then you need to do
some studying

You really should get out more. For example, the "calculations that have
overhead..." is off by three orders of magnitude. The calculations are
completed in microseconds whereas head movement is measured in milliseconds.
There is NO loss of throughput due to the additional calculations. It is
similar to taking a couple of seconds to decide to drive to the next town
versus three hours for the actual trip.

In a RAID 0 configuration, it is possible to copy a monsterous file with
virtually NO head movement (aside from one track to the adjacent one). On a
single drive, there would be a average access time of, oh, 8msec x 2 for
each record. For 10 million records, that's 160 seconds just twitching the
head back and forth.

Granted, not all RAID controllers are equally efficient. I still maintain
that a modern controller can improve throughput, or at least not reduce it.

I do know the difference between a copy and a backup. A copy IS a backup
whereas backups are NOT copies. In a worst-case situation, a copy is a
load-and-go technique. A backup requires hours (sometimes days) of fiddling
with things before you are back up and running.

 

[Leythos]

You really should get out more. For example, the "calculations that have
overhead..." is off by three orders of magnitude. The calculations are
completed in microseconds whereas head movement is measured in milliseconds.
There is NO loss of throughput due to the additional calculations. It is
similar to taking a couple of seconds to decide to drive to the next town
versus three hours for the actual trip.

In a RAID 0 configuration, it is possible to copy a monsterous file with
virtually NO head movement (aside from one track to the adjacent one). On a
single drive, there would be a average access time of, oh, 8msec x 2 for
each record. For 10 million records, that's 160 seconds just twitching the
head back and forth.

Granted, not all RAID controllers are equally efficient. I still maintain
that a modern controller can improve throughput, or at least not reduce it.

I do know the difference between a copy and a backup. A copy IS a backup
whereas backups are NOT copies. In a worst-case situation, a copy is a
load-and-go technique. A backup requires hours (sometimes days) of fiddling
with things before you are back up and running.

Maybe you guys should read this article:

http://archive.nlm.nih.gov/pubs/bangbuck/bangbuck.php

You might also read this one, the part where they copy a 1.5gb file to
both a RAID-0 and RAID-1 array.

http://www.pcw.co.uk/personal-computer-world/features/2160752/raid-
performance-explained

 

[Kerry Brown]

You really should get out more. For example, the "calculations that
have overhead..." is off by three orders of magnitude. The
calculations are completed in microseconds whereas head movement is
measured in milliseconds. There is NO loss of throughput due to the
additional calculations. It is similar to taking a couple of seconds
to decide to drive to the next town versus three hours for the actual
trip.
In a RAID 0 configuration, it is possible to copy a monsterous file
with virtually NO head movement (aside from one track to the adjacent
one). On a single drive, there would be a average access time of, oh,
8msec x 2 for each record. For 10 million records, that's 160 seconds
just twitching the head back and forth.

If you read my post you would see that I excluded RAID 0. With the right
hardware RAID 0 can offer some performance gains at the expense of more risk
of data loss. My post should have read: Striping with parity in particular
involves calculations that have overhead causing slower writes. I thought
you would understand the parity was implied because I had earlier excluded
RAID 0.

Granted, not all RAID controllers are equally efficient. I still
maintain that a modern controller can improve throughput, or at least
not reduce it.

The same hardware with the same drives hooked up would be faster at some
operations if RAID was not involved. With good hardware the question is moot
because as you say the differences would measured in microseconds. With
consumer class hardware like on-board "hardware" controllers I have seen
even RAID 0 setups actually be slower for everyday use than a non RAID setup
on the same hardware. Most of them don't support RAID 5 but those that do
are usually considerably slower than not using RAID 5. Same with RAID 1.
Consumer class controllers aren't aware of head position and consumer drives
don't support supplying that information anyway.

I do know the difference between a copy and a backup. A copy IS a
backup whereas backups are NOT copies. In a worst-case situation, a
copy is a load-and-go technique. A backup requires hours (sometimes
days) of fiddling with things before you are back up and running.

Yes copies are good for quickly restoring data. Backups are for disaster
recovery. I know you understand the difference. In your other post you
stated "There is no easier backup than a RAID 1". This may give people the
idea that if they use RAID 1 there is no need for any other backups. This is
very bad advice.

I stand by my assertion that the real reason for using RAID is for
redundancy not performance or backups. I would never use RAID 0 except in
very special circumstances. For me the performance gains aren't enough to
outweigh the risks. I've seen too many hard drives fail. I use RAID arrays
all the time in servers. I haven't seen a good case yet for RAID on a
workstation. Possibly RAID 0+1 for high end video editing but to be
effective that would require special controllers and drives far beyond what
a typical system would have.

 

[Ken Blake, MVP]

Your points are well taken. In our case, the mirrored drive is
removable. We swap it with another daily.

OK, then what you're doing is sort of a mixture of RAID and backup. But I
would be very wary of suggesting that RAID 1 by itself is adequate backup,
which is how I interpreted the message to which I responded.