RAID newbie again

[Jure Sah]

Hello all,

I have brought a RAID controller to merge many old drives into a big
one, only to learn that the controller is crap enough to crash while
autodetecting those aincient CHS Mode 0 drives...

So here I have a RAID 0,1,0+1,JBOD controller and a disk space crisis on
another computer (where the motherboard BIOS is too old for 8+ GB
drives, there's no way to update, and this RAID add-in card, with it's
stand-alone BIOS, is my only way to use larger drives), where I have an
old 8 GB drive, an even older 4 GB drive and an intention to buy a cheap
fast 40 GB disk (with 8MB cache) or two.

I am looking at diskspace and wondering. I could buy one 40 GB drive and
use it with the existing 8 GB drive in JBOD, but doing this I would miss
all the awesome advantages of RAID.

I am thinking what would happend with availabe disk space if I used a
new 40 GB drive with the 8 GB one on RAID 0 or RAID 0+1. I am also
wondering what would happen to the capacity if I got myself two 40 GB
drives and used them on RAID 0 or RAID 0+1 (when I tried a 500 MB drive
with a 1.6 GB drive on RAID 0, the result was a 1 GB drive so... I don't
udnerstand the logic behind this).

I could get two 40 GB drives and use JBOD, which would give me a
satisfying 80 GB, which is more than I need (a 40 GB system would be
fine, even tho I could use up to 60 GB), but this choice is unattractive
to me, because I would know I have supperior technology and not using it.

Yes I am paranoid about backups (I keep 50% of my files backed up at all
times, which I could do with or without RAID 1) and yes, I love
preformance when it comes cheap (90% of my system is third-party for
that reason).

What do you suggest?

Thanks in advance.

 

[Mr. Grinch]

I could get two 40 GB drives and use JBOD, which would give me a
satisfying 80 GB, which is more than I need (a 40 GB system would be
fine, even tho I could use up to 60 GB), but this choice is
unattractive to me, because I would know I have supperior technology
and not using it.

Yes I am paranoid about backups (I keep 50% of my files backed up at
all times, which I could do with or without RAID 1) and yes, I love
preformance when it comes cheap (90% of my system is third-party for
that reason).

IMHO, usually things are not clearly "superior" to something else. To
me, it's more often a matter of Pros and Cons for everything. Comparing
JBOD, Raid 0, Raid 1, each has it's pros and cons. Which best suits you
depends on your needs and applications. Write down the pros and cons for
each of your options, and go from there.

If your priority is simplicity, flexibility, ease of use and management,
then keeping the disks as individual volumes is the way to go. You can
still get performance benefits, by segregating your disk transactions so
that they are spread out across multiple disks. Operating System on one,
Swap / Page file another, Apps on another, Data on another, Logs on
another, etc. Basically get your reads / writes spread out across your
disks, so you don't have one sitting there thrashing while the others sit
idle. Depending on the applications in question, this kind of arrangment
can give visible performance benefits.

You can use Perfmon under NT/2000/XP/2003 to view your disk counters and
compare the usage stats of your drives. This can help if you're trying
to balance load across disks or track down slow / heavy disk performance
issues.

If your priority is performance, or maximum volume size, you might try
RAID 0. It tries to spread out transactions across two disks, but it
doesn't always do a good job. Depending on your application and your
RAID config, it might help, it might not, and it will vary from app to
app. Do you run that app often enough for it to make a big difference?
Is it worth the cons of RAID 0, which are less flexibility in drive
managment, and higher risk of loss? Do you need a volume bigger than
the biggest single drive you can buy? If the pros outweigh the cons, then
you might go with RAID 0.

If your priority is safety, then RAID 1 (or 5) might be considered. For
business server use, I consider hardware drive mirroring a minimum. But
it's worthless unless you've tested it extensively and know that it
works. That means pulling drives, replacing them, and going through the
rebuild / mirroring process, and making sure you can actually recover in
the event of a failure. A lot of people set up a RAID 5 or RAID 1
system but never test it, only to find out later after a drive fails
that they can't rebuild the array, or it won't boot for whatever reason.
So test it, and test it often. Some RAID 1 controllers can improve
performance on reads, by reading from both drives at once.

Some RAID controllers also have cache. It might be backed up by
batteries, it might not. If it doesn't have battery backup for the
cache, there's a greater chance of loosing data during a loss of power.
Some systems run with all write caches turned off for this reason,
because data integrity is more important than speed sometimes. Some
systems even disable the drive on-board cache for this reason.

Good Luck

 

[Z Man]

Hello all,

I have brought a RAID controller to merge many old drives into a big
one, only to learn that the controller is crap enough to crash while
autodetecting those aincient CHS Mode 0 drives...

<big snip>

Buy two 40GB hard drives and run RAID1. You get the least bang for the buck,
with a 2:1 ratio, but you get the maximum safety. Enough said?

(Dump the old drives, they perform poorly and are worthless. Try to get over
your preoccupation with older drives, they are not worth it.)

 

[Jure Sah]

IMHO, usually things are not clearly "superior" to something else. To
me, it's more often a matter of Pros and Cons for everything. Comparing
JBOD, Raid 0, Raid 1, each has it's pros and cons. Which best suits you
depends on your needs and applications. Write down the pros and cons for
each of your options, and go from there.

Exactly. One of the main reasons I ask questions here is to get answers.

If your priority is simplicity, flexibility, ease of use and management,
then keeping the disks as individual volumes is the way to go.

I'm afraid my RAID card doesn't support that and pluging the disks into
the motherboard controller is not an option since it can't handle large
drives (and updates are not posible because it is a second-hand computer
and the original documentation was not provided).

Also in this case, would you expect me to throw away the RAID controller?

You can
still get performance benefits, by segregating your disk transactions so
that they are spread out across multiple disks. Operating System on one,
Swap / Page file another, Apps on another, Data on another, Logs on
another, etc. Basically get your reads / writes spread out across your
disks, so you don't have one sitting there thrashing while the others sit
idle. Depending on the applications in question, this kind of arrangment
can give visible performance benefits.

That is what I'm using right now, but as said, not an option with a 40
GB drive in use.

You can use Perfmon under NT/2000/XP/2003 to view your disk counters and
compare the usage stats of your drives. This can help if you're trying
to balance load across disks or track down slow / heavy disk performance
issues.

I use Windows 98 FYI.

If your priority is performance, or maximum volume size, you might try
RAID 0. It tries to spread out transactions across two disks, but it
doesn't always do a good job. Depending on your application and your
RAID config, it might help, it might not, and it will vary from app to
app. Do you run that app often enough for it to make a big difference?
Is it worth the cons of RAID 0, which are less flexibility in drive
managment, and higher risk of loss? Do you need a volume bigger than
the biggest single drive you can buy? If the pros outweigh the cons, then
you might go with RAID 0.

You have still not told me what happens to the available capacity (disk
space) if I use a 8 GB and a 40 GB disk on RAID 0. Or what happens to
the available capacity if I use two identical 40 GB disks on RAID 0.

I am aware of the advantages and dissadvantages of RAID 0 and it seems
like a good idea to me, since preformance is sometimes improoved. I
simply do not understand what happens to the disk space.

If your priority is safety, then RAID 1 (or 5) might be considered. For
business server use, I consider hardware drive mirroring a minimum. But
it's worthless unless you've tested it extensively and know that it
works. That means pulling drives, replacing them, and going through the
rebuild / mirroring process, and making sure you can actually recover in
the event of a failure. A lot of people set up a RAID 5 or RAID 1
system but never test it, only to find out later after a drive fails
that they can't rebuild the array, or it won't boot for whatever reason.
So test it, and test it often. Some RAID 1 controllers can improve
performance on reads, by reading from both drives at once.

My RAID controler is cheap and I do not intend to use RAID 1, since it
would basicaly mean throwing away one harddrive worth of disk space. I
was asking about RAID 0+1. Here again same questions as with RAID 0.

Data safety is not as crutial, since I can still back up to CD drives
and possibly my tape drive, not to mention this is a network-backed
computer.

Some RAID controllers also have cache. It might be backed up by
batteries, it might not. If it doesn't have battery backup for the
cache, there's a greater chance of loosing data during a loss of power.
Some systems run with all write caches turned off for this reason,
because data integrity is more important than speed sometimes. Some
systems even disable the drive on-board cache for this reason.

My RAID controller has a small SRAM cache. Data integrity is ensured
with a CRC check.

 

[Jure Sah]

Buy two 40GB hard drives and run RAID1. You get the least bang for the buck,
with a 2:1 ratio, but you get the maximum safety. Enough said?

Not really... I wasn't going to do RAID1 in the first place. I'm only
interested in RAID0 or RAID0+1. And I still have not heard on what
happens to the disk space available then.

(Dump the old drives, they perform poorly and are worthless. Try to get over
your preoccupation with older drives, they are not worth it.)

(and you dump your preocupation with Microsoftomania... the old drives
are the only thing that will work with my zero-cost web/FTP servers)

 

[Mr. Grinch]

I'm afraid my RAID card doesn't support that and pluging the disks
into the motherboard controller is not an option since it can't handle
large drives (and updates are not posible because it is a second-hand
computer and the original documentation was not provided).

Also in this case, would you expect me to throw away the RAID
controller?

You haven't said what applications you use or what your priorities are.
So how could I even guess what answer to give you? If it were me, I'd
use the RAID controller, but as standard disk controller, one volume per
disk configuration, no RAID sets.

That is what I'm using right now, but as said, not an option with a 40
GB drive in use.

You mentioned two drives, an 8 and a 40. That gives you the ability to
split disk transactions between 2 physical drives. Also, if one drive
remains on the motherboard controler, and the other on the RAID
controller, it lets you split transactions between two controllers. That
won't make as much of a difference here though, because the disks
themselves are the bottleneck.

You have still not told me what happens to the available capacity
(disk space) if I use a 8 GB and a 40 GB disk on RAID 0. Or what
happens to the available capacity if I use two identical 40 GB disks
on RAID 0.

All volumes in a RAID set have to be the same size. That size is limited
multiples of the smallest disk. You can put bigger disks into the array,
but the extra space on them is wasted. Any disks that are bigger will be
only get used up to that smaller size, the rest is not used.

If you use an 8 and a 40, you can only make a RAID 0 of 8+8 = 16GB. If
you use 40 + 40, you can make a RAID 0 of 40+40 = 80.

I am aware of the advantages and dissadvantages of RAID 0 and it seems
like a good idea to me, since preformance is sometimes improoved. I
simply do not understand what happens to the disk space.

The volumes have to be the same size. That size is limited to the
smallest disk. That's one of the disadvantages of a RAID set.

http://support.microsoft.com/default.aspx?scid=kb;EN-US;303184

http://graphics.adaptec.com/pdfs/ACSP_RAID_Ch4.pdf

 

[Shailesh]

Not really... I wasn't going to do RAID1 in the first place. I'm only

interested in RAID0 or RAID0+1. And I still have not heard on what
happens to the disk space available then.

http://www.somacon.com/fasttrak_sx4/page4.php

 

[Jure Sah]

You haven't said what applications you use or what your priorities are.
So how could I even guess what answer to give you?

Answers to the questions would be one. But nevermind.

The computer with the RAID array installed is (will be) a standard home
PC, boosted for preformance (according to the benchmarks this is also
what my RAID controller is best for), which is on 24/7 and has some
services installed, part of which use files 600+ MB (CD immages and
archives) and the other randomly accessing the disks to read/write data.

If I understand RAID correctly, a pair of disks with 8 MB cache in a
RAID-0 array should preform like a single disk with 16 MB cache, thus we
got an increase in preformance even with the programs that randomly
access the disks.

If it were me, I'd
use the RAID controller, but as standard disk controller, one volume per
disk configuration, no RAID sets.

The problem is the RAID controller will not provide access to the drives
unless they're in a RAID array.

You mentioned two drives, an 8 and a 40. That gives you the ability to
split disk transactions between 2 physical drives. Also, if one drive
remains on the motherboard controler, and the other on the RAID
controller, it lets you split transactions between two controllers. That
won't make as much of a difference here though, because the disks
themselves are the bottleneck.

Yes. If it worked that way that is.

All volumes in a RAID set have to be the same size. That size is limited
multiples of the smallest disk. You can put bigger disks into the array,
but the extra space on them is wasted. Any disks that are bigger will be
only get used up to that smaller size, the rest is not used.

If you use an 8 and a 40, you can only make a RAID 0 of 8+8 = 16GB. If
you use 40 + 40, you can make a RAID 0 of 40+40 = 80.

I see. That makes sense. Thank you.

The volumes have to be the same size. That size is limited to the
smallest disk. That's one of the disadvantages of a RAID set.

http://support.microsoft.com/default.aspx?scid=kb;EN-US;303184

http://graphics.adaptec.com/pdfs/ACSP_RAID_Ch4.pdf

Thanks.

I'm going for two new 40 GB drives on RAID-0.

You've helped a lot thanks.

 

[Mr. Grinch]

Answers to the questions would be one. But nevermind.

If you don't like the answers don't use them. Anyone can type "raid 0 raid
1 volumes" into google web and groups to get the same answers. When
someone asks for help, the last thing they should do is claim that they
know something when it's painfully clear to everyone that they don't know
much, nor could they be bothered to spend the few minutes it would have
taken to find out for themselves. Last I checked, this forum wasn't
comp.sys.spoonfeedme. But I could be wrong!

If I understand RAID correctly, a pair of disks with 8 MB cache in a
RAID-0 array should preform like a single disk with 16 MB cache, thus
we got an increase in preformance even with the programs that randomly
access the disks.

The onboard drive cache hit rate doesn't change. You've got twice as much
onboad drive cache. It's not unified in any way. You've also got twice as
many disk sectors to cache. There is no net improvement in caching by
going to RAID 0.

Performance for RAID 0 depends on the striping, which attempts to balance
transactions across two physical disks in order to keep both busy. The
goal is to avoid if possible the situation where you have one drive sitting
idle, while the other is busy processing transactions and has more waiting
in queue. This is what determines any peformance benefit from RAID 0. It's
the ability to get both drives processing concurrent transactions with
neither of them sitting idle while the other is queued up.

But getting to that point is mostly a best guess for any application, based
on the size of the transactions it uses most often. If the stripe block
sizes are not in tune with what the application wants to do, or if the
variety of transaction sizes vary so much that there isn't a clear optimum
block size, then the performance may actually end up being the same or
worse than on a single drive. Often the only way people gain any
improvement at all from RAID 0 performance is by re-testing the apps they
use over and over again, re-creating the RAID 0 array each time with a
different block size parameters each time. They continue this process
until they've graphed out the sweet-spot for block size vs performance for
the given app. It could be that size also agrees well with the other apps
they are running. It cold also be that they find their apps require
totally different block sizes for best performance, and the net benefit
ends up being a wash. Caching RAID controllers can widen the range of
block sizes that will tune well, but the size of the controller cash is a
limiting factor.

 

[Folkert Rienstra]

If you don't like the answers don't use them. Anyone can type "raid 0 raid
1 volumes" into google web and groups to get the same answers. When
someone asks for help, the last thing they should do is claim that they
know something when it's painfully clear to everyone that they don't know
much, nor could they be bothered to spend the few minutes it would have
taken to find out for themselves. Last I checked, this forum wasn't
comp.sys.spoonfeedme. But I could be wrong!

Can I use that please or have you copyrighted it?

The onboard drive cache hit rate doesn't change. You've got twice as much
onboard drive cache. It's not unified in any way.

You've also got twice as many disk sectors to cache.

That's besides the point. Cache doesn't double each time capacity doubles.

There is no net improvement in caching by going to RAID 0.

Yes there is.
You can't buy single drives of the same capacity that Raid0 offers with
double the amount of cache.

 

[Jure Sah]

If you don't like the answers don't use them. Anyone can type "raid 0 raid
1 volumes" into google web and groups to get the same answers. When
someone asks for help, the last thing they should do is claim that they
know something when it's painfully clear to everyone that they don't know
much, nor could they be bothered to spend the few minutes it would have
taken to find out for themselves. Last I checked, this forum wasn't
comp.sys.spoonfeedme. But I could be wrong!

Well I'm very sorry if anything that doesn't end with a question mark no
longer deserves an answer in your view. I put up some clear questions
about what happens with disk space and what you gave me back was some
totaly general stuff that I've had a 1000 chances to read online before.

So much about the painfully obvious.

Yes, you've helped, thank you. Now stop feeling so damn over-important.
Other people can think too, you know.

 

[Jure Sah]

That's besides the point. Cache doesn't double each time capacity doubles.


Yes there is.
You can't buy single drives of the same capacity that Raid0 offers with
double the amount of cache.

Can somebody please explain me where is my logic flawed?

Cache of 1 drive is 8 MB, which can hold 128 64KB strips of data. Now
since the data read/write is arranged amongst the two drives, both
drives can exchangivly hold (Entry1 on drive1 + Entry1 on drive2 +
Entry2 on drive1 + Entry2 on drive2 + ...) 256 64KB strips in cache,
which sums up to 16 MB of effective cache.

Any cache the RAID controller has is of course effectively zeroed-out
(since the transport between the drives and the controller is not the
bottleneck), however I don't think my RAID controler has any cache that
could remotely compare to 16 MB in the first place.