Can you stripe together 2 striped sets?

[ryandinan]

Ok...let me see if I can make this more clear... :)
I have four (4) 240GB Maxtor drives. I wanted to create one strip set including all four drives, and be done with it.

However, the Si raid controller card didn't like having all four drives in one set (I guess it was a size issue, since this is close to 1TB). So, I had to break them up into 2 stripe sets (each having 2 drives) in order to get it to work.

Now, in Win2K, I went to device manager to see what I could do with these two striped sets. For simplicity, we'd like to have these 2 sets appear as a single volume - hence the orignial plan of just striping all 4 drives together in one set.
In disk manamgement, I had the option of spanning the two stripe sets together, as well as striping the two stripe sets together.

I didn't realize you could - or think anybody would want to - stripe together multiple stripe sets.

How would that actually work anyway? Would half the data get written to one 'stripe set' and the other half get written to the other set? - and since data is being written to a stripe set - would that data be split in half again across the two drives in the set?? This is making my nose bleed... ;)
If the above is indeed what occurs, is this essentially what would happen if I simply had all 4 drives in one stripe set? In other words, does the data get split up based on how many drives you have in the array? For example, 4 drives would mean that 1/4 the data is written to each drive - 6 drives would be 1/6, etc, etc.?

I understand that the way this is configured, I'd have two striped arrays in hardware (on the Si RAID controller card) - and if I were to select the option to stripe these two arrays together, I'd also be including a software-controlled array (in Win2k).

I chose the span option (as that made more sense at the time) - but I have no problems going back and changing this if there are benifits to striping these two striped sets together :)

Thanks for clearing this up for me!

-Ryan

[OS-Wiz]

First off, software RAID of any flavor isn't a good thing. Second, are you using a 4 or 2 port RAID controller? If its a 4 port controller, do a 10 array; else just do the 2, 2 HD arrays.

[pudad]

you want to do this w/ crappy ata drives? that is just asking for problems. :eek:

[ryandinan]

Originally posted by OS-Wiz
First off, software RAID of any flavor isn't a good thing. Second, are you using a 4 or 2 port RAID controller? If its a 4 port controller, do a 10 array; else just do the 2, 2 HD arrays.

It's a 2 port controller.
Besides, we need all the space, so a 10 array wouldn't fit the bill..
But anyway - was spanning these two striped sets the best way to go?
And I know that software RAID is not the best due to all the cpu cycles it takes up, but this machine is really nothing but a storage array.

pudad:

I always thought that ATA drives were the idea behind RAID (redundant array of inexpensive disks). Using SCSI drives for this application would be cost prohibitive for our needs. Besides - the quality/reliability of ATA drives are very good in comparison.
In fact, we just had a SCSI drive go down in our file server the other day...
All of our IDE RAIDs have never had a problem to date (knock on wood).

-Ryan

[pudad]

Originally posted by ryandinan


It's a 2 port controller.
Besides, we need all the space, so a 10 array wouldn't fit the bill..
But anyway - was spanning these two striped sets the best way to go?
And I know that software RAID is not the best due to all the cpu cycles it takes up, but this machine is really nothing but a storage array.

pudad:

I always thought that ATA drives were the idea behind RAID (redundant array of inexpensive disks). Using SCSI drives for this application would be cost prohibitive for our needs. Besides - the quality/reliability of ATA drives are very good in comparison.
In fact, we just had a SCSI drive go down in our file server the other day...
All of our IDE RAIDs have never had a problem to date (knock on wood).

-Ryan

Yeah I have a hard time believing that, what kind of drives was it? How old? And SCSI raid came before this ide raid ...

RAID == redundant array of independent disks

Raid1 is fine for ide, but raid0 is just too risky, ide drives are just too flaky. Adn this idea you have is ecven worse. Statistically a raid0 array doubles the chances of problems, what you are suggesting quadruples it!!!
:eek:

[pudad]

hmm, I am betting that scsi drive was much older than your ide raid arrays simply because ide raid is a much newer technology than scsi.

[ryandinan]

Originally posted by pudad


Yeah I have a hard time believing that, what kind of drives was it? How old? And SCSI raid came before this ide raid ...

RAID == redundant array of independent disks

Raid1 is fine for ide, but raid0 is just too risky, ide drives are just too flaky. Adn this idea you have is ecven worse. Statistically a raid0 array doubles the chances of problems, what you are suggesting quadruples it!!!
:eek:

The RAID acronym I used was the original definition - it has since been changed to "independant" ;).

I'm just trying to figure out if spanning the 2 stripe sets was better than striping together the 2 stripe sets. I have to do one of those options.

BTW - the SCSI drive in the array that failed was 3 years old. We still have IDE Media raid's working daily (as video capture scratch disks) that are older.

I think IDE drives have come a long way, and are no longer the flaky beasts they used to be..

-Ryan

[pudad]

Originally posted by ryandinan


The RAID acronym I used was the original definition - it has since been changed to "independant" ;).

I'm just trying to figure out if spanning the 2 stripe sets was better than striping together the 2 stripe sets. I have to do one of those options.

BTW - the SCSI drive in the array that failed was 3 years old. We still have IDE Media raid's working daily (as video capture scratch disks) that are older.

I think IDE drives have come a long way, and are no longer the flaky beasts they used to be..

-Ryan

hmm yeah ide is a lot better now. 3 years eh, what is the waranty? I wonder why it is "inexpensive"?

hmm, strping the stripe, do you really need that much throughput? I'd just go w/ 2 sripes...

[SkyDog]

Originally posted by ryandinan
I think IDE drives have come a long way, and are no longer the flaky beasts they used to be..

Actually, I remember looking at some MTBF ratings on some IDE hard drives and seem to remember the newer ones being lower than previous ones. And some of those newer MTBF ratings are specifying that the drives be used only X number of hours per day! I wouldn't count on newer IDE drives being any more durable than their predecessors, and none of them are anywhere close to the reliability of SCSI drives in general. If they were much more reliable, then why would all the major hard drive manufacturers be cutting most of their warranty lengths from 3 years to 1 year?

Originally posted by ryandinan
I always thought that ATA drives were the idea behind RAID (redundant array of inexpensive disks). Using SCSI drives for this application would be cost prohibitive for our needs. Besides - the quality/reliability of ATA drives are very good in comparison.
In fact, we just had a SCSI drive go down in our file server the other day...
All of our IDE RAIDs have never had a problem to date (knock on wood).


The RAID acronym has been used to refer to "inexpensive" disks, but that's not necessarily meant to refer to IDE/ATA. The idea is that smaller, slower disks can be used in place of larger, faster disks to achieve the same type of throughput and potentially better reliability. Those smaller, slower disks that comprise an array would be very inexpensive in comparison to the mammoth single drives they could be compared to.

As for your SCSI array failing while the ATA ones haven't... Let's just say you're bucking the odds. Seagate's good about posting their MTBF ratings on their web site, so let's take a quick look at a couple of randomly selected recent drives:

ST3120022A Barracuda ATA 7200 RPM 120 GB: 600,000 power-on hours
ST173404LW Cheetah SCSI 10000 RPM 73 GB: 1,200,000 power-on hours

So the average ATA drive is expected to die twice as fast as the average SCSI drive under normal usage conditions. And that's before taking into consideration that the documentation states that the SCSI drive's MTBF numbers were built on the assumption that the SCSI drives' "normal usage" consists of 720 operating hours per month. If you do the math, 720 hours works out to 30 days, meaning they're designed to run 24/7.

[pudad]

Originally posted by SkyDog


Actually, I remember looking at some MTBF ratings on some IDE hard drives and seem to remember the newer ones being lower than previous ones. And some of those newer MTBF ratings are specifying that the drives be used only X number of hours per day! I wouldn't count on newer IDE drives being any more durable than their predecessors, and none of them are anywhere close to the reliability of SCSI drives in general. If they were much more reliable, then why would all the major hard drive manufacturers be cutting most of their warranty lengths from 3 years to 1 year?



The RAID acronym has been used to refer to "inexpensive" disks, but that's not necessarily meant to refer to IDE/ATA. The idea is that smaller, slower disks can be used in place of larger, faster disks to achieve the same type of throughput and potentially better reliability. Those smaller, slower disks that comprise an array would be very inexpensive in comparison to the mammoth single drives they could be compared to.

As for your SCSI array failing while the ATA ones haven't... Let's just say you're bucking the odds. Seagate's good about posting their MTBF ratings on their web site, so let's take a quick look at a couple of randomly selected recent drives:

ST3120022A Barracuda ATA 7200 RPM 120 GB: 600,000 power-on hours
ST173404LW Cheetah SCSI 10000 RPM 73 GB: 1,200,000 power-on hours

So the average ATA drive is expected to die twice as fast as the average SCSI drive under normal usage conditions. And that's before taking into consideration that the documentation states that the SCSI drive's MTBF numbers were built on the assumption that the SCSI drives' "normal usage" consists of 720 operating hours per month. If you do the math, 720 hours works out to 30 days, meaning they're designed to run 24/7.

(read my mind)


Very nice skydog! :D

hmm, I thought initially raid was a thing mainly for stripping and data protection stuff because any drives back in the day were flaky as heck... :confused:

[ua549]

If having all the storage appear as a single volume is your goal, you can simply mount the second volume in an empty directory on the first volume. Not only that, the second volume can be accessed by its own drive letter or via the mount point on the first volume.

Use disk manager or commandline diskpart.exe to set it up.

[pudad]

Originally posted by ua549
If having all the storage appear as a single volume is your goal, you can simply mount the second volume in an empty directory on the first volume. Not only that, the second volume can be accessed by its own drive letter or via the mount point on the first volume.

Use disk manager or commandline diskpart.exe to set it up.

very very nice! :)

[ua549]

Just another little used feature of Win2k and later.

[pudad]

Originally posted by ua549
Just another little used feature of Win2k and later.

cool, maybe I will stop bashing it now ;)

[vertices]

Wouldn't another logical "cheap" method here be to just set up all 4 disks as independent disks on the hardware raid controller and then just RAID 5 them through Windows Server?

You would still get speed enhancements, but better yet, if one drive goes down you don't lose everything.

Thats what I would do in that situation.:)

[ua549]

Originally posted by vertices
Wouldn't another logical "cheap" method here be to just set up all 4 disks as independent disks on the hardware raid controller and then just RAID 5 them through Windows Server?

You would still get speed enhancements, but better yet, if one drive goes down you don't lose everything.

Thats what I would do in that situation.:) Your system would be quite slow. Software RAID is a very big burden on the CPU. Not only that, you must also have a basic volume from which to boot because dynamic volumes are not bootable.

[pudad]

Originally posted by ua549
Your system would be quite slow. Software RAID is a very big burden on the CPU. Not only that, you must also have a basic volume from which to boot because dynamic volumes are not bootable.

yeah this was broght up in another forum. This guy wanted a cheaper raid solution than going w/ a LSI megaraid 1600, some people suggested software raid, but there goes part of why we use scsi :(

And I'd think raid5 would especially put a load on the cpu rather than a raid0 too. Not everyone has a 4Ghz vert ;).


to the original poster: do you really need all that space for your server? If this is for some big company I think they would be using something slightly more manly that serial ata and ide. fibre channel scsi? I don't know much about fibre channel (from what I hear it is not for people running scsi on their linux pcs at home for the most part :p ), but all I know is that it is used for serious stuff (that is what they run here at work anyways I hear, yeah yeah I should know more, but I'm just a coop :p ).

[vertices]

Originally posted by ua549
Your system would be quite slow. Software RAID is a very big burden on the CPU. Not only that, you must also have a basic volume from which to boot because dynamic volumes are not bootable.

Yes dynamics are not bootable but it appears he is booting off of another drive anyway since he said this:

"In disk manamgement, I had the option of spanning the two stripe sets together, as well as striping the two stripe sets together. "

If he had the option to stripe them then he must not be booting off them correct? You can't stripe a drive you just booted off of.

I may be wrong, but it still seems to me that you wouldn't lose that much performance (as long you have a halfway decent CPU) by using the Software Raid 5 and it seems a much better option than 4 striped disks. If any one of those disks goes down there is going to be some data loss.:)

[pudad]

Originally posted by vertices




it seems a much better option than 4 striped disks. If any one of those disks goes down there is going to be some data loss.:)

yeah I had to deal with that crap a year ago, crew stripping I say (unless you are deallign w/ some very reliable stufff, but 4 striped disks is horrible... I still don't see why you'd use ata for a enterprise level server? Is this like a small company we are talking about here?

[ua549]

Originally posted by vertices

Yes dynamics are not bootable but it appears he is booting off of another drive anyway since he said this:

"In disk manamgement, I had the option of spanning the two stripe sets together, as well as striping the two stripe sets together. "

If he had the option to stripe them then he must not be booting off them correct? You can't stripe a drive you just booted off of.

I may be wrong, but it still seems to me that you wouldn't lose that much performance (as long you have a halfway decent CPU) by using the Software Raid 5 and it seems a much better option than 4 striped disks. If any one of those disks goes down there is going to be some data loss.:)

First, he does not need dynamic disks to use a hardware stripe set.
Hardware stripe sets are bootable. Having options to span or stripe his current volumes presented in disk manager does not mean that the drives are currently dynamic volumes. Had one of those options been selected disk manager would then ask about converting the basic volumes to dynamic volumes. He would see the appropriate warnings and error messages about not being able to convert the system volume.


Originally posted by vertices
Wouldn't another logical "cheap" method here be to just set up all 4 disks as independent disks on the hardware raid controller and then just RAID 5 them through Windows Server?

You under estimate the impact of software RAID on disk performance as well as CPU utilization. It is quite significant even with a 3 GHz proc.

He would need a basic volume (outside of the RAID 5 array) as a system volume.

You are right about the risk of data loss. A 2 drive stripe set doubles the risk of losing all data on all drives and a 4 drive stripe set quadruples the risk. Personally, I'm not that brave! I much prefer RAID 1 (mirror) and RAID 5 (stripe set with parity) combined into a RAID 5+1 arrangement along with tape backup every day.

[pudad]

hey ua549 do you think it would be too much a risk to raid a pair of Cheetah 15K drives in raid0? They are pretty reliable for that I would think, and I'd have plenty of backup on ide drives too (yeah I neither have a second Cheetah nor a raid card, this is just wishful thinking :) ).

[ua549]

I won't do it myself (and I have 7 Cheetahs) so I can't recommend that anyone else do it.

[vertices]

Originally posted by ua549
First, he does not need dynamic disks to use a hardware stripe set.
Hardware stripe sets are bootable. Having options to span or stripe his current volumes presented in disk manager does not mean that the drives are currently dynamic volumes. Had one of those options been selected disk manager would then ask about converting the basic volumes to dynamic volumes. He would see the appropriate warnings and error messages about not being able to convert the system volume.


I understand the difference between hardware and software RAID.:) I understand that you can boot off a hardware stripe set. But he was talking about being in Disk Managment in Windows and having the option to create a stripe set from his disks which implied to me that he was not booting off them.

I guess I just forgot that it will allow to do it and then give errors. For some reason I remember it not even giving you the option if you are on the system volume.:)

Originally posted by ua549
You under estimate the impact of software RAID on disk performance as well as CPU utilization. It is quite significant even with a 3 GHz proc.

He would need a basic volume (outside of the RAID 5 array) as a system volume.



OK...I guess I underestimated the impact.:) But, if the impact is so HUGE why does anyone use it then?

Yes I understand he would need a separate basic volume, which I thought he already had because I remembered incorrectly how it will allow you to go ahead and then give the error.:)

[pudad]

Originally posted by ua549
I won't do it myself (and I have 7 Cheetahs) so I can't recommend that anyone else do it.

just wondering, how many more can you add on? I got into a big debate w/ someone on thsi forum a while back about ultra3 being able to support up to 15 disks. I think older scsi systems were limited to 7 (yours probably is not one of them) but I was just wondering. I read a whole bunch of stuff about it, but there aren't that many people that run that many scsi disks so you are the person to ask :) . Basically from what I got out of various sources was that a good ~ 3 meter SCSI cable should be good for 15 drives...

edit: oh yeah I just remebered, my system manuals and stuff talk about 15 scsi ids on each channel :)

[pudad]

I wish he would answer to me why he want to use any kind of raid0 setup on a server that he'd be relying on, and why not raid 0+1 (hmm I have heard of raid 10, is that the same thing?).

[vertices]

Originally posted by pudad
I wish he would answer to me why he want to use any kind of raid0 setup on a server that he'd be relying on, and why not raid 0+1 (hmm I have heard of raid 10, is that the same thing?).

http://www.lsilogic.com/products/stor_prod/raid/backgrounder7.html

Apparently RAID 0+1 is more like Raid 5. I always thought 0+1 was a combo of 2 mirrored RAID 0 sets.

Check that link out and check the different RAID configs. I guess I was wrong about some of this stuff.

Edit~~~forget it..I can't read diagrams today apparently...thats what I get for posting and talking on the phone at the same time lol!!:):D

[pudad]

Originally posted by vertices


http://www.lsilogic.com/products/stor_prod/raid/backgrounder7.html

Apparently RAID 0+1 is more like Raid 5. I always thought 0+1 was a combo of 2 mirrored RAID 0 sets.



that is what it is isn't it? I swear it is, or at least that is how the highpoint controller on the KD7-RAID says (before smarktards try to bash me for having a KD7-RAID, know that is was a very nice rma from a KR7A-RAID :cool: ).

[ua549]

RAID 0+1 = mirrored stripe sets.

RAID 10 = striped mirror sets.

There is a difference, but IMO it is not too significant because with that many drives different RAID configurations are better.

[vertices]

Yeah it is....lol!! Apparently I'm having troulbe looking at diagrams today! :)

[pudad]

ah now I see, reverse sort of, hehe learn new stuff all the time.

hey ua549, about the # of disks on a ultra3 channel? 15?

[ua549]

15 addressable drives on any wide channel (many SCSI versions including Ultra 3).
7 addressable drives on any narrow channel (mostly ancient SCSI versions).

[pudad]

Originally posted by ua549
15 addressable drives on any wide channel (many SCSI versions including Ultra 3).
7 addressable drives on any narrow channel (mostly ancient SCSI versions).

sweet! :)

thank you