Are RAID arrays worth it?

[Splash]

Realistically if you're not prepared to either invest in more disk space or rethink your use of current disks you aren't going to be able to backup 650Gb, unless you fancy investing in something like an Ultrium drive or 2...

I would suggest that in your situation you have a think through what data you need to keep backed up and create a seperate partition (better still an entire disk) and run a backup to that once a day/week/month/whatever is most apt for your situation. Either that or buy a shedload of DVDRs...

[latrosicarius]

Does anybody think they'll rework the RAID standards any time in the future to allow them to be more plug & play? For instance, if you have 2 RAIDed drives, there should be some indication on the drives themselves that say "hey, i'm one part of a 2-disk RAID-0 array" so you can move them from system to system without setting them up fresh on a configuration screen and wiping all the data?

Or AT LEAST, is one particular "real" RAID manufacturer (3ware, etc) working on something like this so I can buy a couple?

[directhex]

Does anybody think they'll rework the RAID standards any time in the future to allow them to be more plug & play? For instance, if you have 2 RAIDed drives, there should be some indication on the drives themselves that say "hey, i'm one part of a 2-disk RAID-0 array" so you can move them from system to system without setting them up fresh on a configuration screen and wiping all the data?

Or AT LEAST, is one particular "real" RAID manufacturer (3ware, etc) working on something like this so I can buy a couple?

the problem is that there isn't any "standard"

vendors gain nothing from your proposal - in fact, it defeats vendor lock-in, allowing you to move to a competitor's product with ease (whereas if it;s hard, you might not bother).

this is the thing about software raid solutions; the scenario you describe - "hey, i'm one part of a 2-disk RAID-0 array" - works fine on linux software raid. you can move your disks from a 3ware controller to silicon image, you could clone a PATA disc to SATA (or run one part of an array from a SATA controller and one from a SCSI controller) and it's just work.

[alexkoon]

Been following this thread with interest as I am about to build a terrabyte NAS server to serve media around the house. Thinking of maybe 4 x 300GB drives initially on top of the OS drive (maybe flash or USB flash for this) Will be using Debian linux as I have fair experience setting it up (no RAID experience though) as a server etc.

Was thinking of either RAID 5 or JBOD (I think this means its just one large drive partition). If its JBOD, I can live with losing the data on 1 disk as long as the data on the other disks are ok as I can re-rip the media files as I think thats what JBOD gives you. Important data I was going to back up on an external drive periodically anyway. Would JBOD be a better bet this in this instance?

Also would a P3 800 Mhz (with 640 Mb memory) be good enough for linux software raid? Or would a 2.8Ghz P4 be better? (I have the option of using either for my file server). The other would be a simple web server/SSH server.

[latrosicarius]

Directhex, I disagree with that statement because if consumers want standards and interoperability, than wouldn't companies that respond to those wants be better off for attracting more customers?

For instance, if i'm sure people who are already locked into a particular model of controller card will upgrade their cards every 4 to 6 years anyway. So they are eventually going to be looking for a "new" model. Now, if one company is offering a model that allows interoperability and plug&play standards with multiple controller cards, THAT is the company they are going to buy.

My biggest problem with RAID controllers is this: say you have two different arrays and you need to take one out and put the other in for whatever reason. You can't do that on existing controllers... if one brand allowed people to do this, they would definitly get more customers. The reason that major servers don't do this kind of plug & play today is not because they *don't want to*, but because they have never been able to do so in the past and grew up learning other ways of managing their data. I wouldn't necessarially want to move my arrays from computer to computer each using different controller cards, but just to be able to plug & play different arrays on the SAME controller card without having to configure the arrays every time you plug in a different one, and therefore lose all the data.

Also, what you said about the Linux box sounds like a great idea for now... If I have a Linux RAID box as the "file server" will I be able to access all those files over the network with Windows PCs? I have absolutely no experience with Linux, but I do know that some PC files aren't readable on Apples.

[directhex]

Directhex, I disagree with that statement because if consumers want standards and interoperability, than wouldn't companies that respond to those wants be better off for attracting more customers?

ever heard of "microsoft"?

For instance, if i'm sure people who are already locked into a particular model of controller card will upgrade their cards every 4 to 6 years anyway. So they are eventually going to be looking for a "new" model. Now, if one company is offering a model that allows interoperability and plug&play standards with multiple controller cards, THAT is the company they are going to buy.

when it comes to RAID, then the home user accounts for about 0% of the market - and for corporate users, people will buy what their vendor supplies - usually 3ware or LSI

My biggest problem with RAID controllers is this: say you have two different arrays and you need to take one out and put the other in for whatever reason. You can't do that on existing controllers... if one brand allowed people to do this, they would definitly get more customers. The reason that major servers don't do this kind of plug & play today is not because they *don't want to*, but because they have never been able to do so in the past and grew up learning other ways of managing their data. I wouldn't necessarially want to move my arrays from computer to computer each using different controller cards, but just to be able to plug & play different arrays on the SAME controller card without having to configure the arrays every time you plug in a different one, and therefore lose all the data.

linux software raid is fine with this, if you configured it to be. mostly because linux is made by the sysadmins who want functionality - not the companies who want to force you to lock in

Also, what you said about the Linux box sounds like a great idea for now... If I have a Linux RAID box as the "file server" will I be able to access all those files over the network with Windows PCs? I have absolutely no experience with Linux, but I do know that some PC files aren't readable on Apples.

macos uses linux's SAMBA software - a daemon compatible with windows file sharing, for both serving and accessing. i access files on both SAMBA-on-linux and windows.

[latrosicarius]

ever heard of "microsoft"?

Lol I guess you're right. The commerical market has all kinds of hidden agendas.

Anyway, I love your Linux advice, I want to eventually get going a network attached storage system with at least 1-TB of slave space & another 1-TB of mirrored space, not including separate backups (which may very well be Blue-Ray or HD-DVD disks by the time I get it set up).

[GDVS]

If its JBOD, I can live with losing the data on 1 disk as long as the data on the other disks are ok as I can re-rip the media files as I think thats what JBOD gives you. Important data I was going to back up on an external drive periodically anyway. Would JBOD be a better bet this in this instance?

JBOD stands for "Just a Bunch Of Disks" and does exactly what it says on the tin. You lose a disk, that disk is gone but everything else carries on as normal.

[arthurleung]

I run the following configuration:
Rig1:
1x300 IDE (Boot) (Disk0)
4x300 SATA, 20G from each disk RAID0-ed (Disk1-4, Array1)
2x200 IDE, 20G from each disk RAID0-ed (Disk5-6, Array2)
Rig2:
1x160 IDE (Boot) (Disk7)
4x160 SATA, 20G from each disk RAID0-ed (Disk8-11, Array3)

I only use the windows soft-raid so I can move it around
Performance is fine (230MB/s), and I'm now bottlenecked by the NF4 LAN and CPU. Only disadvantage is its so damn slow to copy from Array1 to Disk 1-4, vice versa so I ended up needing to copy to Array2 FIRST then Array1 for maximum speed. That way I get 40MB/s averaged to the array instead of 15MB/s (semi-random read/write)

I tried the software SIL3114R RAID5 and its just horrible. After Init I did get like 90MB/s read/write BUT after some time it lags so badly down to 2MB/s r/w, and with the chip struggling I don't think that soft RAID5 is anywhere more reliable than my JBOD/RAID0

Will invest in proper RAID5 (Hopefully Broadcom) when the controller come down in price.

[latrosicarius]

I run the following configuration:
Rig1:
1x300 IDE (Boot) (Disk0)
4x300 SATA, 20G from each disk RAID0-ed (Disk1-4, Array1)
2x200 IDE, 20G from each disk RAID0-ed (Disk5-6, Array2)

Just a question... why would you have a 300GB drive for a master? If you put your files on a separate slave drive, you should never need a master with more than 50 or 60 gigs

[Smokey21]

In one word, No.

[arthurleung]

Just a question... why would you have a 300GB drive for a master? If you put your files on a separate slave drive, you should never need a master with more than 50 or 60 gigs

Last year I had only 1 rig so I only buy large drives so reduce the number of drives in my system (while having more space).

You are definitely right that I could do with a tiny master drive, even 10G is fine, but I doubt its going to be as fast as the 300G drive, and the 300G drive is not going to be 30 times more expensive than the 10G drive.

I have a 80G partition for OS (I admit its far too big but I don't want to change it now) and the remaining 201G on static storage.

There are a lot of things need to be think of before you buy your HDDs
1. Do you have many IDE/SATA ports left? (NF4 only have 4 SATA ports per board)

2. Power problem, can your PSU spin up 10 disks at the same time? (It would be almost 2A for each 300G Maxtor)

3. Do you actually have the space to fit that much HDDs? (i.e. 800G by 80GB drives takes 10 slots while by 400G drives it only take 2)

4. What kind of use you're going for? For single-user non-multitasking a single drive would be adequate, if you tax your I/O heavily but only a small scale then a couple of small (or faster) drives would do. If you're doing static nearline storage then you're better off with gigantic drives. My usage is somehow between the last two, and I thought that I could kill 2 birds with 1 stone.

5. Price. Consider a 80G drive cost about 35 quids and a 300G cost about 80 quids, if you want the space then obviously a 300G will be better. If you feel that you won't need 80G for a couple of years that the cheaper one will do the job fine. The 300G should run slightly (a couple of %) faster

[latrosicarius]

Wow, lol you have certainly thoroughly thought out your reasons for having that drive!

Personally, I try to get 10000RPM drives like Western Digital's "Raptor" for the master drives. Then I like to use the slower, 7200RPM, 300GB drives as file slaves.

The 36GB Raptor is fine for most people to use as masters, but I prefer the 74GB version b/c I have A LOT of programs and games installed, and I’m up to about 40 gigs now.

There's a new 150GB Raptor (still 10000RPM) but I passed up on it when upgrading one of my other systems in favor of the tried-&-true 74GB version because the extra size is absolutely useless (to me) because I typically slave all my files. Now, there are also new features such as NCQ and a 16MB cache on this new 150GB Raptor, but, as they would be the only net usability difference, I didn't feel the extra price was justified.

But I definitly see your POV with your drive setup. Cheers