Why Use RAID 5?

[John G.]

Going off another post I made that brought about more questions than answers for me personally...

Why would anyone want to use RAID 5, except in cases of more than, say, six disks? Seems to me that for anything less, it would just be (1) additional hassle, (2) introducing another point of possible failure and resulting data loss, (3) limiting expansion and transplant capabilities, and (4) sizeable reduction of total storage capacity.

The off chance that a drive would fail (and everything else in the array worked as planned) so that you didn't lose data doesn't seem worth all the other risks/downfalls. Everyone tells me that RAID isn't a backup solution and that I should keep backups on hand anyway, so what's the point? Am I missing something?

 

[MixManSC]

It depends on your intended usage and intentions. For me and my graphics business it is a lifesaver. It is not however a backup solution as much as it is a fault tolerance and high uptime solution. For me the advantage is that if our RIP workstation has a drive failure we simply replace the drive and keep working without even shutting the system down, rebooting or anything. I do believe in my opinion though that SCSI RAID with a high end controller gives the best results though. My main workstation has an IBM Servraid controller which is a 64bit 66mhz PCI raid controller with a battery backup and 128MB of cache. I use it to run 7 73GB 15K U320 SCA SCSI drives. 2 are configured as a mirrored boot volume while the other 5 are a RAID 5 array. The advantage is if I am in the middle of a 40 foot by 5 foot large format print and I have a drive failure I do not have to ruin hundreds of dollars worth of material and ink and print the job all over again. The system keeps running with almost no discernable loss of speed (another advantage of a high end raid controller) - I just pull the failed drive and replace it and the controller handles the rebuild automatically in the background. In my case though I go to what most would consider way excessive extremes for home use as this solution is very expensive. This system is actually an IBM server being used as a desktop workstation (runs XP Pro with Onyx Production House RIP to control large format printers and regularly processes 2GB+ images) and it has dual power supplies both on battery backups and each on separate power circuits, chipkill ecc memory, dual fault tolerant fiber gigabit network links, etc, etc. for fault tolerance and near 99.9% uptime.

I've used lower end raid solutions in the past such as promise IDE raid and while it did work well it was not near as robust a solution in that rebuild rates were very slow and the loss of space was not worth it for my home computer budget. I ultimatley found that simply running single drives was much less of a hassle for my home desktop. I just make sure to backup everything regularly as a single drive failure results in the loss of everything on that drive.

 

[NulloModo]

You get all of the benefits of RAID1, but with less overall loss of space. If you want 250gb of useable space with RAID1, you need to lose a 250gb drive for the mirror. If you want 250gb of useable space with RAID5, you lose much less on the parity disk.

 

[draksia]

MixManSC said it, the primary reason for raid5 is uptime. The added efficency compared to raid1 or raid10 is purely a bonus.

You have have high volume stuff that cost you big money when its down raid5 makes alot of since. Many external systems also have dual reduandant powersupplies and dual active-active controllers to make sure that any time of failure will not bring the system down. For many bussiness this extremely. It is also should be noted that raid was really born from a time when disks where rated for 100k hours of MTBF not the million plus we have now. Disk failure was so common that with roughly a 48 drive array you where bound to have a disk failure like once a week.

 

[defakto]

(1) additional hassle,

Yeah, it is, but there are situations where that hassles benifits far outweighs the costs.

(2) introducing another point of possible failure and resulting data loss,

For the most part it is true that anything you add to a system is another point of failure. However, most RAID cards you're goingt to buy, if you buy quality, are from companies that have been doing this for years. Not only are the cards designed to be rock stable but the drivers are tested, retested, stressed to their limits to make sure that you will have stability. 3ware wouldn't be in business if they made hardware that sucked.

(3) limiting expansion and transplant capabilities

Limiting expansion capabilities varies from card to card. Some cards will allow you to add drives while the system is up, some won't. Some will allow online capacity expansion, think adding a drive to an already created array to increase it's size. Some high end cards will actually allow you to span arrays across controllers. Transplan capabilities are really not that big either. You pull the raid card, put it in the new system, hook the drives up just like before and you should be ok. If you have 3TB of data though, are you really going to be moving it from system to system that often?

(4) sizeable reduction of total storage capacity.

Size reduction, so you give up a percentage of storage to decrease the chance you might lose data. Sure it's not a solution to back up BUT it's another added layer of capacity. With raid 5 you're only lose 1 drive worth of data in the array. If you have 3, the minimum drives needed in an array, each 250 gig, you'll have 500 gig available instead of 750gig 33% loss of space. If you have 12 drives, you'll have 2,750gig instead of 3000gig only an 8.33% loss of space.


RAID 5 does have it's place. It has increased sequential read speeds close to that of raid 0, minus the chance of losing everything if a single disk fails. Fault tolerance is always nice to have when data is concerned. It may not be what you are looking for but it definately has it's place in the scheme of things. Be it with 3 drives or 30 drives.

 

[DougLite]

RAID-5 is best used for applications where uptime and capacity are goals. RAID-5 allows a high capacity to be achieved (only one drive worth of capacity is lost to redundancy, regardless of how many drives are in the array) while still protecting uptime.

For home users, its most frequent application is vast media sotrage arrays (DVD rip libraries, etc) that would be extremely time consuming and tedious to restore from removable media, as you have to sit there and feed the system a DVD every few minutes to restore from a failure that is not protected by fault tolerance. If you lose a drive in RAID-5, you simply provide a replacement drive, then your RAID BIOS takes over and automates the restoration.

In the enterprise, RAID-5 is quite popular. It requires the second fewest spindles (3, only beat by RAID-1's 2) to provide fault tolerant uptime protection, while preserving rackspace and limiting power consumption and heat output. May not seem like much, until you've got a rack full of servers, then every drive bay you can save counts. Once again, manual restores suck for busy administrators, and automated ones are preferred.

 

[John G.]

Thanks for the input all-- I think I'm starting to understand. I suppose my main concern is that an array could actually put my data at further risk than it was at before, with the possibility of controller failure. I looked up the specs on the 3ware 7506-4LP (the card I would use should I have the money) and it cites a MTBF of 1.27 million hours. So it's just slightly more reliable than a hard drive? Sounds risky to me...

 

[defakto]

It's important to consider that 1.27 million hours is about 145 years, personally I don't believe anything manufactured yet will last that long, there is no way to prove it at all. It's all simulated tests. WIth that in mind, A raid controller has no moving parts, except maybe a cooling fan which easy to replace. Harddrives are mechanical parts. If you can find me a drive that could run for that many years without failure I would be far more than impressed. Anything on silicon tends to go bad only if improperly cared for. My uncle has a calculater that's close to 30 years old that still works, electronically, like new the buttons, mechanical, are the only things worn on it.

The controllers MTBF may be seem close to that of the harddrive but, in reality and practice that controller will last orders of magnitudes longer than a harddrive. I've got scsi-2 controllers to prove that theory.

 

[NulloModo]

The new Asus i975 board has 8-SATA ports and the capability for RAID5 built in. I've been thinking about it for my next build. Would it be smarter to buy a PCI RAID controller than to use the onboard SATA ports and controller? I mean, when it comes time to move along to a new system, I wouldn't be able to move the boards onboard controller, and other boards with that controller might not exist... is there anyway to unbuild a RAID array and leave all the data back to sitting on the individual drives?

The 8 SATA ports is the only awesome thing about that board, other than that, I'd go for an Intel D975XBX.

 

[GoodOmens]

The new Asus i975 board has 8-SATA ports and the capability for RAID5 built in. I've been thinking about it for my next build. Would it be smarter to buy a PCI RAID controller than to use the onboard SATA ports and controller? I mean, when it comes time to move along to a new system, I wouldn't be able to move the boards onboard controller, and other boards with that controller might not exist... is there anyway to unbuild a RAID array and leave all the data back to sitting on the individual drives?

The 8 SATA ports is the only awesome thing about that board, other than that, I'd go for an Intel D975XBX.

With that many drives your going to completly saturate your PCI bus. I would look into getting a PCI-E 4x card.

 

[GoodOmens]

Going off another post I made that brought about more questions than answers for me personally...

Why would anyone want to use RAID 5, except in cases of more than, say, six disks? Seems to me that for anything less, it would just be (1) additional hassle, (2) introducing another point of possible failure and resulting data loss, (3) limiting expansion and transplant capabilities, and (4) sizeable reduction of total storage capacity.

The off chance that a drive would fail (and everything else in the array worked as planned) so that you didn't lose data doesn't seem worth all the other risks/downfalls. Everyone tells me that RAID isn't a backup solution and that I should keep backups on hand anyway, so what's the point? Am I missing something?

I just bought a raid 5 capable card.

Its a RocketRaid 2220 and offers:

Online Raid Expansion (IE add hard drives to a exisiting array)
Capability of Moving to another raid.
It uses 64bit addressing so it does not have the SCSI 2tb limit.


Raid offers a extra redudency (Hell Raid stands for Redudent array of independant drives). Sure you can't run a raid w/o using a backup BUT what it does offer is far shorter down time.

In a properly set up raid your going to have a hot spare running on the controller card so if a drive does fail it will immediatly kick in and rebuild the array (with no human intervintion). This results in no down time for a loss drive. (just replacing the hot spare, but that can be at a later date).

If you were running single drive with nightly backup scheme, a loss of a drive would not be detrimental but it would result in longer down time (you have to replace the drive, run the backup ect.)

 

[Tilde]

I have had a controller failure and it sucked hardcore, causing the loss of more than 100 gigs of stuff. No money for backups . Luckily, it doesn't happen that often.