Is motherboard RAID safe?

[sebazvideo]

Hello, I recently purchased two Seagate Barracuda 3 TB drives that are the same model and I would like to try a RAID with them, which I've never done before. My motherboard is a Gigabyte GA-890FXA-UD5. What I would like to know is:

1) If I set these two drives as a RAID to have one faster logical drive, is it going to be as dependable as having each one separately? I mean, is there more risk of file corruption in a RAID than each drive separately?

2) Will the total capacity of these two drives in a RAID be twice the formatted capacity of each of them, or less?

3) There are different RAID options. I know one is mirrored in case one drive fails but that option doesn't offer any speed advantage, and I already have an external drive for backup. In the manual it says there's RAID 0, RAID 1, RAID 5, RAID 10, and JBOD. Which one of these is the best for speed and less chance of file corruption?

Thanks,

Sebastian

 

[hhhd1]

3) There are different RAID options. I know one is mirrored in case one drive fails but that option doesn't offer any speed advantage,

That is not true, RAID 1 (aka Mirroring) could offer read speed increase, depending on the RAID chip/software used.

Read in details:

http://en.wikipedia.org/wiki/Standard_RAID_levels

with 2 disks, you can do either RAID1 or RAID0 or JBOD, you can't do RAID5 (needs 3 disks) or RAID10 (needs 4 disks).

Personally, I'd do RAID1.

 

[corkyg]

Safe is a relative term. RAID1 is the safest. RAID0 is the striping of two drives into 1 for more size and speed. It also is higher risk. Suggest you read up on RAID here.

http://en.wikipedia.org/wiki/RAID

This thread gets repeated almost every week.

 

[vss1980]

Safe is a relative term. RAID1 is the safest. RAID0 is the striping of two drives into 1 for more size and speed. It also is higher risk. Suggest you read up on RAID here.

http://en.wikipedia.org/wiki/RAID

This thread gets repeated almost every week.

Indeed, but many people forgot one other aspect - what happens if your motherboard fails?? If the RAID set contains any data you absolutely need to get your hands on, you would need to find another Intel chipset board with RAID capability to be able to mount the RAID array - other manufacturer RAID controllers will not recognize the array.

 

[Nothinman]

Indeed, but many people forgot one other aspect - what happens if your motherboard fails?? If the RAID set contains any data you absolutely need to get your hands on, you would need to find another Intel chipset board with RAID capability to be able to mount the RAID array - other manufacturer RAID controllers will not recognize the array.

True, but at least with Intel the newer chipsets seem to support arrays from some of the older ones so you won't need the exact same version.

And the Linux dmraid driver supports a lot of those fakeRAID array formats so you could get the data back that way as long as it was built with one of the supported formats.

 

[tweakboy]

From what I know motherboard RAID is more stable and faster.

But only thing is if one drive goes bad the array is dead..

 

[Nothinman]

From what I know motherboard RAID is more stable and faster.

But only thing is if one drive goes bad the array is dead..

One drive dying and taking the whole array with it only applies to RAID0, all other levels of RAID have at least some amount of redundancy.

 

[tweakboy]

My bad sure yes your right.. nothinman.. roger that.

 

[vss1980]

True, but at least with Intel the newer chipsets seem to support arrays from some of the older ones so you won't need the exact same version.

And the Linux dmraid driver supports a lot of those fakeRAID array formats so you could get the data back that way as long as it was built with one of the supported formats.

There was a review which tested this theory a while back - can't remember which site had it. I believe the outcome was that you could usually mount a previous array on matching hardware vendor chipsets, usually with no problems if going from previous to next generation hardware. If I remember right they tested the principle on both ATI/AMD and Intel and found it worked in each case (although obviously the ATI/Intel stuff wouldn't recognize each other, although I believe they figured out that the ATI/AMD RAID was based on Promise Tech RAID..???).

The question I have is whether or not todays chipset RAID is still 'fake' RAID and relying on CPU processing or handles or the calculations itself...?
If so, I'd rather just use Windows Dynamic Disk (or Linux dmraid) and take the controller tech out of the question.

 

[Nothinman]

There was a review which tested this theory a while back - can't remember which site had it. I believe the outcome was that you could usually mount a previous array on matching hardware vendor chipsets, usually with no problems if going from previous to next generation hardware. If I remember right they tested the principle on both ATI/AMD and Intel and found it worked in each case (although obviously the ATI/Intel stuff wouldn't recognize each other, although I believe they figured out that the ATI/AMD RAID was based on Promise Tech RAID..???).

The question I have is whether or not todays chipset RAID is still 'fake' RAID and relying on CPU processing or handles or the calculations itself...?
If so, I'd rather just use Windows Dynamic Disk (or Linux dmraid) and take the controller tech out of the question.

Of course they're still software fakeRAID, no manufacturer is going to waste resources on dedicated hardware when the current performance is acceptable and current gen CPUs can keep up with barely appreciable load.

I'm all for software RAID, but would avoid Windows Dynamic Disks because it's a closed format and almost no 3rd party tools support them well.

 

[sebazvideo]

I was out all day so I couldn't reply, but I wanted to say thanks to everybody for your input.

 

[vss1980]

Of course they're still software fakeRAID, no manufacturer is going to waste resources on dedicated hardware when the current performance is acceptable and current gen CPUs can keep up with barely appreciable load.

I'm all for software RAID, but would avoid Windows Dynamic Disks because it's a closed format and almost no 3rd party tools support them well.

Maybe so, but they are portable...

 

[Nothinman]

Maybe so, but they are portable...

Moreso than RAID associated with a piece of hardware, but I still wouldn't trust them.

 

[BonzaiDuck]

From what I know motherboard RAID is more stable and faster.

But only thing is if one drive goes bad the array is dead..

I think you have this turned around -- just a bit. mobo-chipset RAID may be "just as stable." A dedicated controller card with its own (i.e., 256 MB) cache and processor will be faster.

The down side of dedicated PCI-E controllers: it complicates sleep and hibernation, or makes them difficult or impossible.

 

[Old Hippie]

Doesn't really matter.

If you don't have a back-up you're not "safe" anyway.

 

[BonzaiDuck]

Doesn't really matter.

If you don't have a back-up you're not "safe" anyway.

And ditto to that .. . too . . . Of course, you could always add a "spare" to an array, but I prefer hot-swap and server backups.

Thing is, this last couple years, I've tried to reduce the number of HDDs constantly running (as with a 4-drive RAID5), compromising "immediate redundancy" with frequent but periodic backup clones and other solutionss (like WHS).

 

[Emulex]

Can't do scsi-hot-add with sata drives unless its backed by a sas controller. (esxi fail)

if you buy two M5014 LSI 9260-8i you can replace one in failure quickly. Matter of fact many lsi's use the same RIS so you could bum perc h700/6i and go to town.

Battery Backed Cache is a huge thing since it is unwise to enable a drive cache unless it has 100% power uptime. Also preserves data if the machine should crash/lock up that hasn't been written . Those cheap samsung 830's have no super-capacitor. Rotating Drives not have any protection either.

 

[StorageGuru]

Short answer to this no.. software raid can be moved to different machine and still work.

 

[murphyc]

And the Linux dmraid driver supports a lot of those fakeRAID array formats so you could get the data back that way as long as it was built with one of the supported formats.

Pretty sure fakeraid is, or is being, deprecated in dmraid. The md driver, managed with mdadm, does support current IMSM metadata. And at least with RAID 1, the metadata is at the end of the disk, so you just have an end of disk offset for the raid container, meaning that if you were to yank that drive and put it anywhere else, it would behave just like a single disk as there is a 1:1 correlation of array LBA's to on-disk LBA's. The one small "problem" is that if you use GPT, the secondary GPT will not be at the end of the physical disk, due to the IMSM metadata offset. So the kernel might get pissy about the secondary GPT missing, even though the primary GPT contains the true (non-end-of-disk) location of the secondary GPT. It's an ignorable error in any case.

 

[Turbonium]

OP: if you do RAID-0 (striping), you will have twice the capacity, but more probability of losing data. Think of it this way: if you striped the data among 100 drives, you would have 100 times the capacity as 1 drive, but if any of those drives fail, all your data is lost; with 100 drives, the chances of at least 1 of those drives failing is fairly high (as compared to using just 1 drive; either your drive is a lemon or it's not).

Does this make sense? I was never good at stats, lol.

Anyway, I ran motherboard-level RAID-1 (mirroring) for years on an Intel D875PBZ (i875P) board (2 identical SATA drives) and never ran into issues, EXCEPT FOR THIS ONE. I still have no clue what caused it. Either a software-level RAID bug, or two drives failing at the same time, which is highly unlikely, or even something else which I haven't thought of.

I'd advise having an offsite backup at all times. Have your RAID-1 setup, but also have regular backups onto an external HDD or something.

I once read somewhere on the net (from what seemed a reputable source) that any bugs in RAID (software or hardware) will transfer errors to both the drives in the RAID-1 array, which makes sense. If something goes wrong, in any way, both drives will inherit the error. So really, the only thing you're protecting against (redundancy-wise) is mechanical failure, in terms of the HDDs. Anything else is totally still possible.

That is not true, RAID 1 (aka Mirroring) could offer read speed increase, depending on the RAID chip/software used.

Read in details:

http://en.wikipedia.org/wiki/Standard_RAID_levels

with 2 disks, you can do either RAID1 or RAID0 or JBOD, you can't do RAID5 (needs 3 disks) or RAID10 (needs 4 disks).

Personally, I'd do RAID1.

As far as I know, RAID-1 offers a marginal speed increase when reading, and a marginal speed decrease when writing. I assume this is because it has two drives to read the data from, so it can split the workload (or perhaps whichever drive finishes first, cancels the request on the other drive), but when writing, it has to write to both drives, and wait until both are finished until it can move on to the next task (the slowest drive becomes the speed limit).