How good or bad is RAID 10?

[Lindhea]

Hi!
I'm new here, so sorry if i'm posting in the wrong thread. Also excuse my english, I'm swedish.

 

Now to my question:

With 6 drives in RAID 10 one will get 3 drives effective storage and 3 redundant drives, so in a best case scenario up to 3 drives can fail without loosing any data. But in a worst case scenario, with two drives that are mirrored to each other being the ones that fail, will that RAID-setup fail at only two drives?

 

 

If so, that doesn't feel too well thought out one must say... Are there any good alternative solution that compare with RAID 10 regarding throughput, but does not have this possible failing state of only two drives? Or should I go with RAID 6 so that I get two totally redundant drives, even if I get less performance?

[sanderdegrote]

I would go with raid 5/6 because you will lose less drives and still have good performance, but you will need a dedicated RAID controller and they can be pricey.

[Lindhea]

Oh. So the onboard raid controller don't have sufficient bandwidth for this many drives or something like that?

[sanderdegrote]

Oh. So the onboard raid controller don't have sufficient bandwidth for this many drives or something like that?

RAID 5 and 6 require parity calculation, (almost) any onboard RAID controller just isn't fast enough to not be the bottleneck.

[Vindola]

Now what are the odds of 2 drives that are mirroring each other, failing at the same time?

 

 

The chances of this happening are practically none... unless for some reason you like having tantrums on your case every now and then with a sledgehammer.

 

No, Raid 10 is very reliable and brings the best of Raid 1 and 0. Under normal conditions, the only way such catastrophic failure can happen is if there was a power surge that toasts the whole PC. A simple solution for that is a power surge protector. Get one if you don't have already!

 

Also, Raid 10 does not need a Raid card; raid 5 and 6 do. Raid 5 and 6 are slower than raid 10, especially raid 6.

[N7_Zer0]

You'll definitely need a RAID card (that has onboard cache) for 5 and 6. Also a battery backup unit to enable write back cache to speed up the writes as they will be very slow without it.

[wpirobotbuilder]

With 6 drives in RAID 10 one will get 3 drives effective storage and 3 redundant drives, so in a best case scenario up to 3 drives can fail without loosing any data. But in a worst case scenario, with two drives that are mirrored to each other being the ones that fail, will that RAID-setup fail at only two drives?

 

 

Yes. Your worst-case redundancy is still one drive.

 

With that said...

 

If a drive fails in a RAID 10, the process of rebuilding the array will go very quickly. In that case, you are much less likely to have another drive fail during the rebuild than you are with RAID 5.

 

A RAID 10 can be much faster, since there is no parity calculation being done, and will most definitely be faster than RAID 5 with an onboard RAID controller. The obvious downside is efficiency of space usage (50% of total space is usable).

 

This might be of use to you for deciding between RAID levels.

[Lindhea]

RAID 5 and 6 require parity calculation, (almost) any onboard RAID controller just isn't fast enough to not be the bottleneck.

 

What about RAID Z2 - would that work reasonable good through the onboard controller?

 

 

Now what are the odds of 2 drives that are mirroring each other, failing at the same time?

 

 

The chances of this happening are practically none... unless for some reason you like having tantrums on your case every now and then with a sledgehammer.

 

No, Raid 10 is very reliable and brings the best of Raid 1 and 0. Under normal conditions, the only way such catastrophic failure can happen is if there was a power surge that toasts the whole PC. A simple solution for that is a power surge protector. Get one if you don't have already!

 

Also, Raid 10 does not need a Raid card; raid 5 and 6 do. Raid 5 and 6 are slower than raid 10, especially raid 6.

 

Well, if p is the probability of one drive failing, then the probability of one RAID 1 sub array failing would be p^2 (or 6p*p rather), correct me if I'm wrong. P sure is a small number for modern disks, and p^2 is a really small number. But it's nowhere near as small as p^3, which would be the probability of an array with 2 independently redundant disks failing completely (again, please correct me if I'm wrong. Math this late in the evening is not my thing).

I know that the probability p is small, and that p^2 is most likely sufficient for private use, but any risk is an unnecessary risk. And besides - RAID 6 or Z2 would give me more usable space...

[sanderdegrote]

What about RAID Z2 - would that work reasonable good through the onboard controller?

You'll need a ZFS file system for that. (Linux, Free-BSD) I believe RAID-Z is software based so it doesn't really care if you use the onboard controller.

[Lindhea]

You'll need a ZFS file system for that. (Linux, Free-BSD) I believe RAID-Z is software based so it doesn't really care if you use the onboard controller.

 

Yes it is, I just wasn't sure whether or not it was limited by the PCI lane throughput of the onboard controller.

 

 

Yes. Your worst-case redundancy is still one drive.

 

With that said...

 

If a drive fails in a RAID 10, the process of rebuilding the array will go very quickly. In that case, you are much less likely to have another drive fail during the rebuild than you are with RAID 5.

 

A RAID 10 can be much faster, since there is no parity calculation being done, and will most definitely be faster than RAID 5 with an onboard RAID controller. The obvious downside is efficiency of space usage (50% of total space is usable).

 

This might be of use to you for deciding between RAID levels.

 

That's a really good remark. RAID Z2 would however be even less likely to fail during rebuild. Or does RAID Z2 take significantly longer to rebuild a drive?

 

Thanks for the link!

[wpirobotbuilder]

Yes it is, I just wasn't sure whether or not it was limited by the PCI lane throughput of the onboard controller.

 

 

 

That's a really good remark. RAID Z2 would however be even less likely to fail during rebuild. Or does RAID Z2 take significantly longer to rebuild a drive?

 

Thanks for the link!

Z2 will be slower. The more parity drives you have on a volume, the longer it takes to rebuild a volume of a given size.

 

With Z2, if you had a drive fail during a rebuild, all would not be lost, but then you have to throw another drive in and hope no other drives fail.

 

There is always some sort of risk during a rebuild.

[Vindola]

What about RAID Z2 - would that work reasonable good through the onboard controller?

 

 

 

Well, if p is the probability of one drive failing, then the probability of one RAID 1 sub array failing would be p^2 (or 6p*p rather), correct me if I'm wrong. P sure is a small number for modern disks, and p^2 is a really small number. But it's nowhere near as small as p^3, which would be the probability of an array with 2 independently redundant disks failing completely (again, please correct me if I'm wrong. Math this late in the evening is not my thing).

I know that the probability p is small, and that p^2 is most likely sufficient for private use, but any risk is an unnecessary risk. And besides - RAID 6 or Z2 would give me more usable space...

 

Unless you are building a server or something meant for storage. It is very foolish to go with Raid 6. Raid 6 is very slow. At least go with Raid 5 goddamn it.

 

And no, your probability of failure formulas you got are wrong. Each type of raid got it's own rate of failure formulas; there's long but feel free to look them up.

 

In a private house setup the only advantage of raid 6 over 10 is more storage. That's it, since the chances of 2 drives failing at the same time (given you gotta surge protector) is basically impossible. So if you really want to avoid raid 10 go with raid 5, because no one in their right mind uses raid 6 for general home use. 

[wpirobotbuilder]

What about RAID Z2 - would that work reasonable good through the onboard controller?

Yes. Here is an LTT user's storage setup using the onboard controller for Z2.

 

http://linustechtips.com/main/topic/32904-to-anyone-that-has-fears-of-zfs-performance-issues/#entry996378

[IdeaStormer]

@Lindhea

 

It all depends on what your goals, needs and constraints (price, speed, equipment) are, otherwise you're just brainstorming RAID. I don't see how the discussion can really go far without a end goal you want to accomplish within your constraints/needs are.

[Lindhea]

Unless you are building a server or something meant for storage. It is very foolish to go with Raid 6. Raid 6 is very slow. At least go with Raid 5 goddamn it.

 

And no, your probability of failure formulas you got are wrong. Each type of raid got it's own rate of failure formulas; there's long but feel free to look them up.

 

In a private house setup the only advantage of raid 6 over 10 is more storage. That's it, since the chances of 2 drives failing at the same time (given you gotta surge protector) is basically impossible. So if you really want to avoid raid 10 go with raid 5, because no one in their right mind uses raid 6 for general home use. 

 

I am building a server meant for storage... And why would I want to take unnecessary risks with my data just because I am a private person not a company?

 

 

Yes. Here is an LTT user's storage setup using the onboard controller for Z2.

 

http://linustechtips.com/main/topic/32904-to-anyone-that-has-fears-of-zfs-performance-issues/#entry996378

 

Very nice! Thank you.

 

 

@Lindhea

 

It all depends on what your goals, needs and constraints (price, speed, equipment) are, otherwise you're just brainstorming RAID. I don't see how the discussion can really go far without a end goal you want to accomplish within your constraints/needs are.

 

Yes, I suppose I maybe should have been a bit more specific in my original post. Open discussions like this tend to go all over the place... But then again I was mostly brainstorming RAID, so...

[Eric1024]

Z2 will be slower. The more parity drives you have on a volume, the longer it takes to rebuild a volume of a given size.

 

With Z2, if you had a drive fail during a rebuild, all would not be lost, but then you have to throw another drive in and hope no other drives fail.

 

There is always some sort of risk during a rebuild.

when ZFS rebuilds, it only rebuild the data you have on the array, not the entire capacity of the array like most hardware controllers, so in that respect it's usually faster.

 

to OP: RaidZ2 with a decent amount of ram (say 8GB) and a L2 Cache (as in a 120GB ssd) will be wicked fast. ZFS is cheaper than other raid solutions because you don't need a hardware controller, it's arguable safer than hardware raid (It's not vulnerable to corruption on power loss (the raid 5 write hole)), and it has build in (optional) deduplication (for the whole array or only certain folders) and compression (also for the whole array or certain folders), as well as snapshots (again for the whole array or certain folders) and you can force the filesystem to check to make sure all the parity matches (to check for silent corruption) at any given time.

 

Edit: I think I'm gonna make a dedicated thread on ZFS, so get back to me in a few hours.

[wpirobotbuilder]

when ZFS rebuilds, it only rebuild the data you have on the array, not the entire capacity of the array like most hardware controllers, so in that respect it's usually faster.

But for a volume with a certain amount of data on it, rebuilding it will be slower if it is a Z2 volume vs a Z1 volume, correct?

 

I think I'm gonna make a dedicated thread on ZFS, so get back to me in a few hours.

Yes!

[Eric1024]

But for a volume with a certain amount of data on it, rebuilding it will be slower if it is a Z2 volume vs a Z1 volume, correct?

 

Yes!

hmmmm... I'm actually not sure on that. If you're only rebuilding for one disk loss, I think they would take about the same time. Reason: for both rebuilds, stripes and parity have to be checked and the missing stripe has to be recalculated, the only difference between Z1 and Z2 is that an extra parity block would have to be read from another drive, but since that's being done in parallel I don't think it should take any longer. There is a bit more calculation that has to happen when rebuilding in Z2 because I believe both parity blocks are checked, but CPUs are so much faster than disks that the rebuild is only bottlenecked by the disks, not processing speed. I don't have any data to confirm this though, so this is just my theoretical understanding.

[Vindola]

I am building a server meant for storage... And why would I want to take unnecessary risks with my data just because I am a private person not a company?

 

Well you should have opened up with that. I thought this was going to be a PC for standard use. If you are mainly gonna make it a storage server, then yeah sure, go raid 6. 

[Lindhea]

Sorry about that.

[bill]

If your worried about data loss during a rebuild make sure you've got a good backup solution in place so that even if the array fails you are still protected.  

 

My home server has a raid 5 array in it but I still run nightly backups to an iosafe hard drive next to my server and a robocopy command that duplicates my most important data over a VPN to a NAS at my mom's house.  Raid protects your from drive failure but not other events that will cause you lose your data.