Impossible RAID10 Configuration

[Caligulae]

Hello Community!

Well after getting the TrueNAS Core installed (and booting...!), I got excited and wanted to get going with my RAID config. I wanted to configure, what I thought made sense, 2 x 4Tb (WD Black) Stripped Disk (Raid 0) + 1 x 8Tb (WD Red plus) Mirrored (Raid 1) of the Stripped disks...

This config would allow me to have quick access to the Raid 0 while having the Raid 1 giving me redundancy without the use of a Backup.

Got really disappointed after researching Google and this forum that it might not be possible as Raid 10 requires 4 disks??? I thought that I could mirror a Stripped set of disks to one physical drive with the same capacity... Lots of posts with old version of FreeNAS but nothing really new with TrueNAS...

Would the new TrueNAS Core (I am running TrueNAS-12.0-U6.1) allow the setup I am looking for? If not, what you be the best configuration to achieve what I would like? Should I just buy another 2 x 4Tb or bite the bullet and buy another 3 x 8Tb and get more storage? If I buy more disks, how should I configure the other drives?

Thank you in advance for your help!

 

[jgreco]

Well, first off, ZFS doesn't support RAID0, RAID1, or RAID10, and we strongly discourage the use of these terms because it poorly communicates what you're trying to do, and may actually confuse people.

Terminology and Abbreviations Primer

We realize that new users have a lot to learn when they come to FreeNAS. There's a certain amount of confusion added to discussions when users pick random/approximate terms to describe things. I've spent a lot of time quietly trying to translate terms on the reader's side when reading posts...

www.truenas.com

Both TrueNAS and FreeNAS are based on ZFS, and ZFS pool design is very straightforward. There are no new magic configurations in TrueNAS that allow weird combinations of disks to be summed together in arbitrary ways; pretty much every ZFS pool design guide you find on the Internet has all the fundamentals correct. ZFS was designed by Sun Microsystems for large storage arrays containing dozens or hundreds of identical disks, and it wasn't really designed for hobbyist or SOHO use where someone might want to add a disk at a time.

No ZFS configuration provides "a backup", just redundancy. Backups are also kind of orthogonal to system design. You need to arrange for backups on a separate system if you want them -- think about what happens if your power supply suddenly starts letting line voltage through to your hard drives.

There are lots of good guides to help you decide between different pool design strategies. If you are looking to add small amounts of disk at a time, you are probably better off with mirror pairs. This allows you to add two disks at a time as an additional vdev to an existing pool, or to replace the existing disks in a vdev with larger disks to gain more capacity. This means you could have a vdev of your two 4TB HDD's, buy an additional 8TB HDD to supplement the one you have, and then have a second vdev of the two 8TB HDD's, giving you a pool with two vdevs totaling 12TB.

 

[Caligulae]

Thx jgreco, I googled ZFS pool design guide and got some good reading material.

Also, I will investigate more on a Backup Strategy for the Striped drives, outside of the main server...

Thx again!

 

[sretalla]

You can (sort of) do what you're talking about with replication tasks and a relatively short schedule of snapshots... not exactly a mirror, but only one snapshot behind current at any moment.

Your first pool should be a mirror though if you want to benefit from the primary storage being redundant, but you could replicate to a single-disk striped pool in line with your original description.

I'm not sure that what you're thinking of doing is recommended by anyone, but that's a way it could be made to happen.

 

[jgreco]

I'm not sure that what you're thinking of doing is recommended by anyone, but that's a way it could be made to happen.

The main problem with this is that having a "backup" pool on the same system as your data makes you vulnerable to hardware issues with the system, such as the spectacular (and admittedly slightly ridiculous) PSU failure I mentioned earlier.

If you rsync or replicate your pool to a second host with nonredundant disk, that qualifies as "backup".

 

[danb35]

I thought that I could mirror a Stripped set of disks to one physical drive with the same capacity

Well, you can, but it's an unsupported configuration with, IMO, no real reason to recommend it, and you'll need to do it all at the CLI. But if you manually partition the 8TB disk into two 4TB partitions, you can add one of the partitions as a mirror of the first 4TB disk, and the other partition as a mirror of the other 4TB disk. I wouldn't attempt to speculate on the performance or integrity characteristics of such a pool, but it can be built.

 

[jgreco]

Well, you can, but it's an unsupported configuration with, IMO, no real reason to recommend it, and you'll need to do it all at the CLI. But if you manually partition the 8TB disk into two 4TB partitions, you can add one of the partitions as a mirror of the first 4TB disk, and the other partition as a mirror of the other 4TB disk. I wouldn't attempt to speculate on the performance or integrity characteristics of such a pool, but it can be built.

Yeesh. Can we avoid suggesting off-road canyon adventure driving to the user who is just learning the fundamentals of how to drive their ZFS car?

This is not going to be properly manageable via the web UI and is going to have various hidden landmines associated with it.

 

[Arwen]

@danb35 - Anyone contemplating that trick, needs to know all the details themselves.

For example, having vDev 0's 4TB disk fail, along with vDev 1's 4TB partition of the 8TB disk fail. You have to know what order to replace the disks. If the 8TB was pulled first, total pool loss. And of course, when it is time to replace the 8TB disk, total loss of redundancy for those 2 vDevs.

This is like contemplating De-Dup. If they don't understand it enough and are asking questions, then they should not be implementing De-Dup.

 

[AlexGG]

thought that I could mirror a Stripped set of disks to one physical drive with the same capacity...

Technically you can. I would not recommend it though. The performance is abysmal, due to added seek requirements. Also, failure modes multiply and flourish.

such as the spectacular (and admittedly slightly ridiculous) PSU failure

Spectacular indeed, but it happens more often than you apparently think it does. Another thing is a brownout, causing all disks to sort of flicker on and off, for lack of better description. The third thing is multiple repeated unscheduled power cycles, which is not supposed to bring the pool down, but it does. I think power issues of one sort or other are the second most frequent reason for data loss in the real world (the first being operator error).

 

[jgreco]

Spectacular indeed, but it happens more often than you apparently think it does.

I'm well aware of the actual failure modes of these sorts of things. I've been working with systems and hardware professionally for decades.

There are certainly lots of crappy low quality power supplies out there where proper design hasn't been a consideration, and these things can certainly be hazardous. If we switch tracks for a bit to something where it may be easier for an electronics novice to actually see this ... my favorites are some of the teardowns that have been done on "fake Apple" iPhone-style 5W chargers for 240VAC countries. These almost always use a two-PCB design with poor high voltage separation. Dave Jones did an excellent teardown many years ago on one of these, which is my standard go-to when explaining to people why they should just bite the bullet and pay the twenty bucks for a legit charger, or at least buy a charger from a known-quality manufacturer like Anker. That $4.99 charger you picked up at the dollar store ... not a good idea.

This of course carries over into PC computer gear, because a lot of PC components are designed for the whitebox PC market. These are often poorly designed and use low quality components, which is the main way to hit that $12.75 price point.

If you've ever looked at the design of Apple's 5W charger, it's pretty clear that Jobs let a team of engineers loose on the problem, and I wouldn't be shocked if several man-years of effort went into the design of that thing. Part of that, of course, is the Jobs-ian desire for small, sleek, and sexy, but also that it has to be super-safe and reliable and not causing houses to burn down.

But an engineer might cost $100K/yr, so putting even a single man-year of effort into a PC PSU where the production run might be 100,000 units means that the cost of the PSU goes up by a dollar just to pay the engineer. (I'm using rounded numbers here to make a point).

And for a cheap generic PSU, it mainly has to survive its warranty period. It doesn't have a brand name that's going to suffer significant reputational damage when it fries your Windows PC and takes your data with it.

That's part of why I aim people at non-cheap, well-designed, name-brand power supplies in the Proper Power Supply Sizing Guidance resource. These sorts of failures are not unheard-of even with decent PSU's, but they are RELATIVELY uncommon when you are using well-designed PSU's.

It's still a hazard, which is why I mentioned it as a system-destroying failure mode in post #2.

So I'm going to take exception to the quoted bit. I probably have a very good handle on this, having dealt with this kind of thing professionally for decades.

 

[danb35]

Anyone contemplating that trick, needs to know all the details themselves.

Fair enough. And to OP, to be clear, I wouldn't recommend doing what I mentioned. It's possible, in the way described, but I really can't think of a situation where it'd be a good idea. Lots of pitfalls there.

 

[jgreco]

I'm going to take exception to the quoted bit.

Upon re-reading, this came off a bit harsher than intended.

Not like Superman-frying-your-head-to-a-crisp-with-heat-vision level exception.

Just good natured exception. But the topic was and is serious, and I definitely see both the risks of PSU's, especially cheap PSU's, along with the reduced risk inherent in a well-designed PSU, which I heartily endorse.

None of this will save you when a high voltage transmission line falls on your local electrical distribution, or lightning strikes, or [....more unlikely but not unheard-of stuff....] so I definitely want to emphasize that @AlexGG is right in being paranoid about "things happening".

And now the OP is wondering "OMG what did I get involved with here" heheheh

 

[Caligulae]

Wow! Thank you all for the great stories, advices and knowledge dumping!

As a very wise creature said: "A bit overwhelmed, I am..."

I have lots to think about and read on the ZFS environment (thx for the slap on the hand and SMH @jgreco ...!). I will re-think of my strategy and for sure, I will investigate the use of an offsite server (Planethoster) where I could do my snapshot/backup.

FYI, I do have a APC UPS on which my FreeNAS is connected to and my 850W Corsaire is about 1 year old.

So, to summarize:

1. Wanted fast access within the house
2. Wanted full restore in case of a disaster without going to several sources/programs

My Christmas vaca will be pretty busy!

 

[jgreco]

My Christmas vaca will be pretty busy

And your wallet lighter? Don't forget, NewEgg and Best Buy sell gift cards, and/or you can tell all your friends and family what sort of gifts you need to feed your NAS-ty habit. ;-)