RAID .. just moving the single point of failure ??

[poyntesm]

Hello all,

Hope I can get some help/opinions.

I am researching my needs for building a NAS with RAID. The total overall storage requirement will not be huge (2TB). Future expansion needs to be possible. So for best chance of recovering from a disk failure I want to use RAID6. I plan on use of 4 disks to start. 2 for data and 2 for partity

Key for me is how any failure in the chain to data access will be handled. My current understanding is that RAID shifts the single point of failure away from disk. Great. My concern is where is the next SPOF ... no point just moving it in my mind, it needs to be removed as good as possible. However I am not interested in true HA via a cluster, but I am interested in making sure the SPOF is a comodity piece of standard hardware .. i.e the motherboard.

If you have a RAID with a single controller if that controller fails you are in the same boat as with no RAID and a disk failure. You lose access to data. Not good. The controller you bought might no long be made or available and what happens then? you are up the smelly stream with the wodden object

I take it each brand will not be compatible with others, and I would even hazard a guess differing modesl from same brand.

So can you have 2 RAID controllers in 1 PC, with both connected to the 4 disks. if one controller should fail the second takes over, i.e there is 2 paths to each disk and in reality the parirty is stored in 2 disk. Meaning to lose data I need to lose either 2 controllers or 2+ disks. I am not worried about a hot failover, cold via reboot is enough, although hot would be even better.

If a motherboard fails well they are comodity items and I should be able to easily replace as I imagine the RAID config is on the cards? I hope anyway.

I persume I could just add more disks to get more storage also. So run out of space and I will add more disk.

I will also need to ensure a off site regular backup for real saftey.. that I know.

Hope that all makes sense.. I am not trying for overkill but I see no point in moving the SPOF to a specialised piece of kit.

Can I get your feedback/thoughts on the above.

Thanks,
Esmond

 

[VirtualLarry]

Well, mobo RAID is tied to the particular chipset on the mobo used to implement the RAID. Intel chipsets and Intel RAID software have been good about backwards compatibility. I've heard stories of people that upgrading their Intel-chipset mobo, and migrated their RAID arrays without issues. I don't know about NV or AMD RAID-capable chipsets, whether they are likewise.

If you go with an add-in controller card, then yes, I would recommend purchasing a spare, just in case. It's pretty important not to lose access to your data, if the controller card were to fail.

I'm not aware of any consumer controller cards that allow you to connect to consumer HDs in a manner that allows fail-over. For that, you would need to step up to Fibre-channel HDs, which is a whole nother ball game.

Also, why RAID6 with only four drives? Why not RAID 10?

 

[mvbighead]

Well, mobo RAID is tied to the particular chipset on the mobo used to implement the RAID. Intel chipsets and Intel RAID software have been good about backwards compatibility. I've heard stories of people that upgrading their Intel-chipset mobo, and migrated their RAID arrays without issues. I don't know about NV or AMD RAID-capable chipsets, whether they are likewise.

If you go with an add-in controller card, then yes, I would recommend purchasing a spare, just in case. It's pretty important not to lose access to your data, if the controller card were to fail.

I'm not aware of any consumer controller cards that allow you to connect to consumer HDs in a manner that allows fail-over. For that, you would need to step up to Fibre-channel HDs, which is a whole nother ball game.

Also, why RAID6 with only four drives? Why not RAID 10?

RAID6 can withstand the loss of any two drives. RAID10 can at most sustain a loss of 2 drives, where if two drives of the same portion of data are lost, all data is lost.

As for the controller issue, I can certainly see the OP's point. The controller is indeed a single point of failure. Not sure if ZFS or any other other software forms of RAID can readily be swapped to entirely different hardware and still be recovered... would be interested to know if someone has done that though...

Otherwise, buy two add-in controllers and when you lose one, swap in the backup controller and write your data to a single large disk, and then upgrade and buy another two controllers. Certainly not a very likeable plan, but it sure beats not being able to find the exact controller you are looking for.

 

[Mark R]

As for the controller issue, I can certainly see the OP's point. The controller is indeed a single point of failure. Not sure if ZFS or any other other software forms of RAID can readily be swapped to entirely different hardware and still be recovered... would be interested to know if someone has done that though...

Most software RAID/advanced filesystems should be completely transferrable to an equivalent operating system.

E.g. I have a home NAS (QNAP) - it actually runs a customised version of linux on an ARM processor. For a SHTF data recovery test, I pulled the drives out of the NAS, plugged them into a spare PC (regular core 2 duo) and booted a linux live CD - the RAID array was detected, configured and the partition automounted. The files were instantly accessible.

I even tried it with a missing drive (3 out of 4 drives connected) the PC and live CD could read the data straight away with no configuration.

 

[mvbighead]

Most software RAID/advanced filesystems should be completely transferrable to an equivalent operating system.

E.g. I have a home NAS (QNAP) - it actually runs a customised version of linux on an ARM processor. For a SHTF data recovery test, I pulled the drives out of the NAS, plugged them into a spare PC (regular core 2 duo) and booted a linux live CD - the RAID array was detected, configured and the partition automounted. The files were instantly accessible.

I even tried it with a missing drive (3 out of 4 drives connected) the PC and live CD could read the data straight away with no configuration.

Very cool. I had kinda figured that was how it would work, but since my data is stored on a simple mirrored config, I have never ventured into configuring a ZFS or similar file system for a file server. Tempting though.

 

[Paperlantern]

The controller is a single point of failure, yes, but the idea of RAID is to move the single point of failure from the part that has constant moving parts, higher operating temp, and a higher failure rate, to a part that has no moving parts and lower operating temp, and should generally not fail (or fail MUCH less often). Its not a matter of moving the point of failure to another single point of failure (though i wont argue the fact that that IS what theyve done because all components CAN fail), its a matter of moving it to a part with a smaller possibility of failure.

 

[mfenn]

OP, the key point is what you addressed at the bottom of your post, namely RAID is not a backup. (Yes, it is that important).

RAID is for increasing the availability of data. Backups are for preserving the integrity of data. By going with any sort of RAID at all, you've already significantly reduced the chance that the data will become unavailable. The 80/20 rule applies here, so I would not recommend going beyond a single RAID 5 array unless and until you're in a true enterprise environment. That money would be better spend on a solid backup solution. If the controller dies and you can't find a replacement, no big deal, just restore from backup, which shouldn't take more than an hour or so for 2TB of data.

So, in short, don't blow all your money on a fancy RAID setup before you spend money on backup.

 

[Zap]

OP, the key point is what you addressed at the bottom of your post, namely RAID is not a backup. (Yes, it is that important).

RAID is for increasing the availability of data. Backups are for preserving the integrity of data.

Couldn't have said it better myself.

 

[poyntesm]

Thanks all. I fully understand RAID is not a backup solution that some think it is.

So knowing I will still need a backup I have decided to worry less about the controller failure and decided a failure of that will result in use of backup after a new controller is purchased, rather than try engineer around it.

 

[mfenn]

Thanks all. I fully understand RAID is not a backup solution that some think it is.

So knowing I will still need a backup I have decided to worry less about the controller failure and decided a failure of that will result in use of backup after a new controller is purchased, rather than try engineer around it.

:thumbsup: Sounds good!