Some interesting insight what kind of RAID/redundancy & what HDDs

[jeyare]

there is an really interesting answer why I don't like RAID5 (SHR >3xHDD), specially with low cost HDD inside.
Many discussion was here about reliable redundancy architecture or usefulness of indicator MTBF for a disk choice. MTBF is just a statistical number of AVERAGE time = then it's same like an average rate of crashed planes for travelers.

Yeap, here is most sophisticated description of my alerts for use of RAID5/SHR everywhere as standard. Yes, it's cheap from the available space point of view. But really risky base for a failure events.
When you will take into consideration, that the risk of the failures increase with size of the every single HDDs in the arrays. You have an answer why you don't combine >4TB low cost HDDs in RAID5/SHR. It's about physic.
You need count with less lifespan of such disk in such arrays.

The Math on Hard Drive Failure

Since I haven't seen any discussion on this issue (well, anything more than speculation), I thought I'd go ahead and do some math on exactly what is the likelihood of HDD failure (and, by extension, array failure). I'll be using the data from Google's HDD study...

www.truenas.com

[automerge]1611830354[/automerge]
and here is another point of view, w/o any details to operational conditions:

2020 Hard Drive Reliability Report by Make and Model

The AFR for 2020 dropped below 1% down to 0.93%. In 2019, it stood at 1.89%. That’s over a 50% drop year over year.

www.backblaze.com

for me it's just a marketing data w/o insights to deep dive - like arrays structure, operation temperature, workload, ...
then pls. read the data carefully

 

[WST16]

interesting answer why I don't like RAID5

So what do you like?

 

[jeyare]

there isn't easy answer, like apple or pears:

Lot of variables there:
1. different scenario for some who needs store just media downloaded from web. There is better approach to store in RAID0 than RAID5, because array rebuild for RAID5 is really higher than new RAID0 and download of the media again.
Simple:
20Mbps WAN download speed = 2,5MB/s = 211GB/day .... 9 movies in 4k/HDR/Atmos .... for next 9 days of cinema watching
100Mbps ----> 1055GB/day .... 45 movies in 4k/HDR/Atmos
500Mbps ----> 5273GB/day .... 225 movies in 4k/HDR/Atmos
yeap, but with RAID5 is there the redundancy. True. But same time for the array rebuild.

2. sensitive data for Drive operation. When you have more than 100GBof such data, then you still can use RAID1 with simple and better operation than RAID5.

3. When you have just bunch of everything in single volume, then you can use RAID5.

4. ....

5. ...

 

[WST16]

Ok. I got you. You don’t like it but you’d still use it if it fits the situation.
Life is pretty much like that.

 

[jeyare]

sure, it's like some corporate reports

 

[fredbert]

Given I'm already on RAID 5/SHR with 3x8TB Ironwolf 'NAS' disks and I suspect am not alone here, what do these reports mean by low cost disks? Non-NAS; Non-NAS 'pro'; Non-NAS 'enterprise'?

With a RAID 5 there is the heighten risk of full array failure when two disks fail out of N (e.g. in a 3 disk array 66.6% have to fail; 50% in a 4 disk array) versus a RAID 1 of two disks where 100% have to fail. From a physical failure point of view the more you have there is more risk one or more failing.

I'm wondering if there's positive mileage in building up the RAID with extra disks over time: don't overly populate with disks until storage is required. While this would put stress on disks during rebuild the newer disks would be from different batches and different power-on times.

Then there's the alternative approach of what to do instead of RAID 6 but you need more space. If you have a 6-bay then two RAID 5 of 3 drives means even if 2 fail there's a maximum of 50% of storage lost (more than 6 disks in RAID 6) but 3 fail and it's still 50% max (but 100% in RAID 6). Similar on a 5 bay NAS. In fact worth looking at two smaller RAID before opting for one RAID 6.


Then there's all the P(URE) stuff, which looks neat as 'pure', but really just

 

[jeyare]

Low cost disk in such discussion is an usage of desktop HDD instead NAS recommended e.g. Barracuda vs. IronWolf/Pro vs. Enterprise level (Exos). More things for an evaluation are there: vibration, thermal output, helium filled, Workload level, block level geometry, ... maybe low cost isn't the right label, because:
- you can purchase Exos datacenter line for better price than IronWolf.

Another point (main topic of this thread) is physical failures probability of HDDs, operated in different arrays based on:
- how many disk you have in the array
- how many of them are for the redundancy ... you can operate RAID1 with 3 HDDs also, then you have 2 redundant HDD
- what is the block structure
.... and finally - how an operation temperature makes impact for the life span - you can read it in the primary link proposed (there is really interesting analyze from their data centers):

What is problem for some, who would like silent NAS in sleeping/living room, enclosed in cabinet. Don't mention about silent rack mount cases. Then number of HDDs in the used RAID scenario make heavy impact for the temperatures = AFR increases. Specially in the silent mode.

Yes, it's up to your business model for the NAS. I don't like generalize that there is just single rule for all.
Take this thread as point to thinking about, when RAID5 is one and only "economy" driven solution. For whom?

 

[Dexxer]

I'm on RAID5 with 6 x 16TB Exos drives. They are very loud, but good drives.
Room temps 23C, disk temps 32C.

I did not go RAID6 because I really needed the space. Already used 52TB only have 18TB left on a 6 bay :-(
I'll need to buy a DX517 expansion by next year.

I know in case of failure, it might take 5 days to rebuild a 16TB drive... But I have other backups meanwhile.
DSM 7 says they solved that with a special algorithm that goes faster.