Buying new hdd's for z2 pool - Unrecoverable Read Errors interesting read

[Jayos]

I have recently updated my hardware and now ready to purchase new drives to create a new pool with more storage.

Whilst researching different drive options I have decided to go for a 5 disk z2 pool with 3TB hdd's as it seems to be cost effective and within budget.
My current z1 pool is 3x2TB WD Red.

I have stumbled across a few articles that discuss URE and the chance of disk failure when re-silvering. Based on the details provided by WD my current drives have a value of <1 in 10^14
whereas the WD Red Pro & WD Gold have the value <1 in 10^15
This may not seem too important but look at a thread in another forum it could make a significant difference :(https://forum.synology.com/enu/viewtopic.php?t=96171)

Hard drives have built-in logic that is intended to correct certain types of read errors - the errors that escape without automatic correction are classified as unrecoverable read errors (URE). If you take a look at the manufacturer's specifications, you will see that the specifications list less than one in 10^n bits for the URE statistic:
Western Digital Red
http://www.wdc.com/wdproducts/library/S ... 771442.pdf
<1 in 10^14

Western Digital Red Pro
http://www.wdc.com/wdproducts/library/S ... 800002.pdf
<1 in 10^14 (it seems like this document a couple of months ago used to state <10 in 10^15)

Western Digital RE4
http://www.wdc.com/wdproducts/library/S ... 701338.pdf
<1 in 10^15

So, that <1 in 10^14 URE statistic could mean that one URE might be encountered in every 10^16 bits read. You might want to look at that statistic more as a measure of how well the drive manages bit rots (that is the eventual decay of magnetism that controls whether a bit is read as a 0 or a 1, not necessarily the chance of encountering an error when reading the same bit 10^n times).

Let's go back to the formula that I provided in my initial reply:
(1 - (99,999,999,999,999 / 100,000,000,000,000) ^ 48,000,000,000,000) = 0.380979164
So, for this RAID 5 array, rebuilding the array (when a failed drive is replaced) has a 38.1% probability of failure.

What about if enterprise class drives are used, with a 1 URE in 10^15 bits read statistic:
(1 - (999,999,999,999,999 / 1,000,000,000,000,000) ^ 48,000,000,000,000) = 0.046829645
Note that the probability of failure is now 4.7% - a considerable improvement.

The formula for a RAID 10 array results in the number 0.015860094, so that is roughly 1.6% probability of failure during the rebuild of the RAID 10 array.

It seems it may be much safer to go for the more expensive Enterprise drives, but with 5xhdd's this then changes the total cost significantly. Also I'm struggling to find 3TB Hdd's in the pro variant or another manufacturer with <1 in 10^15 values.
What are your thoughts?

 

[Chris Moore]

This may not seem too important but look at a thread in another forum it could make a significant difference

My first thought is that you are building a FreeNAS system not a Synology system, so why are you looking in the Synology forum for information. Additionally, the referenced post is talking about RAID-5 where you were discussing RAID-z2 which is very different from RAID-5, so I would thing that, None of that is applicable to the ZFS file system that is used by FreeNAS.

Second thought, go with 4TB drives. They are a better value per TB vs 3TB drives. Also, Seagate drives are an even better value than WD drives.

Seagate Desktop HDD ST4000DM000 4TB 64MB Cache - $111.99

Seagate BarraCuda ST4000DM005 4TB 64MB Cache - $114.00

Seagate IronWolf 4TB NAS Hard Drive 5900 RPM 64MB Cache - $139.99

Seagate BarraCuda Pro ST4000DM006 4TB 7200 RPM 128MB Cache - $179.99

Seagate IronWolf Pro ST4000NE0025 4TB 7200 RPM 128MB Cache - $192.76

 

[danb35]

None of that is applicable to the ZFS file system that is used by FreeNAS.

Nonsense. The hardware error rate of the disks is most certainly relevant to ZFS, just as it is to any other filesystem. ZFS is great, but it isn't magic--it can't just resurrect data from dead disks (or even dead sectors).

 

[Pezo]

But he's got a point that it's more of a problem with RAID5 as opposed to ZFS, since ZFS can at least detect silent corruption.

 

[danb35]

Sure, and that's why I think the insistence on "RAID5/RAIDZ1 is dead" is a bit overstated. One URE on a remaining disk in a degraded RAID5 volume can kill the entire volume. That won't happen with ZFS because (1) most likely it would have been caught in a previous scrub, and (2) there are at least two copies of all metadata (and more than two of the really critical stuff), and it's all checksummed. That makes RAIDZ1 considerably more resilient than RAID5, but it still isn't magic.

If enterprise disks actually have a URE rate 10x better than consumer-level disks, that's a perfectly valid reason to prefer them, and that's true whether you're using ZFS, hardware RAID, or non-ZFS software RAID. I'm a little skeptical that this is the case, but that's a separate issue.

 

[Evertb1]

. Based on the details provided by WD my current drives have a value of <1 in 10^14
whereas the WD Red Pro & WD Gold have the value <1 in 10^15
This may not seem too important but look at a thread in another forum it could make a significant difference :(https://forum.synology.com/enu/viewtopic.php?t=96171)
It seems it may be much safer to go for the more expensive Enterprise drives, but with 5xhdd's this then changes the total cost significantly. Also I'm struggling to find 3TB Hdd's in the pro variant or another manufacturer with <1 in 10^15 values.
What are your thoughts?

My thoughts are that you just should buy the disks you can afford. Preferably of a brand and type that are known to be halfway decent. Getting your self wind up about things like URE, in my opinion, is useless for a server that is not mission critical (and even then). You have chosen for ZFS. Together with a for home use reasonable Raidz setup and an bunch of nice disks you are on the right track I think. Now take care that you have a good backup plan and of you go. And think of URE as just another 3 letter abbreviation.

 

[Stux]

Don't worry about it, use raidz2, and maintain a backup.

 

[Ericloewe]

Yes, the money spent on more expensive disks is better spent on an extra layer of backup, past the basic NAS models.

 

[Jayos]

My thoughts are that you just should buy the disks you can afford. Preferably of a brand and type that are known to be halfway decent. Getting your self wind up about things like URE, in my opinion, is useless for a server that is not mission critical (and even then). You have chosen for ZFS. Together with a for home use reasonable Raidz setup and an bunch of nice disks you are on the right track I think. Now take care that you have a good backup plan and of you go. And think of URE as just another 3 letter abbreviation.

Yes, I have a good backup plan and moving to RAIDZ2 will improve things anyhow. I guess I was just getting hung up on the risk of failure 34% vs 4.7%.