WD Elements all SMR drives - looking for advice on management

leprejohn

Cadet
Joined
Jun 1, 2019
Messages
9
Hello IX community how are you doing? I brought last Sept, 4TB elements drives which turned out are all WD40EMAZ, finding out these are all SMR drives and I've currently got them in a raid-z1 array.

With a drive failure and the rebuild times are quite long I was wondering sort of what would my options be on these encase I get one that failed. Unfortunately I don't think I can just replace these for something like IronWolf drives due to funds all at once.

Going forward would freenas be okay with the SMR drives or should I look into another solution say windows storage spaces or unraid?
 

Constantin

Vampire Pig
Joined
May 19, 2017
Messages
1,829
I’d consider gradually replacing the SMR drives with CMR drives as they fail. That way, you don’t incur a huge expense and the rebuild time will be low because the CMR drive does not suffer from the DMSMR pool Rebuild performance deficit.

That said, if SMR is having a big performance impact on your day to day use case, there is always the option of eBaying the whole lot of drives and starting with a new pool of CMR drives. If you go that route, I’d also suggest taking a harder look at going Z2.
 

leprejohn

Cadet
Joined
Jun 1, 2019
Messages
9
I’d consider gradually replacing the SMR drives with CMR drives as they fail. That way, you don’t incur a huge expense and the rebuild time will be low because the CMR drive does not suffer from the DMSMR pool Rebuild performance deficit.

That said, if SMR is having a big performance impact on your day to day use case, there is always the option of eBaying the whole lot of drives and starting with a new pool of CMR drives. If you go that route, I’d also suggest taking a harder look at going Z2.

Performance doesn't seem to suffer during daily use, backups etc I couldn't see any issues with any of the drives which is why I don't want to replace all of them at once.

Looking at it, I will be okay for now but I'll start maybe replacing them as I can afford to do so
 

Heracles

Wizard
Joined
Feb 2, 2018
Messages
1,401
Hey Leprejohn,

I’d consider gradually replacing the SMR drives with CMR drives as they fail.

I really disagree with @Constantin on that one... The SMR drives are the most problematic during resilvering, so you should avoid waiting for that critical operation. Second, it is the moment where your pool, so your data, is at its most vulnerable. Again, I would not do the most problematic when the data are the most vulnerable. Finally, Raid-Z1 is not strong enough to protect your data efficiently, so I would take the opportunity to convert the pool to another structure.

So I would deploy a brand new pool, Raid-Z2 or mirrors, and migrate the data from the old pool to the new one.

How many drives can you fit in that server ? Do you have access to a second system that can hold your data temporarily ? Do you have complete backups of your data ?

Good for you that you can address the problem before the catastrophe...
 

Constantin

Vampire Pig
Joined
May 19, 2017
Messages
1,829
Heracles, I guess I must be confused about what resilvering is then?

Isn't resilvering the process of reconstituting a missing / failed drive based on parity data saved on other drives? SMR drives behave fine as long as their CMR-based cache is not exhausted while they are being written to. Granted, SMR drives may also suffer from further performance losses due to garbage collection and similar housekeeping tasks, but I'd like to think that those kinds of maintenance processes generally execute when the SMR HDD is otherwise idle.

Coming back to the OP question: If your NAS is only reading from SMR drives while writing to a replacement CMR drive, then where is the performance harm?
 

FJep

Dabbler
Joined
Mar 7, 2019
Messages
38
Have you tried contacting WD-Support?
I understood that they gave some people free CMR replacements for there SMR drives.
Perhaps it's worth a try.
 

Apollo

Wizard
Joined
Jun 13, 2013
Messages
1,458
Just watched this youtube video earlier today.

 

Constantin

Vampire Pig
Joined
May 19, 2017
Messages
1,829
Have you tried contacting WD-Support?.
I agree! The age of the drives might make that more difficult. But multiple users have had luck insisting with WD support to get their SMR drives replaced with CMR drives, so it’s worth a try.

I’d emphasize how your NAS is not as responsive / performant as it ought to be, that you couldn’t figure out why until it was disclosed that the drives were SMR, etc. The key being that DMSMR drives having recently been disqualified by multiple major NAS vendors. DMSMR drives have no business in a NAS application yet that is what they were marketed towards.

Patrick at ServeTheHome has not only a good video and article on how SMR is no good for ZFS, he also found a video from 2015 in which a HGST employee details why DMSMR drives are problematic for ZFS. So there is hope that WD will replace the drives, especially since they're facing a class-action lawsuit at this point.
 
Last edited:

Heracles

Wizard
Joined
Feb 2, 2018
Messages
1,401
I guess I must be confused about what resilvering is then?

Isn't resilvering the process of reconstituting a missing / failed drive based on parity data saved on other drives? SMR drives behave fine as long as their CMR-based cache is not exhausted while they are being written to.

Hi @Constantin,

The NAS will keep running during the re-silvering process. As such, it is not dedicated to read requests. The drive being already loaded with requests, it may very well turn to problematic when mixing the massive read requests with the regular ones, the regular write requests, the garbage collection and the rest of its regular activity.

Should the drive suffer ANY hiccup during that process, ZFS may well mark it as non-functional, pushing you to 2 failed drives in a single Raid-Z1 structure.

I would clearly not take that risk.
 

Constantin

Vampire Pig
Joined
May 19, 2017
Messages
1,829
In an ideal world, the OP will be able to replace the SMR drives with CMR drives on the basis of goodwill from WD. That's the first thing I'd try and thank you @FJep for suggesting that step. Chances are, the replacement drives will be refurbished, but so are the SMR drives being sent back in. Hopefully, there will be some sort of warranty. Additionally, I'd build the CMR-only pool with Z2 VDEV(s) unless there are multiple backups nearby.

As for the gradual replacement of SMR drives via CMR drives, the OP has mentioned that the SMR drives work fine on a day-to-day basis. To me, that signals that the OP's use case / workload is light enough not to repeatedly trigger known DMSMR issues such as CMR-cache transfer, garbage collection, and so on. Thus, I still consider the gradual replacement of SMR drives with CMR drives an option since the pool use appears to be light. But yes, I'd leave that pool alone while it resilvers (I do that for my regular, CMR-based pool also).

On higher-workload machines, the impact of SMR drives becomes very apparent once the drives start filling up, stuff gets fragmented due to heavy work, and the SMR controller never gets the quiet time to defragment, flush the cache, collect garbage, whatever. That's how Patrick explained the time delay between WD starting to ship allegedly-qualified DMSMR NAS drives and customers starting to complain to their vendors./ forums re: cratering pool performance.

Customers like Backblaze would catch this stuff quickly (they have foresworn the use of SMR drives altogether) but they likely do not buy sub-8TB drives at this point since they're transitioning to 12TB drives and beyond. Thus, there was no quick market pushback from high-volume buyers to alert the management group at WD that the DMSMR-for-NAS-drives decision was a really bad one. Like Patrick Kennedy, I would like to give WD the benefit of the doubt re: boneheaded business decisions.

However, the more cynical among us would infer that WD knew this thanks to presentations given in 2015 re: the unsuitability of DMSMR for NAS applications and hence
  • only saddled the lower-capacity NAS Reds with DMSMR technology
  • did not update their spec sheets to reflect this very important change
  • took longer than they should have to acknowledge the issue
More importantly, I wish that WD simply bit the bullet and offered all its customers with SMR Reds the option to replace them with refurbished / new CMR Reds without a lot of fuss. That is, set up a web site, let the buyers enter the relevant S/Ns, collect a CC number for cross-shipment, and presto, some customer goodwill is restored. The current drip-drip-drip approach to press releases screams that there is tumult @ WD re: who is going to be held responsible.
 
Last edited:
Top