Can you make SMR drives work

shrekfx

Cadet
Joined
Aug 5, 2021
Messages
1
I'm new to the whole NAS thing and will be building my first here shortly. I currently have some 3-4TB hard drives (which I'm guessing are SMR drives) and a newer Seagate 6TB Barracuda drive, (SMR) As with a lot of new NAS builders out there and I'm one to jump in and start playing with stuff, was not aware of the SMR and CMR drives. I have found that SMR does not play nice on NFS systems. I've tried to do more searching on this, but all I find is regarding the WD Reds that have SMR and the issues with those. But not much other info out there that I can find.

With out dropping a ton of money on new drives right away (wife and kids eat up most of the cash on hand) and use what I currently have, can I do the following?

I have an empty 6TB Seagate that I want to start out with and start moving data from some of my external drives over to this one. When those are empty, I want to add those externals one by one until I get everything moved over (I know I can't just take the filled drive and place it in because it will get cleared when pooled if I am correct.) So in all, I should have at least 4 drives. (Plan on shucking my externals I currently have so not fully sure what they are but guessing they are SMR since they are 3-4TB.)

Thanks for any thoughts. Again, I am going to get CMR drives down the road, but will come slowly.
 

HoneyBadger

actually does care
Administrator
Moderator
iXsystems
Joined
Feb 6, 2014
Messages
5,110
The 3-4TB external drives may be CMR, more likely if they're older models. It's worth shucking them first (or checking to see what expected drives are inside) before making too many negative assumptions.

The challenge I see here is how you will need to organize data to end up with some manner of redundancy in the end. If you have 3 3-4TB drives that are close to full, you're looking at 9-12TB of data. Even if we assumed you use a not-recommended RAIDZ1 you still only end up with 6-8TB of usable space across those three drives once pooled.

If you're willing to completely forgo redundancy then of course you can move a drive at a time, using the 6TB disk as a "temporary location" - but that's giving up a lot of the usefulness of ZFS.
 
Joined
Jan 4, 2014
Messages
1,644
@HoneyBadger Are mirrored SMR drives okay on TrueNAS? I'm not clear whether ZFS comes into play here?
 

Constantin

Vampire Pig
Joined
May 19, 2017
Messages
1,828
I doubt it.

The whole premise of ZFS/COW is the appliance knowing what the drive is doing. DM-SMR drives go AWOL when they're flushing their CMR cache to the SMR zone of choice, blocking any I/O until the flush is complete. In turn, a blocked mirror will block any further I/O to the affected pool. Mirrored pools may behave better since both SMR drives are likely to go AWOL at the same time and there are no VDEV-wide parity writes like there are in RAIDZ.

For a RAIDZ, both data and parity writes can trigger a pool-wide I/O block due to SMR cache flushes. I expect that's the reason the STH resilver test took 17hrs for CMR drives and 230hrs for SMR drives. I recommend that article also for the depth it goes into re: SMR.
 

HoneyBadger

actually does care
Administrator
Moderator
iXsystems
Joined
Feb 6, 2014
Messages
5,110
@HoneyBadger Are mirrored SMR drives okay on TrueNAS? I'm not clear whether ZFS comes into play here?
SMR mixed with a copy-on-write (COW) filesystem like ZFS is a bad combination regardless of vdev geometry, until the filesystem is rewritten to take advantage of it.

It isn't "writes" that causes the SMR stalls, it's "rewrites" or "overwrites." New writes into empty areas, especially done in larger recordsizes, can be handled because it doesn't need to re-shuffle any existing data on those sectors. So if your intention is to fill a pool with multi-GB "Linux ISOs" or other archival workloads, then your drives should be able to happily lay those onto a series of overlapping 256MB SMR zones. Even if you delete those large chunks of files, you should hopefully be freeing entire zones at a time, and only have to reshingle the small amount of zones where "File A" ends and "File B" begins. Give enough time between the deletes and the new data ingest for your drives to fix themselves, and you won't really notice.

The problem comes when you start introducing small records or overwriting chunks of larger files frequently. Basically, the more often you "poke holes" in the SMR zones, the more frequently they have to reshingle, and that's what causes the drive to stall out doing internal housekeeping.

Once HA (Host Aware) and HM (Host Managed) SMR drives are commonplace and ZFS has code to support them, ZFS may be able to handle some of that housekeeping internally - controlling the data placement on the platters themselves to say "okay, this is a larger record, I can drop it on an empty SMR zone nice and fast, giving me X ms of free time, I also know the size of the CMR cache area is and how full it is, so let's do a little housekeeping now to preserve my ability to keep dumping data into an empty SMR zone."

DM-SMR doesn't communicate anything back to the filesystem; it's more of a compatibility layer pretending to be a regular CMR drive. Works okay as a single drive for many filesystems, but ZFS needs to know what's going on under the hood to use it efficiently.
 

olddog9

Dabbler
Joined
Apr 23, 2021
Messages
28
My opinion ...based on several years as a project manager in storage engineering (hard disks and tape) ...
Magnetic rotating disks that are now branded as 'shingled magnetic recording' SMR ...
... are exactly the type of design that disk design engineers (yes the old guys) tried to AVOID !

Design engineers have done an AMAZING job of packing data denser and denser. I applaud them for their skill, training, knowledge, and innovation. They improved magnetic recording head design and manufacturing processes. However there are inherent limits. You may have heard the vocal expression "Packing 10 pounds of stuff, in a 5 pound bag".

A consequence of this 'SMR' design is ...eventually the recording head writes on the correct track yet re-records the EDGE of the adjacent tracks (one inner edge, one outer edge). The magnetic field 'bleeds' over to the neighboring tracks and corrupts the edge of that track. This is BAD ! This over packed design introduces a higher risk of unintentionally damaging data on adjacent tracks. This manifests itself as read errors and increased latency because the drive firmware says 'read it again', so we wait another another full rotation to attempt the read ... again.

I've been away from storage design for years, until I recently needed a new server. When looking at new disks, I could not believe the disk companies were selling these so called 'SMR' designs.
 

Constantin

Vampire Pig
Joined
May 19, 2017
Messages
1,828
It speaks to the lack of investment at the remaining big three OEMs to choose a software solution to solve a hardware problem. The software is cheap and many consumers cannot articulate the difference in HDD performance.

As I see it, a lot of milking of cows while the folk at the top get rich. No real competition from SSDs either on account of limited fab capacity and building. So they’ll keep milking this cow as long as the market allows them to.
 

Arwen

MVP
Joined
May 17, 2014
Messages
3,600
Don't forget that their is a SERIOUS problem with WD Red SMR drives and ZFS. This does not seem to apply to the Seagate Archive SMR drive line, (I have an 8TB model used with ZFS and have not had problems other than writes being slow).

Here is what I think is happening:

If the host attempts to read a block it has not written, the drive returns an error. Unlike a CMR / conventional drive, this change in behavior causes ZFS read ahead to fail. ZFS can combine 2 or more reads of separated groups of blocks into one drive read request. This can include blocks ZFS has not written, that are in-between the other, real blocks. Because ZFS is one of the few, (possibly only), file systems that does this, it was the one most affected by this "bug". (It's a bug in my opinion...)

Could this problem with WD Red SMR drives be overcome by writing to the entire drive?

I don't know. But, I dislike firmware bug work arounds, because we don't know what else is waiting for us to find, (or experience the hard way!).
 

no_connection

Patron
Joined
Dec 15, 2013
Messages
480
How is SMR doing in 2022?
I am thinking of getting (pretty cheaply) four 5TB Seagate 2.5" drives and stick them into my G8 host. Pass through SATA controller to TN and set as a Z1.
It would be used as bulk storage of movies/videos/vods/ and misc/backups/backupsofbackups/ that gets put on there and stays until end of time.
I would not care too much about performance as long as I can read whatever file faster than it plays, which is limited by 1Gb connection anyway.
 

HoneyBadger

actually does care
Administrator
Moderator
iXsystems
Joined
Feb 6, 2014
Messages
5,110
How is SMR doing in 2022?
The situation hasn't changed any.

Your proposed SMR RAIDZ would likely be "acceptable" as your use case never requires a fast response on an overwrite or rewrite - dumping large files onto the disks will go quickly, and reads will of course be fine.

The SMR issue will only rear its ugly head if/when file deletions occur, and the drives start freeing up space that doesn't align with the SMR "zone" boundaries. If the pool is given sufficient "free time" then the drives can reshingle and be ready for the next set of writes.
 

ChrisRJ

Wizard
Joined
Oct 23, 2020
Messages
1,906
At the end of the day the question to ask yourself is: Why do I want to use ZFS? Because yet another YouTube video was released that just scratches the surface? Or because you value your data? In the latter case the answer should be rather obvious.

This may sound arrogant, although I can assure you that it isn't meant this way. I just got the impression over the last weeks that more and more people come to TrueNAS with little understanding about how it is different from "normal" file systems. That is not bad and of itself. In fact, I think it is great to many new folks here.

But without knowing what I really need (or at least want), and what ZFS can and cannot deliver, buyer's remorse is almost guaranteed. Given the amount of message you already wrote, you are likely not the "typical" candidate in this respect. Still I wanted to minimize the risk of you being disappointed.

Good luck!
 

Arwen

MVP
Joined
May 17, 2014
Messages
3,600
One last comment of using 4 x 5TB 2.5" Seagate SMR drives in a RAID-Z1.

Make sure you have enough cooling.

During ZFS scrubs, (which should be done one or more times a month), all data & parity will be read. Because these disks are SMR, the data may be scattered across the cylinders, instead of contiguous, (or close to), sectors in a sequence. This SMR side effect can cause higher than non-SMR head movement, such that that the head movement motor gets more of a work out than non-ZFS setup. And creates more heat.

So, make sure your drives have enough cooling.
 

no_connection

Patron
Joined
Dec 15, 2013
Messages
480
I will remember the heat thing in case I end up with a different server that uses SMR in the future. DL360e should be fine given it's absurd obsession with cooling.
It's pretty much four drive slots in a host that is already running making it a compelling option compared to another server that uses power.

I am tempted to drill a hole and use a breakout cable tho. Well see.

I was hoping ZFS had evolved some SMR understanding so it won't freak out and go kick the poor pool whenever it had a bad day or a few too many drinks. Oh well.
 

HoneyBadger

actually does care
Administrator
Moderator
iXsystems
Joined
Feb 6, 2014
Messages
5,110
I was hoping ZFS had evolved some SMR understanding so it won't freak out and go kick the poor pool whenever it had a bad day or a few too many drinks. Oh well.
If said understanding does ever happen, it's likely to require Host Aware or Host Managed (HA/HM) SMR on the part of the drives. All current consumer models are Drive Managed (DM) SMR, which basically means they hide their "SMR-ness" from the connected machine.
 

Constantin

Vampire Pig
Joined
May 19, 2017
Messages
1,828
All current consumer models are Drive Managed (DM) SMR, which basically means they hide their "SMR-ness" from the connected machine.
That’s the fatal flaw for all consumer SMR drives ATM. HA/HM SMR drives are going on a B2B basis directly into data centers who have the requisite hardware and software to deal with HA/HM SMR.

Hence, there is ZERO reason for iXsystems to start working on supporting HA/HM SMR unless a very big customer makes it a TrueNAS adoption requirement. IMO this is unlikely to happen.

But even if iXSystems were to support HA/HM Drives in the future, you’d have to wait for said drives to trickle used out of the data centers unless drive OEMs start shipping HA/HM drives to the unwashed masses. This too seems unlikely given the implementation headaches that may be caused in the marketplace re: consumers mistakenly buying the wrong drive type for their computers.

It’s also the reason that DM-SMR was developed in the first place, ie a software algorithm that stuffs 20% more data into a SMR vs. CMR drive at virtually no cost to the OEM and a performance hit that most customers seem to be OK with.

Initially (before channels such as serve the home, etc raised the alarm) many consumers bought SMR drives for the same money as better-performant CMR drives, a deception the drive OEMs were only too happy to engage in.
 
Last edited:
Top