Simple FreeBSD based NAS hardware, what would you recommend in 2026?

I have six 4tb disks in my ZFS2 and out of space. Movie server.

Bought three 18tb disks on sale before AI raped the pricing. Those are Unobtanium today due to ghastly high prices.

At my age I am content with what I have which will still last a long time. Yesterdays big Xeon is perfect for workstation with plenty of muscle.
 
bgavin said:
I have six 4tb disks in my ZFS2 and out of space. Movie server.

Bought three 18tb disks on sale before AI raped the pricing. Those are Unobtanium today due to ghastly high prices.

At my age I am content with what I have which will still last a long time. Yesterdays big Xeon is perfect for workstation with plenty of muscle.
Click to expand...
Now you have to keep an eye for 3 more. That way replace one, resilver, replace another, resilver.....eventually you have replaced them all and expanded pool

"What matters for a home fileserver is idle power usage. Which unfortunately is being driven up by each harddrive."
I think this is where we have the spinning disk vs ssd comes in. Cost of device against cost of power and timeline to ROI.
I don't know the answer to this but do know SSDs are quieter and typically generate less heat which feeds into power usage.
 
Not too long ago I came across this blog post for a cheap, low powered ZFS/NAS build. The case used alone is perfect for this type of build. I plan on using this as inspiration for my first cheap vanilla ZFS/NAS build in the near future.
 
Freedom said:
I have never been into "storage".
My understanding of it is basically there are mechanical spinning disk drive and some based on memory chips (faster and more expensive, especially now). And that ZFS is awesome.

I am now feeling the need for a NAS for the first time in decades but I have little idea how to approach this in regards to the rocketing cost of memory.
[...]
Click to expand...
Memory costs rocketing also applies, to a somewhat lesser extent, to spinning rust platters. Driven by increasing density and reducing cost for the consumer market (even the pro-sumer one) drive manufacturers came with this "artefact" that is named SMR; manufacturers were even hiding this important property from their customers. When selecting a HDD you may be not aware of, or overlook this aspect. See also here. Recently (Jan 2026) an OpenZFS issue has been opened: Docs: Add prominent warning about drive-managed SMR HDDs (CMR vs SMR) for RAIDZ/mirror resilver/scrub/send workloads*.

My strong advice for a ZFS based set up: do not use SMR drives!
(I suggest you do not use them at all.)

Perhaps one can tolerate the performance loss under "normal running" circumstances, they are absolutely desasterous for the resilvering process. A defective HDD transforms a ZFS pool with redundancy to a pool with less or no redundancy: a degraded state. When the defective drive has been replaced by a new one, the process of resilvering starts and you wil be confronted by the difference between SMR and CMR drives.

___
* There may come a time that ZFS becomes "SMR-aware", however:
Host-Aware SMR by Tim Feldman - OpenZFS Dev Summit 2014 - video - slides
I haven't noticed any results sofar, so I'm not holding my breath.
 
Erichans said:
Memory costs rocketing also applies, to a somewhat lesser extent, to spinning rust platters. Driven by increasing density and reducing cost for the consumer market (even the pro-sumer one) drive manufacturers came with this "artefact" that is named SMR; manufacturers were even hiding this important property from their customers. When selecting a HDD you may be not aware of, or overlook this aspect. See also here. Recently (Jan 2026) an OpenZFS issue has been opened: Docs: Add prominent warning about drive-managed SMR HDDs (CMR vs SMR) for RAIDZ/mirror resilver/scrub/send workloads.

My strong advice for a ZFS based set up: do not use SMR drives!
(I suggest you do not use them at all.)

Perhaps one can tolerate the performance loss under "normal running" circumstances, they are absolutely desasterous for the resilvering process. A defective HDD transforms a ZFS pool with redundancy to a pool with less or no redundancy: a degraded state. When the defective drive has been replaced by a new one, the process of resilvering starts and you wil be confronted by the difference between SMR and CMR drives.
Click to expand...
The ISSUES with SMR HDDs was on the agenda already 15 years ago.
the SMR drives were not suited for use with ZFS then , and the maker of SMR needs to make practical and empirical test
to prove to me that their non-suitability status have changed.

my previous employer Fujitu, always used Toshiba SAS CMR HDD's in SPARC servers , and most drives ( 90 %+ ) lasted more than 10 years 24/7/365 usage.
 
For a small number of home users with intermittent NAS usage, perhaps consider a 'wake on lan' solution. Then you can use pretty much any hardware and the power draw won't be quite so much of an issue.

Though in some instances, suspending the machine can put more stress on it than leaving it running, so factor that in when choosing a standby duration.
 
SMR drives are GREAT, when used correctly. They give the user more capacity for fewer $/GB. They are not so good when used wrongly, since they can have lower performance, in some usage patterns MUCH lower.

I have one SMR drive at home, being used with ZFS, as my backup device. It is not in a redundant configuration, so resilvering doesn't apply. It has few updates applied to it, since in my backup system, files are hardly ever deleted and never modified.

The big psychological / sociological problem with SMR drives is that people bought them and used them wrong, and then were surprised by the bad performance. It doesn't help that neither the manufacturers nor the channel explained to consumers how they are different from the old-fashioned CMR drives. Nor that some file systems and some usage patterns work great on them, others horrible, and usually in between.

It also doesn't help that the file systems used in consumer-grade free software were slow to be updated to take advantage of SMR and work efficiently with it. Ext4 is doing reasonably well (thank you Ted!), although I haven't used it for my own personal use, so I can't report performance experience on a typical home server workload. From what I know about ZFS, using it on SMR drives for a typical small file / development style workload might be a bad idea, but don't take my word for that for lack of testing.
 
ralphbsz said:
SMR drives are GREAT, when used correctly.

The big psychological / sociological problem with SMR drives is that people bought them and used them wrong,
Click to expand...
So for my edification, in relation to SMR drives what is "used correctly" vs "used wrongly".
I use ZFS typically mirror pairs, default block/stripe sizes, home use so mix of development (so smaller text files), photo editing, some streaming media.
I have trouble correlating that to "good/bad" on different devices.
 
Look similar thread here:

H

Other Thread 'How would you build your own NAS?'

FreeBSD, as any unix, offer a lot of possibilities for building something like a NAS,
special in FreeBSD is ZFS with its zpool scrub.

You can boot with pxeboot, mount with nfs, or send files with rsync, etc.

But the biggest problem I always have is mechanical.

You need to build a system with two discs for the mirror and a third for the OS (if no pxeboot). A big PC case?

Commercial NAS systems have compact cases, but their CPU does not allow to install the OS you want. And they are expensive.

I would be happy to have a "2 bay docking station" with power supply and 2 esata...
 
mer said:
So for my edification, in relation to SMR drives what is "used correctly" vs "used wrongly".
I use ZFS typically mirror pairs, default block/stripe sizes, home use so mix of development (so smaller text files), photo editing, some streaming media.
I have trouble correlating that to "good/bad" on different devices.
Click to expand...
Incorrectly: If you know that you are using SMR disks, and you know that your file system is not SMR aware (or not highly modified for using SMR), then don't quickly modify and delete small files. Correctly: Write large sequential files (where "large" is with respect to the 256 MiB SMR zone size). Now this is clearly not an option for most people, since their file system layout and read/write/delete ratios and file sizes can't be easily changed.

Correctly: Use a file system that is knowledgeable about the SMR zones. Within each zone, it only appends data to the end of the written part of the zone, doesn't overwrite existing data. It is somewhat knowledgeable about how many zones each drive can keep "open" at a time. All these parameters can be easily determined by asking the drive.

A little more advanced would be to estimate the expected workload (what fraction is sequential writes, what fraction is random writes, that is after they have gone through the file system mechanism), and format hybrid SMR drives accordingly, tuning the relative size of CMR and SMR portions correctly.

Seriously advanced would be go to host-managed SMR. That is probably outside the realm of open source general purpose file systems. I don't even remember whether host-managed SMR is still a widely used thing, or whether it was a dream.

Note that this is highly file system internal; it's not something a user can accomplish by tweaking a little bit of tuning parameters. Ted (of ext4 fame) has done a lot of work on this already; his publications can be found easily. And also note that we've known about SMR for the last 15 years.
 
  • Like
Reactions: mer
ralphbsz said:
The big psychological / sociological problem with SMR drives is that people bought them and used them wrong, and then were surprised by the bad performance. It doesn't help that neither the manufacturers nor the channel explained to consumers how they are different from the old-fashioned CMR drives. Nor that some file systems and some usage patterns work great on them, others horrible, and usually in between.
Click to expand...
My recollection is somewhat less sympathetic to the manufacturers, especially Western Digital, who had always sold Reds as NAS drives. They switched Reds from CMR to SMR in certain common capacities and failed to mention the change. I viewed that as active deceit by WD, not customers choosing to "use them wrong". WD later back-pedaled and brought back the CMR drives (beside the SMR drives in the product line) with clear labels identifying the products, and appropriate technical descriptions. But they only did that after after some "extremely negative" feedback from the customer base.
 
Agree, WD keeping the "brand name" Red while changing the underlying technology was quite dumb of them. So dumb that they got sued and lost. I hope they learned a lesson.

There is an underlying problem, which is the extreme clustering in the disk drive market. A few years ago I learned that over 90% of all non-laptop non-SSD disk drives (meaning the 3.5" disk drives) in the world are sold to 8 end-user customers. The largest well-known company among those customers bought half of the world's annual production of 3.5" disk drive on a single day a few years ago. Note that these 8 companies are not resellers, they are users of these disk drives. Their identities are obvious, the FAANG and friends. Armed with this information, I wonder why Seagate and WD even bother selling disk drives into the "open market", meaning to the small server manufacturers (HP, Dell, Lenovo...) and to the end-user channel (Newegg, CDW, ...). And I think the answer is that the profit margin in those channels is much higher than selling to the huge FAANG customers who buy disks millions at a time.
 
You must log in or register to reply here.
Back
Top