Making the most of SMR drives

[Zandr Milewski]

I recently built a home backup/media server using Seagate 8TB SMR drives, and have been pondering how to tweak FreeNAS to make the most of these drives, given their unique write characteristics. I'll describe what I've done so far, but everything is still in testing, so if nuking the drives and rebuilding the system is in order, that's fine.

The box has 6 8TB drives in a RAIDZ2, 16GiB of RAM (maxed out), and a pair of 120GB mSATA SSDs. It's booting FreeNAS 9.3.1 from a 16GB USB drive.

Workload is backup/archive and media server. Clients are almost exclusively Macs. Once it sees "production", it will be on a connected UPS.

I've already moved the system dataset to the SSDs (configured as a mirror). I'm not sure if there's any point in booting from the SSDs; seems like the boot volume should be pretty much read-only.

I've turned off atime on the SMR volume.

It seems like I should get swap off of the SMR drives, and on to the SSDs. If I understand correctly, the write cache is limited to a fraction of real memory, not memory+swap, so using lots of the SSD for swap won't help any.

Is there a good way to measure what portion of my workload is sync, to determine whether putting an SLOG on SSD would be worthwhile? (or have I misunderstood that thread?)

Thanks,
-Zandr

 

[anodos]

You can use zilstat to see if you'd benefit from a slog. My guess is probably not. Don't mess with swap or l2arc. You want something with more RAM. IOPs performance will probably suck donkey nuts.

 

[Robert Trevellyan]

It seems like I should get swap off of the SMR drives, and on to the SSDs.

I think it's more a case of "add RAM until swap usage goes to zero".

 

[jgreco]

You don't want to mess with the swap space. You do want to make sure your system is not actively swapping as part of normal operation. The swap space serves as a safety valve, a way for the system to cope with unexpected memory pressure, while also allowing for replacement drives of slightly different sizes.

 

[Zandr Milewski]

As I said, RAM is maxed out. For this box, that's 16GiB. Not ideal, but there it is.

Having said that, there's no memory pressure as of now. I'm pretty sure I don't want to start swapping to the SMR drives if things go sideways, though.

zilstat - thanks, that's what I was after.

 

[L]

In the last openzfs office hours I think I remembered shingled disks being discussed and at the time they are not currently recommended for use with zfs. I don't have time to review, but you might want to watch the replay. http://open-zfs.org/wiki/OpenZFS_Office_Hours

 

[jgreco]

In the last openzfs office hours I think I remembered shingled disks being discussed and at the time they are not currently recommended for use with zfs. I don't have time to review, but you might want to watch the replay. http://open-zfs.org/wiki/OpenZFS_Office_Hours

Yeah, but we're discussing a system that's already built.

 

[cyberjock]

I've worked with 3 shingled systems. ZFS works fine. There are performance limitations that start to rear their head if you are running 10Gb LAN and transferring massive amounts of data (think 100s of GB+) at nearly saturation speed.

So I don't agree that they aren't "recommended". I certainly wouldn't recommend them for VMs. But for backups (or data that is rarely written to or at least not in massive quantities) they are very good options and work very well. In fact, I'm helping my best friend expand his system, and he's going with 8TB shingled drives. :P

 

[diskdiddler]

I got the impression these things are hugely bad due to how they operate? Constant excessive read / writes?
I'm really hoping to see some disks exceeding 6TB which don't use SMR, don't use helium. To my knowledge, 6TB is the largest available "normal" hard disk.
There are articles implying in the next 3 or 4 years big, BIG SSD's will become surprisingly cheap but right now, it's still platter disks

 

[Bidule0hm]

Just make 5.25" platters instead of 3.5" and bingo, 30-40 TB drives are available with the same surface data density :D

 

[diskdiddler]

My FreeNAS box is already big, hot and noisy enough :)

 

[cyberjock]

Just make 5.25" platters instead of 3.5" and bingo, 30-40 TB drives are available with the same surface data density :D

I was actually looking into this. Turns out it's been like 10 years since the last 5.25" platter manufacturing plant in the world shutdown. Someone is trying to get money to build one again for high density 5.25" disks for archival purposes, but no clue on if/when it will happen. I can bet that if suddenly 50TB archival disks came out that were 5.25", someone somewhere would be ready to buy them as fast as they are manufactured.

 

[Bidule0hm]

That's very interesting, thanks for the info ;)

 

[jgreco]

Just make 5.25" platters instead of 3.5" and bingo, 30-40 TB drives are available with the same surface data density :D

Oh, yay, the return of the Quantum BigFoot.