Mirror & stripe 2 disks at the same time?

[Dariusz1989]

Hey

I'm going over this guis blog https://pthree.org/2012/12/07/zfs-administration-part-iv-the-adjustable-replacement-cache/
Pretty great read and I was wondering, at the end of this post he shows that he has a 2 OCZ ssds in both mirror and stripe? - or do I read it incorrectly?

pool 824G 2.82T 11 60 862K 1.05M
raidz1 824G 2.82T 11 52 862K 972K
sdd - - 5 29 289K 329K
sde - - 5 28 289K 327K
sdf - - 5 29 289K 329K
sdg - - 7 35 289K 326K
logs - - - - - -
mirror 1.38M 3.72G 0 19 0 277K
ata-OCZ-REVODRIVE_OCZ-69ZO5475MT43KNTU-part1 - - 0 19 0 277K
ata-OCZ-REVODRIVE_OCZ-9724MG8BII8G3255-part1 - - 0 19 0 277K
cache - - - - - -
ata-OCZ-REVODRIVE_OCZ-69ZO5475MT43KNTU-part2 2.34G 49.8G 0 0 739 4.32K
ata-OCZ-REVODRIVE_OCZ-9724MG8BII8G3255-part2 2.23G 49.9G 0 0 801 4.11K


How can I make it ? I have a freeNas on Oracle VM VirtualBox and just messing around with commands trying to learn how to do stuff.

His idea of having mirror & stripe on the same 2 hdds feels quite interesting and I want to learn/see how its made. But I'm not sure if I understand it correctly nor how he did it?

Can any1 have a look or correct me?

I'm trying to do some tests to see how to create slog/l2arc on the same 2 disks + configure/optymize and how to deal with ram needs. As its virtual box I think its good testing environment for now.

Regards
Dariusz

 

[Dariusz1989]

Looks like it is possible ! :O
So here are some helpfull links>
https://www.penguinpunk.net/blog/freenas-using-one-ssd-for-zil-and-l2arc/
https://blog.nathancoad.com/item/freenas-and-ssd-l2arc/
https://clinta.github.io/freenas-multipurpose-ssd/

Good luck coding! :- )

 

[Ericloewe]

That is a horrible idea. The workloads are completely different and unaware of each other.

This more or less works with ultra-high-end SSDs and is pure pain with consumer SSDs.

 

[Dariusz1989]

That is a horrible idea. The workloads are completely different and unaware of each other.

This more or less works with ultra-high-end SSDs and is pure pain with consumer SSDs.

Hey

I'm still reading on the slog/l2arc. The blog post in original post is from 2012 and hes using a OCZ driver... which if memory serves me right were running at something in the range 400 mb/rw. Comparing to current m.2 3/2gb rw were quite a bit ahead?

I mean it all depends on workload no?

On a side note I read some old posts where sLog/l2arc could damage main pool if removed improperly - is that still the case? I was able to run a test on my vm Freenas with remove poolName mirror-1 (slog) and remove poolName gptid/driveID of l2arc so it looks like it's possible to add and then remove the sLog/l2arc if performance suffers? - is that correct with FreeNas 11.3?

Thanks for info.

 

[danb35]

What he's done is partitioned two SSDs. He's used the first partitions of both, mirrored, for SLOG. He's used the second partitions of both, striped, for L2ARC. It's a bad idea, as @Ericloewe says, but yes, it's possible. If you manually partition your disks, you could even do something crazy like that for your data disks--but you shouldn't.

 

[Stux]

Thing about using the same disks for SLOG and L2ARC is the read/write traffic literally happens at the same time, and Sata disks just don't have the IOP capacity for that.

If you had a high performance NVMe drive, then you could perhaps get away with it...

https://forums.freenas.org/index.ph...n4f-esxi-freenas-aio.57116/page-4#post-403374

Of course, it'd work with Sata, but the issue is that the Sata slog is already getting you only a fraction of the performance that you'd like and any additional loss to that (any using l2arc) is undesired.

Also, you mirror the slog so that if the drive fails on unexpected restart you don't lose any data. And the l2arc is striped for performance/capacity reasons.

And finally, there's no point using a crappy non-PLP drive for slog.

 

[Pezo]

Thing about using the same disks for SLOG and L2ARC is the read/write traffic literally happens at the same time

Why is that? When there are only writes, why would the L2ARC be busy, and vice versa?

 

[Ericloewe]

If you have a write-only workload, what's the point of the L2ARC? If you have a read-only workload, what's the point of the SLOG?

 

[Stux]

Why is that? When there are only writes, why would the L2ARC be busy, and vice versa?

Writes get buffered in ARC. ARC gets shuffled to L2ARC continuously at 5MB/s as it fills