SMR for backups / archives?

[Constantin]

Copying an 80GB VHD file goes lovely, for a while, eventually dropping to horrific speeds.

Sounds like a fast front end write cache with the slow flash in the rear. Works great until you have a sustained workload. Then, just as SMR, the performance hits a brick wall as the drive goes unresponsive while flushing the cache to the back end.

 

[diskdiddler]

Sounds like a fast front end write cache with the slow flash in the rear. Works great until you have a sustained workload. Then, just as SMR, the performance hits a brick wall as the drive goes unresponsive while flushing the cache to the back end.

That's exactly my thoughts, however.....

Like we're talking about a 6 disk array here!

6!

You'd think the worst possible write speed would be like 100MBs surely... Not this

 

[rvassar]

That's exactly my thoughts, however.....

Like we're talking about a 6 disk array here!

6!

You'd think the worst possible write speed would be like 100MBs surely... Not this

What kind of 6 disk array? If it's a single vdev RAIDz/z2, write speed is equal to a single device, as all drives have to complete the write to ensure data safety. Now if it's 6 drives in a 3vdev mirror... Yes, you should see 3x single device write.

 

[diskdiddler]

I

What kind of 6 disk array? If it's a single vdev RAIDz/z2, write speed is equal to a single device, as all drives have to complete the write to ensure data safety. Now if it's 6 drives in a 3vdev mirror... Yes, you should see 3x single device write.

Believe it's a Z2, with 2 redundant.

I'd have thought the data is at least 4x the speed of a single?

6 disk's, 2 redundant.

 

[rvassar]

I

Believe it's a Z2, with 2 redundant.

I'd have thought the data is at least 4x the speed of a single?

6 disk's, 2 redundant.

Nope. No write performance boost in RAIDz/z2 untill you have multiple vdev's. In order to remain 2 drive redundant, all the drives have to have all the data needed to reconstruct from parity. You do get a read boost, as you can read from 4 devices and reconstruct in memory faster than reading from all 6 drives. So ZFS will play games like that on reads and you can probably get almost 3x faster reads. But writes are held to the gold standard so that any two drives can fail and it's all still recoverable.

This why we recommend mirrors for VM datastores, etc... Each pair gets round-robin written, and on read each device gets a different read queued up to it. Downside is it wastes a bunch of disk space.

 

[Davvo]

I suggest a reading of the following resource.

ZFS Storage Pool Layout

This amazing document, created by iXsystems in February 2022 as a "White Paper", cleanly explains how to qualify pool performance touching briefly on how ZFS stores data and presents the advantages, performance and disadvantages of each pool...

www.truenas.com

 

[diskdiddler]

So is there a different way I can lay the disks out and increase performance, while still having 2 disk redundancy?

2x3 disk pools? In raid 0?

Means I can only lose 1 of each pool?

 

[ChrisRJ]

Didn't the resource provided by @Davvo help here?

 

[Davvo]

Two three-way mirror VDEVs would give you the ability to endure the loss of up to 2 drives in each VDEV.

This would also improve the IOPS performance of the pool.

It's all explained in the previously linked resource.