How Does TN Fill Drives?

[Zain]

I am in the process of upgrading drives to larger disks in an R720. Once they are all swapped, I will install the old drives in an MD1200 so that when the additional storage is needed again I can just boot that up as well.

My question here is, how does SCALE fill up the drives? I added drives as storage was needed, so as you can see below, the data is not balanced across each drive. Yes, I know, I shouldn't worry about the data being balanced as it will automatically hand itself with time - I'm not worried about that. I am more just curious how the data is written to each of the VDEVs when the data isn't balanced, and there are different size drives in the pool.

Is data written to drives based on the percentage of the capacity of each drive/vdev? Or, is the data written so that there is equal bytes written across each vdev?

Thanks.

 

[sretalla]

I shouldn't worry about the data being balanced as it will automatically hand itself with time

Not necessarily... maybe it will if you write enough and wait long enough.

You can use something like this to help it along though...

Re-balance striped vdevs?

I have a zpool with a single 15x wide RAIDz2 vdev. I filled the pool with data and it is now almost full. I am thinking about adding a second 15x wide RAIDz2 vdev to double the capacity of the zpool. However, I am unclear how new data written to the zpool will be handled. Will it be evenly...

www.truenas.com

Or just copying the data off and back on again manually will work too

Is data written to drives based on the percentage of the capacity of each drive/vdev?

No. ZFS is looking to put transactions together in transaction groups which are contiguous on the disk(s), so depending on the size of the requested writes and the available unfragmented free space to put them, many different outcomes can occur (and you shouldn't spend too much time trying to predict it unless you want to become super-familiar with the deep code... it's on github if you want to do that).

is the data written so that there is equal bytes written across each vdev?

All things being equal (having started the pool with all VDEVs present), this is the most likely outcome and also the most desirable for IOPS.

 

[Zain]

Not necessarily... maybe it will if you write enough and wait long enough.

You can use something like this to help it along though...

Re-balance striped vdevs?

I have a zpool with a single 15x wide RAIDz2 vdev. I filled the pool with data and it is now almost full. I am thinking about adding a second 15x wide RAIDz2 vdev to double the capacity of the zpool. However, I am unclear how new data written to the zpool will be handled. Will it be evenly...

www.truenas.com

Or just copying the data off and back on again manually will work too


No. ZFS is looking to put transactions together in transaction groups which are contiguous on the disk(s), so depending on the size of the requested writes and the available unfragmented free space to put them, many different outcomes can occur (and you shouldn't spend too much time trying to predict it unless you want to become super-familiar with the deep code... it's on github if you want to do that).


All things being equal (having started the pool with all VDEVs present), this is the most likely outcome and also the most desirable for IOPS.

Again, I'm not concerned about balancing the data. I was just curious how the drives got filled up. I would expect drives to be filled up so that there are equal bytes spanned across each drive/vdev in the pool, and as you mentioned would be performance beneficial, but wasn't sure.