Pool, vdev, datasets, sharing

[Fastline]

Hello,

I've got a few questions regarding my setup

1. Can i use single pool for my multi-purpose needs or i need to have different pools for different needs? This will be like 16x16TB and two clients having one normal archival storage shared between both clients, two different time machine setups and a rendering storage shared with one client. What is a safer and reliable practice here?

2. How much vdevs should i add in a single pool to maintain the optimal performance of the pool?

3. Let's say i have three different needs (Normal Storage, Time Machine and Rendering). Do i need to create 3 different datasets and 3 different sharing or it will be one single dataset but 3 different sharing?

Thanks

 

[Fastline]

Bump!

 

[Patrick M. Hausen]

One share per dataset and vice versa. The question of one pool or more of them is not so easily answered. There is no "optimal performance" - it's always:

Pick any two of:

- fast
- reliable
- cheap

I would set up a single pool. All your use cases are quite similar. You don't have a requirement for "as fast as it can possibly go" anywhere. For private or small company setups people frequently compromise on "fast", because nobody wants "not reliable", right?

So 2x 8 disk wide RAIDZ2. 192 TB of usable space.

I would not consider RAIDZ1 with drives that large and also for your use case not consider mirrors. Economically RAIDZ wins over mirrors (of course) at the expense of performance. But you are not planning storage for virtualisation so we can give performance some slack to get more capacity for the money.

If you want to be even more conservative - and with 8 disk wide RAIDZ you can afford that - 2x RAIDZ3. 160 TB net. But your data is as safe as ZFS can probably deliver.

HTH,
Patrick

 

[Etorix]

1. "Different needs" would be: File storage vs. block storage (VM, iSCSI). Your requirements appear to be all file storage, so one pool may fit all.

2. That depends on your performance requirements. More IOPS = more vdevs.
With 16 TB drives, one would recommend raidz2 (or 3-way mirrors), and raidz# vdev should generally not be larger than 10-12. So, 2 * 8-wide raidz2 looks like a reasonable design.

3. TimeMachine probably wants its own dataset, with suitable properties and quotas.
Normal and Rendering may share a dataset, or may use different datasets if different properties may be useful (e.g. larger recordsize for larger files in Rendering), if you want to have different backup policies (snapshots) and/or be able to wipe Rendering clean.