Question re ZFS with mixed drive speeds & types

David Christensen dpchrist at holgerdanske.com
Wed Aug 5 08:32:13 UTC 2020


On 2020-08-04 23:55, Greg Marsh wrote:
> Hello everyone. I hope you are well.
> 
> I’m budgeting to upgrading my home file server. Sort of for more data capacity, but mostly for more physical space, flexibility and to try some things storage wise. It’s currently got a 6x2tb SATA raidz of WD reds w/ssd for l2arc & logging.
> 1st goal - mirror that ssd.
> 
> Now, I’ve got a big box of still working Seagate Cheetah 15k scsi drives & several controllers, both raid and not. The drives are only 146gb, but 15k. However, I’m wondering if I were to make a VDEV pool of like 7 of them, add that to my volume and what, if any positive benefit it might to my ZFS storage?
> 
> I ask because about 10 years ago, I was involved in a project that had a HUGE storage component. Many of the vendors we brought in to pitch, including Sun, were promoting tiered storage. Sun in particular were quite proud of this tech. Their system ran ZFS and had a hybrid of ssd, sas & SATA, with the system dynamically moving data around the different speed/capacity drives, based on their activity, all transparent to the application or user. Most often used data coming from the ssd & sas drives, with less active files kept on the SATA drives.
> 
> I’m wondering if this capability made it into the open source version of ZFS. Because that would be fantastic.
> 
> Thank you all for any insight or advice.
> 
> Cheers,
> Greg


Everything I have read recommends using the same size and speed devices 
in a given pool.


To upgrade a pool, one technique is to replace one device at a time with 
a "better" device, wait for the new device to resilver, and repeat until 
all devices have been upgraded.  Beware that if you have raidz1, you 
would be at risk of losing everything if a drive failed while upgrading 
another.


Using the SSD for cache and using two or more SCSI drives in a mirror 
for log makes sense if you want to minimize hardware cost.


David


More information about the freebsd-questions mailing list