Relationship between Cache, Storage Density, Rotational Speed and PMR / SMR

lumdol

Member
Jun 9, 2016
30
2
8
37
Seems the SMR drives have inflated cache to compensate for lower performance.

The SMR drives however have really appealing densities and form factors (4tb in a 2.5” package, that’d be 96tb in a 2U 24 bay Supermicro chassis).

So, first question: are they really so bad? I’m guessing the response will be yes for anything looking like an active workload.... but maybe in a big zfs pool not so much?

Are most SATA drives with greater than 4tb moving to the SMR technology?

Does this technology have the capacity to improve much in terms of performance parameters?

Anyone with experiences with the 4tb 2.5” Seagate in an array approaching or exceeding 24 disks with ssd support?

(Any data in a configuration like this would be much appreciated).

What about the relationship between a PMR drive with less cache and less rotational speed?

Or, many PMR drives with less cache equalling the same density with more total cache?

I.e. 32mb 1tb x 8 = 8tb with 256mb of cache, versus one 8tb drive with 256mb cache?

Is there a general (however general) rule of thumb any of you follow to calculate performance based upon this criteria?

How do you all evaluate the parameters for the best disk for your needs?

How would you do so if available space were not yet a consideration (beginning a build, no chassis or form factor selected)?
 

thulle

New Member
Apr 11, 2019
22
14
3
I got 6x Seagate Archive 8TB in a RAID-Z. Works fine for media storage at 1MB recordsize, but large send/recv backups fills the write cache and then I start seeing multi second latencies.
When OpenZFS 0.8 releases with metadata vdevs and ability to allocate small blocks on that vdev I suspect that the performance would be better, but until then I'd recommend thinking through the use case.