We bought a new server from HPE (Gen11) which is equiped with 12 nvme 1.92 TB Disks. Now I am asking myself wether to configure it for maximum diskspace - like 11x in a raidz5 plus 1 hostspare vs. IO-performance - something like 5striped vdevs à 2 disks in mirror config plus 2 hotspares. So while evaluating the need of the business (space vs. performance of guests) I thought,
I should ask here as its a once in a longtime oportunity to do some tests to see the difference of the prerformance to check if its worth to config it like this. On the other hand we are talking about nvme's which are rather fast, aren't they so I tought maybe a big vdev in raidz5 might be quick enough in terms of IO performance to run lets say 10 to 15 VM guests simulatniously.
As I read the IO performance of one vdev is approximately at the same speed as one nvme disk, it might be best creating something like 2 striped vdevs with 5 disks plus one hotspare in raidz5 config or three striped raiz5 vdevs containing 4 nvme's each? That way I would loose less diskspace but have 2 or 3 vdevs so approximatelly the speed of 2-3 nvme disks?
If I would like to go through the hassle of testing it, how would such a test have to look like?
I should ask here as its a once in a longtime oportunity to do some tests to see the difference of the prerformance to check if its worth to config it like this. On the other hand we are talking about nvme's which are rather fast, aren't they so I tought maybe a big vdev in raidz5 might be quick enough in terms of IO performance to run lets say 10 to 15 VM guests simulatniously.
As I read the IO performance of one vdev is approximately at the same speed as one nvme disk, it might be best creating something like 2 striped vdevs with 5 disks plus one hotspare in raidz5 config or three striped raiz5 vdevs containing 4 nvme's each? That way I would loose less diskspace but have 2 or 3 vdevs so approximatelly the speed of 2-3 nvme disks?
If I would like to go through the hassle of testing it, how would such a test have to look like?
Last edited: