[SOLVED] ZFS Wizard Wanted

[rmundel]

Hey guys, I was digging a little more into ZFS and got some questions:
- Isn't the IOPS for a HDD too high? (zpool iostat -v 1)


- With high I/O load the PVE web gui become unresponsive, there's a 'fix'? Maybe more ARC? (12x2TB disks in RAIDZ6 with ZIL and L2ARC ssd's limited to 24GB of ARC)
- In the same hardware with ashift=12 my data uses more than double of disk space (real data inside a VM = 7ish TB, ZFS is reporting 15ish TB). Reinstalling with ashift=9 I solved this 'issue'. My disks are 512b, with ashift=12 shouln't I have the same disk usage/block alignment?

Thanks!

 

[wolfgang]

Hi

- Isn't the IOPS for a HDD too high? (zpool iostat -v 1)

it depends if the io is sync or async. you can check this with the -r parameter.

With high I/O load the PVE web gui become unresponsive, there's a 'fix'? Maybe more ARC? (12x2TB disks in RAIDZ6 with ZIL and L2ARC ssd's limited to 24GB of ARC)

you can have a lock at the arc_summery.
There you get a good overview of a bottleneck in the arc.

In the same hardware with ashift=12 my data uses more than double of disk space (real data inside a VM = 7ish TB, ZFS is reporting 15ish TB). Reinstalling with ashift=9 I solved this 'issue'. My disks are 512b, with ashift=12 shouln't I have the same disk usage/block alignment?

If the vdev blocksize is smaller sure it is possible to save disk space. But space you pay with system speed.
With ZOL it is recommended to use 4k = ashit12. And if the system gets populated the space advantage gets lost.

 

[rmundel]

Thanks!

 

[LnxBil]

it depends if the io is sync or async. you can check this with the -r parameter.

and also if it's random (worst case) or sequential (best case). Often it is a mixture of both. IOPS values for disks of 150-200 is for random 4K IOPS, sequential IOPS are much, much higher.