ok, ashift values are ok and not different within datapool and at 12.
(can happen if you add another vdev to a pool)
The ashift=0 values are not clear to me but seems uncritical as they do not affect data or disks
Performance:
- I asume you have enabled sync on the SSD pool. This will result in a poor but secure write performance. To validate, disable sync and retry a write. (On a VM storage you should enable sync or your guest filesystem can become corrupted on a crash during write)
To benchmark pool performance use menu Pool > Benchmark.
This is a series of large and small read/write io with sync enabled vs sync disabled
Check:
Is your vnic a vmxnet3 or e1000. First is faster
Have you run System > Appliance tuning with defaults?
This will improve NFS and ip performance between storage and ESXi
An SSD pool with an additional Slog is "suboptimal" as this has the effect that every write musz be done twice, once fast via rambased writecache and once on every write commit. As I see your pool SSDs are desktop ones and only the log is the enterprise edition with powerloss protection. In such a case the slog improves write security a little but a small risk remains as the pool itself has no powerloss protection so no guarantee that last writes are save on a crash.
Best regarding security and performance would be using a mirror of SSDs with powerloss protection without an extra Slog and sync enabled.
(can happen if you add another vdev to a pool)
The ashift=0 values are not clear to me but seems uncritical as they do not affect data or disks
Performance:
- I asume you have enabled sync on the SSD pool. This will result in a poor but secure write performance. To validate, disable sync and retry a write. (On a VM storage you should enable sync or your guest filesystem can become corrupted on a crash during write)
To benchmark pool performance use menu Pool > Benchmark.
This is a series of large and small read/write io with sync enabled vs sync disabled
Check:
Is your vnic a vmxnet3 or e1000. First is faster
Have you run System > Appliance tuning with defaults?
This will improve NFS and ip performance between storage and ESXi
An SSD pool with an additional Slog is "suboptimal" as this has the effect that every write musz be done twice, once fast via rambased writecache and once on every write commit. As I see your pool SSDs are desktop ones and only the log is the enterprise edition with powerloss protection. In such a case the slog improves write security a little but a small risk remains as the pool itself has no powerloss protection so no guarantee that last writes are save on a crash.
Best regarding security and performance would be using a mirror of SSDs with powerloss protection without an extra Slog and sync enabled.
Last edited: