Merge some ZFS-on-SSD performance optimizations from FreeBSD
Sequential scrub increased I/O aggregation size in ZFS from 128KB to 1MB. It makes sense for HDDs, but for SSDs it is mostly a waste of time on additional memory copy operation to aggregate memory chunk, which GEOM will still split into 128KB pieces. Proposed patch introduces new limit of 128KB for non-rotating VDEVs and allows it to be tuned even lower later, if we see it beneficial.
Make vfs.zfs.metaslab_lba_weighting_enabled to be automatically ignored for SSD-based vdevs, since it has no sense there.
#3 Updated by Alexander Motin over 2 years ago
- Status changed from In Progress to Ready for Testing
PR for 11.2-stable: https://github.com/freenas/os/pull/189
QE: There should be no externally visible changes. Internally, profiler should show disappearance of one big memcpy() when writing to zvol/dataet with block size >= 128K on SSD-backed pool. I hope nap@ to test it.