I'm currently experimenting with different ways of improving write speeds to a fairly large, rotating disk-based, software-raid (mdadm) array on Debian using fast NVMe devices.
I found that using a pair of such devices (raid1, mirrored) to store the filesystem's journal yields interesting performance benefits. The mount options I am using to achieve this are noatime,journal_aync_commit,data=journal.
In my tests, I've also discovered that adding the barrier=0 option offers significant benefits in terms of write performance. However, I'm not certain that this option is safe to use in my particular filesystem configuration. This is what the kernel documentation says about ext4 write barriers:
Write barriers enforce proper on-disk ordering of journal commits, making volatile disk write caches safe to use, at some performance penalty. If your disks are battery-backed in one way or another, disabling barriers may safely improve performance.
The specific NVMe device I'm using is an Intel DC P3700 which has built-in power-loss protection which means that in the event of an unexpected shutdown, any data still present in temporary buffers is safely committed to NAND storage thanks to reserve energy storage.
So my question is, can I safely disable ext4 write barriers if the journal is stored on a device with battery-backed cache, while the rest of the filesystem itself sits on disks which don't have this feature?