Hi, just a curious post about some things that were suggested with regards to performance here and there.
In an old blogpost: Performance considerations for larger UrBackup server instances | UrBackup Developer Blog There it says one of the considerations is to “Save the filesystem metadata of your backup storage on an SSD and only the actual data on a spinning disk RAID-5/6 to get the maximum performance. This is only possible with btrfs on Linux.”
I did find this: https://wiki.tnonline.net/w/Btrfs/Allocator_Hints#Benefits_of_Allocator_Hints, but it seems to me such a feature isnt part of standard btrfs implementation, is not stable and only for experimental use. (Btw that blog is an awesome resource for btrfs enthusiasts.)
Is that what the blog implied or is there another “trick” to have btrfs “prefer” certain members of the array for metadata storage (only)? I would really love to know. Because i think it makes a lot of sense for urbackup, considering the nature of the mechanism with its intensive metadata operations.
And the other thing is the manual suggests setting up the database for less concurrency. UrBackup - Server administration manual : “UrBackup’s database is in a mode which enables high concurrency. Since the cleanup procedure can sometimes be bottlenecked by the database it may be advisable to switch the database into a mode which allows less concurrency but is fast for some operations for the cleanup procedure.”
Now i dont think my setup will benefit much from such a change, because its mainly (metadata) IO bottlenecked, but i cant help to notice the manual does not give any pointers on how to do what it suggested. A quick search suggests “turning off WAL (Write Ahead Logging)” and “setting synchronous mode to FULL”, i wonder if those are indeed the things it wanted to imply there?