On 29/01/25 04:42PM, Christoph Hellwig wrote:
I don't think per-cpu is the right shard here. You want to write
related data together. A fіrst approximation might be inodes.
We have plan to affine inode to a particular writeback thread, like on
first encounter of inode we select one writeback context, and use the same
context when inode tries to issue IO again.
The number of writeback contexts can be changed from per-cpu and can be
decided using other factors as in this discussion.
https://lore.kernel.org/all/20250131093209.6luwm4ny5kj34jqc@green245/
