Re: [PATCH 0/2] btrfs: reduce extent map lookup overhead for data write

[David Sterba <dsterba@suse.cz>]

On Thu, Aug 08, 2024 at 03:35:58PM +0930, Qu Wenruo wrote:
> Unlike data read path, which use cached extent map to reduce overhead,
> data write path always do the extent map lookup no matter what.
> 
> So this patchset will improve the situation by:
> 
> - Move em_cached into bio_ctrl
>   Since the lifespan of them is the same, it's a perfect match.
> 
> - Make data write path to use bio_ctrl::em_cached
> 
> Unfortunately since my last relocation, I no longer have any dedicated
> storage attached to my VMs (my laptop only has one NVME slot, and my main
> workhorse aarch64 board only has one NVME attached either).
> 
> So no benchmark yet, and any extra benchmark would be very appreciated.
> 
> Qu Wenruo (2):
>   btrfs: introduce extent_map::em_cached member
>   btrfs: utilize cached extent map for data writeback

This looks like a good optimization, we're approaching code freeze (next
week) so it would be good to get it merged by then. As it's an
optimization we can also postpone it if you're busy with other things.