Re: [PATCH v4 4/6] xfs: generic AG set based steering

[Kanchan Joshi <joshi.k@samsung.com>]

On 8/19/2026 12:51 PM, Christoph Hellwig wrote:
> On Fri, Jul 17, 2026 at 06:25:36PM +0530, Kanchan Joshi wrote:
>> Improve allocator concurrency and reduce interleaving by introducing
>> fixed sized AG set.
>> Use low bits of the inode as a hash to select AG within the AG set.
>> Overall, a file will try to use the same AG (and contiguity is maintained),
>> but multiple files will be spread across all AGs in the target AG set.
> I'm a bit worried this kind of magic - the inodes close to each other
> might not all have data or very different amounts of data.
> 
> I suspect the root cause this helps with is that the XFS initial data
> placement algorithm that tries to place file data next to the inode is
> actively harmful these days.  So doing an initial AG selection similar
> to how the RT device does the initial RTG selection might be more
> useful here (and thinking about moving to an inode32-like scheme
> where small metadata with a lot of churn is constrained to a few
> AGs).

Thinking to drop this for now, as this may need a distinct series.

This patch was here because stream based AG-set extended what this patch 
was doing (i.e., the minimal fixed-size AG-set). That relationship goes 
with the change you mentioned.

Replacing default AG allocator with an automatic RG-style rotor (no 
user-supplied write-stream) might be a radical change in terms of impact.