Received-SPF: pass (google.com: domain of linux-kernel+bounces-182910-linux.lists.archive=gmail.com@vger.kernel.org designates 139.178.88.99 as permitted sender) client-ip=139.178.88.99;
Precedence: bulk
MIME-Version: 1.0
References: <20240517-amtime-v1-1-7b804ca4be8f@kernel.org> <Zkg7OCSYJ7rzv6_D@casper.infradead.org>
In-Reply-To: <Zkg7OCSYJ7rzv6_D@casper.infradead.org>
From: Linus Torvalds <torvalds@linux-foundation.org>
Date: Sat, 18 May 2024 09:03:45 -0700
Message-ID: <CAHk-=wjgUoZE3jq7Ynm=LtavYK6yZMboogwWtm1fGT1yqh7NoQ@mail.gmail.com>
Subject: Re: [PATCH] fs: switch timespec64 fields in inode to discrete integers
To: Matthew Wilcox <willy@infradead.org>
Cc: Jeff Layton <jlayton@kernel.org>, Christian Brauner <brauner@kernel.org>, 
	Alexander Viro <viro@zeniv.linux.org.uk>, Jan Kara <jack@suse.cz>, linux-fsdevel@vger.kernel.org, 
	Amir Goldstein <amir73il@gmail.com>, linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"

On Fri, 17 May 2024 at 22:23, Matthew Wilcox <willy@infradead.org> wrote:
>
> Smaller is always better, but for a meaningful improvement, we'd want
> to see more.

I think one of the more interesting metrics for inodes is actually not
necessarily size per se, but cache footprint.

A *lot* of the inode is never actually touched in normal operation.
Inodes have all these fields that are only used for certain types, or
perhaps only for IO.

So inodes are big, but more important than shrinking them is likely to
try to make them dense in the cache for normal operations (ie
open/close/stat in particular). They cache very well, and actual
memory use - while still somewhat relevant - is less relevant than
cache misses.

             Linus