Re: [RFC][PATCH v2] fsnotify: optimize the case of no content event watchers

From: Jens Axboe <axboe@kernel.dk>
To: Amir Goldstein <amir73il@gmail.com>, Jan Kara <jack@suse.cz>
Cc: Christian Brauner <brauner@kernel.org>, linux-fsdevel@vger.kernel.org
Subject: Re: [RFC][PATCH v2] fsnotify: optimize the case of no content event watchers
Date: Mon, 15 Jan 2024 13:48:51 -0700	[thread overview]
Message-ID: <b45bd8ff-5654-4e67-90a6-aad5e6759e0b@kernel.dk> (raw)
In-Reply-To: <CAOQ4uxhLum65Nou=DRaAT6W5xTvWjjP4+5mxYv2K3j4PB89s1A@mail.gmail.com>

> Jens, if you have time for another round please test.
> My expectation would be to see something like
> 
> +1.46%  [kernel.vmlinux]  [k] __fsnotify_parent

Sure - so first I can current -git, which we know has that extra
fsnotify call as previously discussed. This is with SECURITY=y and
FANOTITY_ACCESS_PERMISSIONS=y for both.

Performance:

IOPS=65.01M, BW=31.74GiB/s, IOS/call=31/32
IOPS=65.42M, BW=31.94GiB/s, IOS/call=32/32
IOPS=65.02M, BW=31.75GiB/s, IOS/call=31/32
IOPS=65.37M, BW=31.92GiB/s, IOS/call=32/32
IOPS=65.44M, BW=31.95GiB/s, IOS/call=31/31
IOPS=64.91M, BW=31.70GiB/s, IOS/call=32/32
IOPS=64.25M, BW=31.37GiB/s, IOS/call=32/31

and profile:

+    4.51%  io_uring  [kernel.vmlinux]  [k] fsnotify
+    3.67%  io_uring  [kernel.vmlinux]  [k] __fsnotify_parent

With this patch applied, I see:

IOPS=73.09M, BW=35.69GiB/s, IOS/call=32/31
IOPS=72.94M, BW=35.61GiB/s, IOS/call=32/32
IOPS=72.95M, BW=35.62GiB/s, IOS/call=31/31
IOPS=72.54M, BW=35.42GiB/s, IOS/call=32/32
IOPS=73.01M, BW=35.65GiB/s, IOS/call=32/31
IOPS=73.07M, BW=35.68GiB/s, IOS/call=32/32

and profile:

+    2.37%  io_uring  [kernel.vmlinux]  [k] __fsnotify_parent

and perf diff shows the following top items:

     3.67%     -1.30%  [kernel.vmlinux]  [k] __fsnotify_parent
     4.51%             [kernel.vmlinux]  [k] fsnotify

Let me know if this is what you wanted, or if the base should be with
that extra fsnotify removed.

-- 
Jens Axboe