[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20241007175250.GP21853@frogsfrogsfrogs>
Date: Mon, 7 Oct 2024 10:52:50 -0700
From: "Darrick J. Wong" <djwong@...nel.org>
To: Jan Kara <jack@...e.cz>
Cc: Tang Yizhou <yizhou.tang@...pee.com>, hch@...radead.org,
willy@...radead.org, akpm@...ux-foundation.org,
chandan.babu@...cle.com, linux-kernel@...r.kernel.org,
linux-fsdevel@...r.kernel.org, linux-xfs@...r.kernel.org
Subject: Re: [PATCH v2 3/3] xfs: Let the max iomap length be consistent with
the writeback code
On Mon, Oct 07, 2024 at 06:36:09PM +0200, Jan Kara wrote:
> On Sun 06-10-24 23:28:49, Tang Yizhou wrote:
> > From: Tang Yizhou <yizhou.tang@...pee.com>
> >
> > Since commit 1a12d8bd7b29 ("writeback: scale IO chunk size up to half
> > device bandwidth"), macro MAX_WRITEBACK_PAGES has been removed from the
> > writeback path. Therefore, the MAX_WRITEBACK_PAGES comments in
> > xfs_direct_write_iomap_begin() and xfs_buffered_write_iomap_begin() appear
> > outdated.
> >
> > In addition, Christoph mentioned that the xfs iomap process should be
> > similar to writeback, so xfs_max_map_length() was written following the
> > logic of writeback_chunk_size().
>
> Well, I'd defer to XFS maintainers here but at least to me it does not make
> a huge amount of sense to scale mapping size with the device writeback
> throughput. E.g. if the device writeback throughput is low, it does not
> mean that it is good to perform current write(2) in small chunks...
Yeah, I was wondering if it still makes sense to throttle incoming
writes given that iomap will just call back for more mappings anyway.
--D
> Honza
>
> --
> Jan Kara <jack@...e.com>
> SUSE Labs, CR
>
Powered by blists - more mailing lists