[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <ZdjYBwNTaxbZHpo4@google.com>
Date: Fri, 23 Feb 2024 09:38:15 -0800
From: Jaegeuk Kim <jaegeuk@...nel.org>
To: Matias Bjørling <m@...rling.me>
Cc: Juhyung Park <qkrwngud825@...il.com>,
Niklas Cassel <niklas.cassel@....com>,
Damien Le Moal <damien.lemoal@....com>, daehojeong@...gle.com,
linux-kernel@...r.kernel.org,
linux-f2fs-devel@...ts.sourceforge.net,
Matias Bjørling <matias.bjorling@....com>
Subject: Re: [f2fs-dev] [PATCH 3/3 v2] f2fs: kill zone-capacity support
On 02/22, Matias Bjørling wrote:
> On 21-02-2024 18:27, Jaegeuk Kim wrote:
>
> > > Doesn't this break practically all ZNS NVMe devices?
> >
> > Yes, so here I'm in questioning who is really using w/ zone capacity. If there's
> > no user complaining, I'd like to deprecate this, since this adds code complexity
> > and unnecessary checks.
> >
>
> Hi Jaegeuk,
>
> I like to make a couple of points to hopefully keep the support in a little
> while longer.
>
> - NVMe-based zone devices continue to be developed with the pow2 zone size
> and zone size != zone cap features. There was some divergence in the
> first-gen drives. However, all the second-gen drives I know of are
> implemented with those features in mind.
>
> - A very active community is doing work using f2fs, and many of those
> members are working with the ZN540s device (which exposes a pow2 zone size).
>
> - For drives with a capacity of less than 16TiB, f2fs is an excellent file
> system to use and is really useful for various use cases. We're using the
> f2fs daily for a couple of our workloads.
>
> Work is ongoing on btrfs and XFS to support zoned storage devices, but they
> have yet to be through the trenches as much as f2fs has been with its zone
> support. So it would be great to have f2fs continue to support the pow2 zone
> sizes, as it is a valuable feature for the currently used and second-gen
> drives that have been released or are soon becoming available.
>
> If there is a performance concern with the feature re: ZUFS, maybe the pow2
> implementation could be slightly more computationally expensive, as the
> feature, anyway, typically is used on more beefy systems.
Thanks, Matias for the background. It seems to be fair for keeping this for a
while tho, IMO, non-pow2 could address both parties.
>
> Regards,
> Matias
Powered by blists - more mailing lists