Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 23 Jan 2023 11:49:42 -0500
From:      Paul Mather <paul@gromit.dlib.vt.edu>
To:        Peter <pmc@citylink.dinoex.sub.org>
Cc:        Mateusz Guzik <mjguzik@gmail.com>, freebsd-stable@freebsd.org
Subject:   Re: close() taking minutes (ZFS related)
Message-ID:  <C558EB67-330F-48AE-A251-E244089311CB@gromit.dlib.vt.edu>
In-Reply-To: <Y85nY7AgWMvhWruF@disp.intra.daemon.contact>
References:  <Y8oUemqHiPilKCK7@disp.intra.daemon.contact> <CAGudoHE4Sx_KY1vAsF1r8zXW6Y3vA4odRj6=bfm0fmLXFcVN9g@mail.gmail.com> <Y8qknwgR4quVXxay@disp.intra.daemon.contact> <CAGudoHH_ZP9%2BdZHb=6y10urYcYCwjdE1qOS2T6n4ennOSzVJ_A@mail.gmail.com> <Y85nY7AgWMvhWruF@disp.intra.daemon.contact>

next in thread | previous in thread | raw e-mail | index | archive | help
On Jan 23, 2023, at 5:54 AM, Peter <pmc@citylink.dinoex.sub.org> wrote:

> It seems I have found the reason for the strange behaviour: it is
> dedup.
>=20
> The pool is configured to use dedup, because it carries all my jail OS
> installation, which are very similar. But dedup makes no sense on
> database filesets, and apparently it has very bad effects on real
> workload: it creates immense write-amplification (about factor 4)
> and probably other bad effects (memory is not the issue here).
>=20
> I don't think it should lockdown the entire pool, but it is
> understandable that it creates a logical relation between the
> different filesets.


Note that "dedup" is a fileset-level property and so if the database =
filesets are poor candidates for deduping you should be able to set =
dedup to "off" for them to mitigate their bad impact on the system.

Even though dedup is a fileset-level property, enabling dedup on any =
fileset can have major impact system-wide (due to large resource =
consumption) and so should be done with due diligence.

Cheers,

Paul.




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?C558EB67-330F-48AE-A251-E244089311CB>