Date: Sun, 3 Jul 2016 15:24:56 -0700 From: Jordan Hubbard <jkh@ixsystems.com> To: Julien Cigar <julien@perdition.city> Cc: Ben RUBSON <ben.rubson@gmail.com>, freebsd-fs@freebsd.org Subject: Re: HAST + ZFS + NFS + CARP Message-ID: <65906F84-CFFC-40E9-8236-56AFB6BE2DE1@ixsystems.com> In-Reply-To: <20160703214723.GF41276@mordor.lan> References: <20160630144546.GB99997@mordor.lan> <71b8da1e-acb2-9d4e-5d11-20695aa5274a@internetx.com> <AD42D8FD-D07B-454E-B79D-028C1EC57381@gmail.com> <20160630153747.GB5695@mordor.lan> <63C07474-BDD5-42AA-BF4A-85A0E04D3CC2@gmail.com> <678321AB-A9F7-4890-A8C7-E20DFDC69137@gmail.com> <20160630185701.GD5695@mordor.lan> <6035AB85-8E62-4F0A-9FA8-125B31A7A387@gmail.com> <20160703192945.GE41276@mordor.lan> <20160703214723.GF41276@mordor.lan>
next in thread | previous in thread | raw e-mail | index | archive | help
> On Jul 3, 2016, at 2:47 PM, Julien Cigar <julien@perdition.city> = wrote: >=20 > I guess that ZFS will split the read requests accross all devices in > order to maximize performance... which could lead to contrary to what = is > expecpted when iSCSI disks are involved, no? > Is there some sysctl params which could prevent this unexpected > behavior? Nope. You will suffer the performance implications of layering a = filesystem that expects =E2=80=9Crotating media or SSDs=E2=80=9D (with = the innate ability to parallelize multiple requests in a way that ADD = performance) on top of a system which is now serializing the requests = across an internet connection to another software layer which may offer = no performance benefits to having multiple LUNs at all. You can try = iSCSI-specific tricks like MPIO to try and increase performance, but ZFS = itself is just going to treat everything it sees as =E2=80=9Ca disk=E2=80=9D= and so physical concepts like mirrors or multiple vdevs for performance = won=E2=80=99t translate across. Example question: What=E2=80=99s the point of writing multiple copies of = data across virtual disks in a mirror configuration if the underlying = storage for the virtual disks is already redundant and the I/Os to it = serialize? Example Answer: There is no point. In fact, it=E2=80=99s a = pessimization to do so. This is not a lot different than running ZFS on top of RAID controllers = that turn N physical disks into 1 or more virtual disks. You have to = make entirely different performance decisions based on such scenarios = and that=E2=80=99s just the way it is, which is also why we don=E2=80=99t = recommend doing that. - Jordan
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?65906F84-CFFC-40E9-8236-56AFB6BE2DE1>