Date: Wed, 18 Oct 2023 20:32:51 -0500 From: Eric Borisch <eborisch@gmail.com> To: freebsd-fs@freebsd.org Subject: Re: free space considerations writing bhyve image to a zvol Message-ID: <CAMsT2=kc3rrTL=S%2B=Woxz0te=Jgr=u_HRLibDZ=XiifdBkob-Q@mail.gmail.com> In-Reply-To: <ZS_ivB48NWpD1D8s@int21h> References: <ZS_WZD-JlHEo_Em5@int21h> <1733602178.6691.1697636557003@localhost> <ZS_ivB48NWpD1D8s@int21h>
next in thread | previous in thread | raw e-mail | index | archive | help
--000000000000ac44ef060807bcf3 Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable Be aware that compression of a zvol can be _very_ different from compression of a file -- zvols must work on volblocksize-d segments to compress, and are also forced into integer multiples of the underlying pool's sector size (2**ashift). Ashift =3D 12 -> 4k sector size are typical these days, so a zvol with volblocksize=3D4k will not compress (save trivial all-0 areas and the like)= . With volblocksize=3D8k, only if an 8k segment reaches 50% reduction can zfs successfully compress, 25% for 16k volblocksize, etc. On top of this, raidz setups bring their own allocation size requirements, which hit harder on small recordsize / volbolcksize settings (which again, are typically set smaller on zvols). If you don't need any of the shows-up-as-a-device and more predictable RW IOPS (operations on the volblocksize won't risk write amplification in the ZFS layers or require decompressing a larger record) features of zvol, and are more concerned about compression, leaving the VM's filesystem in a plan file will (assuming you haven't dialed down recordsize) give you better compression. - Eric On Wed, Oct 18, 2023 at 8:51=E2=80=AFAM void <void@f-m.fm> wrote: > On Wed, Oct 18, 2023 at 03:42:37PM +0200, Ronald Klop wrote: > >You can test how much it will compress your linuxvm.img. > > > >$ [/tmp] ls -l ./bla.txt > >-rw------- 1 ronald wheel 5242880 Oct 18 15:39 ./bla.txt > > > >$ [/tmp] lz4 -c ./bla.txt | wc -c > > 20604 > > > > > >So do "lz4 -c linuxvm.img | wc -c" and it will print about the size of > the zvol you need. > > thanks for that. I'm concerned about wrecking the linux vm's > filesystem internally in this process though. I know it's easier to > expand than contract; what i don't know is the detail > > >NB: What keeps you from just trying to write it to the zvol and see if i= t > works? > > downtime & space considerations on the system. And curious if people have > tried similar, and if they can recall the results. Searching hasn't > come up with a near-enough matching context. > -- > > --000000000000ac44ef060807bcf3 Content-Type: text/html; charset="UTF-8" Content-Transfer-Encoding: quoted-printable <div><div dir=3D"ltr"><div>Be aware that compression of a zvol can be _very= _ different from compression of a file -- zvols must work on volblocksize-d= segments to compress, and are also forced into integer multiples of the un= derlying pool's sector size (2**ashift).</div><div><br></div><div dir= =3D"auto">Ashift =3D 12 -> 4k sector size are typical these days, so a z= vol with volblocksize=3D4k will not compress (save trivial all-0 areas and = the like). With volblocksize=3D8k, only if an 8k segment reaches 50% reduct= ion can zfs successfully compress, 25% for 16k volblocksize, etc.</div><div= ><br></div><div>On top of this, raidz setups bring their own allocation siz= e requirements, which hit harder on small recordsize / volbolcksize setting= s (which again, are typically set smaller on zvols).</div><div><br></div><d= iv>If you don't need any of the shows-up-as-a-device and more predictab= le RW IOPS (operations on the volblocksize won't risk write amplificati= on in the ZFS layers or require decompressing a larger record) features of = zvol, and are more concerned about compression, leaving the VM's filesy= stem in a plan file will (assuming you haven't dialed down recordsize) = give you better compression.</div></div></div><div><div dir=3D"ltr"><div><b= r></div><div>=C2=A0- Eric<br></div></div><br><div class=3D"gmail_quote"><di= v dir=3D"ltr" class=3D"gmail_attr">On Wed, Oct 18, 2023 at 8:51=E2=80=AFAM = void <<a href=3D"mailto:void@f-m.fm" target=3D"_blank">void@f-m.fm</a>&g= t; wrote:<br></div><blockquote class=3D"gmail_quote" style=3D"margin:0px 0p= x 0px 0.8ex;border-left-width:1px;border-left-style:solid;padding-left:1ex;= border-left-color:rgb(204,204,204)">On Wed, Oct 18, 2023 at 03:42:37PM +020= 0, Ronald Klop wrote:<br> >You can test how much it will compress your linuxvm.img.<br> ><br> >$ [/tmp] ls -l ./bla.txt<br> >-rw-------=C2=A0 1 ronald=C2=A0 wheel=C2=A0 5242880 Oct 18 15:39 ./bla.= txt<br> ><br> >$ [/tmp] lz4 -c ./bla.txt | wc -c<br> >=C2=A0 20604<br> ><br> ><br> >So do "lz4 -c linuxvm.img | wc -c" and it will print about th= e size of the zvol you need.<br> <br> thanks for that. I'm concerned about wrecking the linux vm's<br> filesystem internally in this process though. I know it's easier to<br> expand than contract; what i don't know is the detail<br> <br> >NB: What keeps you from just trying to write it to the zvol and see if = it works?<br> <br> downtime & space considerations on the system. And curious if people ha= ve<br> tried similar, and if they can recall the results. Searching hasn't<br> come up with a near-enough matching context.<br> -- <br> <br> </blockquote></div> </div> --000000000000ac44ef060807bcf3--
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CAMsT2=kc3rrTL=S%2B=Woxz0te=Jgr=u_HRLibDZ=XiifdBkob-Q>