Date: Wed, 15 Aug 2018 08:10:30 -0700 From: Mark Millard <marklmi@yahoo.com> To: Trev <freebsd-arm@sentry.org> Cc: freebsd-arm@freebsd.org Subject: Re: RPI3 swap experiments (grace under pressure) Message-ID: <01136DB2-F635-4F2A-9167-C34195993914@yahoo.com> In-Reply-To: <16004AB0-B499-4F30-8C14-AAE5E45B0A3B@yahoo.com> References: <20180812173248.GA81324@phouka1.phouka.net> <20180812224021.GA46372@www.zefox.net> <B81E53A9-459E-4489-883B-24175B87D049@yahoo.com> <20180813021226.GA46750@www.zefox.net> <0D8B9A29-DD95-4FA3-8F7D-4B85A3BB54D7@yahoo.com> <FC0798A1-C805-4096-9EB1-15E3F854F729@yahoo.com> <20180813185350.GA47132@www.zefox.net> <FA3B8541-73E0-4796-B2AB-D55CE40B9654@yahoo.com> <20180814014226.GA50013@www.zefox.net> <02fe39af-a02c-fb6a-70b0-da3b7fd06c22@goodgas.com.au> <20180814213107.GA51051@www.zefox.net> <D04BB7AF-503F-4779-B31B-AEF2A15EA7C6@jeditekunum.com> <bc2b4aa9-5284-502b-1745-aa531c12c5b8@m5p.com> <2f3bed05-b27e-420b-e831-1c8286edf35e@sentry.org> <16004AB0-B499-4F30-8C14-AAE5E45B0A3B@yahoo.com>
next in thread | previous in thread | raw e-mail | index | archive | help
[Add notes about the clan g versions around in 2016-Oct.] On 2018-Aug-15, at 7:39 AM, Mark Millard <marklmi at yahoo.com> wrote: > On 2018-Aug-14, at 5:20 PM, Trev <freebsd-arm at sentry.org> wrote: >=20 >> George Mitchell wrote on 15/08/2018 08:33: >>> On 08/14/18 18:17, Jedi Tek'Unum wrote: >>>> I firmly disagree with the entire concept of out of memory killers. = They are simply evil and in my opinion a complete cop-out. I first = encountered this kind of kludge back in the =E2=80=9880s with AIX. It = was bad then and it still is today. Frankly I find it ridiculous that = they still exist. >>>> [...] >>> However: consider the subject (Raspberry Pi). -- = George >>=20 >> When researching whether 512M of RAM was considered "usable" for a = FreeBSD buildworld, I came across [https://reviews.freebsd.org/D8302] = from October 2016 where just 256M was considered a test case for i386 = and amd64 (-j1 I'm assuming) buildworlds which should succeed. >=20 > What I see there is pho writing: >=20 > QUOTE > I have run stress2 testes on i386 and amd64. > I ran a buildkernel on both i386 and amd64 with 256MB RAM / UP. > Buildworld was run with various small RAM configurations. > END QUOTE >=20 > It explicitly lists buildkernel for 256 MiBytes of RAM, not > buildworld. I'm not sure what the "UP" is for. "RAM / UP" > looks like it might be a ratio but may be it was indicating > not SMP? It not clear if this buildkernel testing included > kernel-toolchain as well. >=20 > buildworld was listed separately with nothing explicit about > what "small RAM" was considered to be. >=20 > If stress2 is configurable, there is no information on what > the specific test was for retrying the same test or knowing > just what the test was. I'm not sure which toolchain was used for any buildworld or any kernel-toolchain activity. But looking around inside https://svnweb.freebsd.org/base/projects/ shows that https://svnweb.freebsd.org/base/projects/clang390-import/?view=3Dlog had a last check-in on 2016-Nov-24 and was created on 2016-Aug-16. So clang 3.9.0 or an earlier clang. (clang was the default for x86's so I've assumed that it was used.) I'm not sure how similar that was to now with clang 6. It might be interesting if pho's tests could be re-run but for modern head. >> Regardless, my RPi3B+ OOMA issues were completely eliminated by = replacing the 16G SanDisk "Ultra" card with a "faster" 32G SanDisk = "Extreme" card which contains all file systems as well as a swap = partition and runs -j4 buildworld flawlessly. I think this lends some = support to Warner's view. =3D=3D=3D Mark Millard marklmi at yahoo.com ( dsl-only.net went away in early 2018-Mar)
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?01136DB2-F635-4F2A-9167-C34195993914>