Gentoo Archives: gentoo-amd64

From: Mark Knecht <markknecht@×××××.com>
To: Gentoo AMD64 <gentoo-amd64@l.g.o>
Subject: Re: [gentoo-amd64] Re: How much value does llvm provide for a low-use laptop?
Date: Wed, 26 Apr 2017 21:31:13
Message-Id: CAK2H+ed-NR1mgoKJwScD6kNWbavpT7cFsUg6bW1T0EPbbVU59Q@mail.gmail.com
In Reply to: [gentoo-amd64] Re: How much value does llvm provide for a low-use laptop? by Duncan <1i5t5.duncan@cox.net>
1 On Wed, Apr 26, 2017 at 6:04 AM, Duncan <1i5t5.duncan@×××.net> wrote:
2 >
3 > Mark Knecht posted on Tue, 25 Apr 2017 10:43:29 -0700 as excerpted:
4 >
5 > > 3) Respectfully, I'm not sure your answer encompasses the problems and
6 > > frustrations of having to maintain OTHER people's computers. I don't
7 > > hear you speak of that very often. The problem with KDE is on my wife's
8 > > computer. When it's building KDE it's unavailable to her. In the past 2
9 > > weeks I've had two massive builds that each took about 24 hours. That
10 > > amounts to about 15% downtime on her machine,
11 >
12 > What I'm trying to say, tho, is that if you set it up right, you'll only
13 > be building once, for your machine, or at least /on/ your machine if it's
14 > a package you don't yourself use, and will then be using already built
15 > binpkgs on her machine.
16 >
17 > So effectively it's like using a binary distro on everything except your
18 > build machine, only the binary distro will be customized with your chosen
19 > gentoo profile, USE flags, etc.
20 >
21 > IOW, bigger picture, the gentoo as metadistro idea, with you effectively
22 > creating the customized binary distro out of it with the build on your
23 > machine, that you then install on your wife's machine, and however many
24 > more you have around that you maintain or help maintain.
25 >
26 > Now depending on how similar the machines and layouts are, you may still
27 > end up building a /few/ packages individually for each or at least some
28 > of the machines, but if you choose your battles (packages) well, it'll be
29 > perhaps 10% of them, and "big" packages like gcc, firefox or chromium,
30 > etc, will only be built once. Tho if you have say kde on some and xfce
31 > on others, you might be building one or the other of them for perhaps one
32 > machine only, and certainly, kde at least is big, but still, if you're
33 > building for say 10 machines and a few packages are only used on one or
34 > two, with another few that you have to rebuild custom for each one, you
35 > might be building in total say 120% or 150% or even 200% of what you'd
36 > build for a single machine, but that's still way better than the 1000%
37 > (100% * 10) that you'd be building if you did each one individually.
38 >
39 > And while not /exacty/ the same as you'd get with all individual builds
40 > (the 1000%), it'd still be way closer to fully customized individual
41 > builds then the generic target mass distribution build you'd get using a
42 > normal binary distro.
43 >
44 > Meanwhile, the per-machine update and admin time, for other than that
45 > first build machine, would be very nearly the same as you'd spend with a
46 > mass binary distro anyway, and actually possibly less than the time you'd
47 > spend if you were splitting distros and having to keep up with the
48 > different ways different distros did things.
49 >
50 > At that point the update and admin time on your wife's machine would
51 > probably be /less/ staying with gentoo, because you'd be doing binpkg
52 > installs with already-built packages done on your main machine, and being
53 > gentoo, you'd know it better and be more effective at admin, so you'd
54 > actually spend less time on the admin side than you would if it were the
55 > only machine you had running ubuntu (or fedora or whatever), and thus
56 > dealing with any changes to config for the first and only time on her
57 > machine.
58 >
59
60 Yeah, I think the idea is at least worth investigating. However this is one
61 of
62 those ideas that I always thought sounded good on paper but would likely
63 be a problem in real life. However maybe these machines are close enough
64 to make it a good option.
65
66 Other than the aes flag the two CPUs appear to support the same features. I
67 have no idea how that maps into system performance or application space.
68 I'll
69 have to research that a bit. Not using it might make my machine a little
70 slower
71 on web pages or something like that but I doubt I'll see it. I could
72 investigate
73 turning that off and seeing what my machine wants to rebuild. On the other
74 hand, it might make a bigger difference inside the Windows VMs.
75
76 mark@c2RAID6 ~ $ cpuinfo2cpuflags-x86
77 CPU_FLAGS_X86="aes mmx mmxext popcnt sse sse2 sse3 sse4_1 sse4_2 ssse3"
78 mark@c2RAID6 ~ $ ssh laptop
79 Password:
80 mark@slinky ~ $ cpuinfo2cpuflags-x86
81 CPU_FLAGS_X86="mmx mmxext popcnt sse sse2 sse3 sse4_1 sse4_2 ssse3"
82 mark@slinky ~ $
83
84 I'll need to see about bringing the two machines into parity in terms of
85 portage
86 config files. Other than LibreOffice I'm pretty sure our world files are
87 close
88 and Office is binary anyway.
89
90 She actually uses very few apps. The big sticking point here is the
91 time it takes to build KDE which I'm going to do anyway on my machine so
92 if the binary packages work that would be great.
93
94 As I'm travelling next week there's no rush on this. I'll do some reading
95 and think about possibly giving it a try.
96
97 Cheers,
98 Mark