Re: [gentoo-user] Re: Gentoo speed comparison to other distros

[Pandu Poluan]

On Mar 15, 2013 7:31 AM, "Mateusz Kowalczyk" <fuuzetsu@×××××××××××.uk>
wrote:
>
> On 14/03/13 23:52, Dale wrote:
> > Mateusz Kowalczyk wrote:
> >> On 14/03/13 22:41, Dale wrote:
> >>> Grant Edwards wrote:
> >>>> On 2013-03-14, Dale <rdalek1967@×××××.com> wrote:
> >>>>
> >>>>> I was wondering.  Has anyone ever seen where a test as been done to
> >>>>> compare the speed of Gentoo with other distros?  Maybe Gentoo
compared
> >>>>> to Redhat, Mandrake, Ubuntu and such?
> >>>> I just did a test, and they're all the same.
> >>>>
> >>>> CDs/DVDS of various distros dropped from a height of 1m all hit the
> >>>> floor simultaneously [there are random variations due to aerodynamic
> >>>> instability of the disk shape, but it's the same for all distros]. If
> >>>> launched horizontally with spin to provide attitude stability (thrown
> >>>> like a frisbee), they all fly the same.
> >>>>
> >>>> The point being, you're going to have to define "speed".
> >>>>
> >>>> Does speed refer to
> >>>>
> >>>>  Installation time?
> >>>>
> >>>>  Boot time?
> >>>>
> >>>>  Linpack?
> >>>>
> >>>>  Dhrystone?
> >>>>
> >>>>  Whetstone?
> >>>>
> >>>>  Time for me to figure out how to fix a configuration problem?
> >>>>
> >>>>  Time to do to an update on a machine that's been unplugged for a
year?
> >>>>
> >>>>  Time to to produce a packaged version of some random C program that
> >>>>  comes with a Makefile that uses autotools?
> >>>>
> >>>>  Time for a reported bug to get fixed?
> >>>>
> >>>
> >>> OK.  It appears not very many can figure out what I asked for.  So,
let
> >>> me spell it out for those who are challenged.  LOL   ;-)  Read some
> >>> humor into that OK.
> >>>
> >>> Install a OS.  Run tests on a set of programs and record the time it
> >>> takes to complete a certain task.  More tasks the better.
> >>>
> >>> Then install another OS on the same hardware.  Run tests on a set of
> >>> programs and record the time it takes to complete a certain task.
 More
> >>> tasks the better.
> >>>
> >>> The object of this is, does Gentoo with the customization it allows
run
> >>> faster than some binary install that does NOT allow those controls?
 In
> >>> other words, can a Gentoo based install perform more efficiently than
a
> >>> binary based install like Redhat, Ubuntu or some other distro?
> >>>
> >>> I am NOT concerned about compile times or the install itself.
> >>>
> >>> Does that put the dots closer together for the challenged ones?  ROFL
> >>>
> >>> Dale
> >>>
> >>> :-)  :-)
> >>>
> >> The point of the challenged ones was that while we can take
measurements
> >> like these, it's rather meaningless to do so. The result will be
> >> different for every single person out there depending on their
> >> configuration, USE, CFLAGS and who knows what else.
> >>
> >> I can compile a package with support for 3 different DEs, few WMs, oss
> >> and alsa and about a billion things I will never use. Does this make
for
> >> a more or less of a meaningful test than doing the same test with no
> >> flags what so ever? There is no correct answer as it varies per user
> >> basis. The most meaningful measurements that we can probably take would
> >> be between different USE flags configurations. Maybe we can say that
> >> package ‘foo’ with certain USE and CFLAGS runs in less average time
than
> >> the same package on a distro Bar.
> >>
> >> In my opinion, it would be far more meaningful to measure the effect of
> >> different USE flags on the same package, *in relative time* on the same
> >> system. This would give us more idea about the impact of each flag as
> >> opposed to a very limited view of ‘package foo with certain specific
USE
> >> flags runs 10ms faster than the same package on the same hardware on a
> >> binary distribution’. If you still want such measurements and you want
> >> them to be somewhat meaningful to you, it is you who will have to take
> >> them. Unless there are some gross inconsistencies in run times on
> >> different distributions, we have no use for such measurement.
> >>
> >> Everyone understood what you asked for. It's _you_ that misunderstood
> >> their explanation for why it's meaningless to ask such a question in
the
> >> first place.
> >>
> >
> > I didn't miss anything.  I get what some are saying.  The reason for my
> > question is this.  Gentoo allows a person to customize the OS to the
> > specific hardware it is being run on.  Redhat and other binary distros
> > don't allow this, unless you compile your own packages which is no
> > longer really a binary install.
> >
> > So, if I install Redhat on my machine, would it be less efficient than
> > my Gentoo install which is customized for my hardware?  Has someone else
> > tested this and made it public?
> >
> > If people can't get this, never mind.
> >
> > Dale
> >
> > :-)  :-)
> >
>
> I don't think that it's plausible to take such measurement. We could set
> every USE flag possible for the package we are benchmarking to try and
> replicate the support for everything that the binary package is likely
> to have. We also have to do this for all its dependencies (and their
> dependencies and so on) to have nothing that could potentially influence
> the measurement. Assuming that portage complies with this (it won't), we
> compile the package with optimizations for our hardware. The result? We
> probably have the same result on Gentoo and the other distro.
>
> Why? The reason is simple: binary distributions provide packages
> compiled with optimizations turned on for specific architectures. Unless
> you are doing some unheard of optimizations for your obscure model of
> the CPU, I don't imagine you'd get much advantage at all. If the
> maintainer of the binary package compiles it with optimizations for an
> i7 and you do the same, why should the performance be different?
>
> If you install RedHat on your machine, it probably will be less
> efficient than Gentoo but for a different reason. It won't be
> (noticeably) more efficient because you've got your CFLAGS set in a
> particular way but it because you only install the packages you actually
> want and remove support for things you don't need. Why bloat a package
> and waste cycles having it try to poll a service you might never
> actually use? Gentoo lets you get rid (or never install in the first
> place) of this kind of… bloat while you don't have much choice on a
> binary package short of compiling everything by hand at which point you
> should be using Gentoo to do it for you.
>
> RedHat maintainers aren't stupid (you can probably tell I've never used
> RH) – they will release packages optimized for architectures they will
> run on. Overall you might get very slight performance boost because of
> some CFLAG you enable but you might as well have worse performance
> because you don't know as much about optimizations as the RH maintainers
> and developers. Bah, you can even find examples on Gentoo wiki where
> compiling certain packages with certain flags actually makes them slower
> and not faster where usually the opposite is the case.
>
> --
> Mateusz K.
>

AFAIK, binary distros have only two kinds of targets for Intel/AMD
processors: x86 or amd64.

I think most binary distros set their -march or -mtune to nothing newer
than a Pentium Pro (x86 case) or Pentium 4 (amd64 case). This means, newer
generation instruction sets are not used.

Of course, the amount (%) of code that can be optimized is very small
compared to the rest of the code. On most programs, there will be no
benefit of mtune/march to a specific processor.

However, if you happen to frequently use a program whose major processing
time happens to be optimizable with the newer instructions, you hit a
jackpot.

TL;DR : Just like everybody say, Gentoo might be faster, but depends
totally on the jobs one do on it. On majority of cases, the faster speed is
totally inconsequential.

Rgds,
--