On Sun, Sep 23, 2007 at 08:18:05PM +0200, Andre Poenitz wrote:
> On Fri, Sep 21, 2007 at 03:51:26PM +0200, Enrico Forestieri wrote:
> > On Fri, Sep 21, 2007 at 02:13:55AM +0200, Andre Poenitz wrote:
> >
> > > On Thu, Sep 20, 2007 at 02:23:28PM +0200, Enrico Forestieri wrote:
> > > > > What about deciding that we do not install the libraries? Of course,
> > > > > this would imply that an installable LyX is a static one, but I am not
> > > > > sure we care much about distributing dynamic versions anyway. We could
> > > > > for example decide that shared is turned on only for development
> > > > > unix builds.
> > > >
> > > > Agreed. And I am not even sure of the usefulness of a dynamic build
> > > > when you gain 15 seconds or so.
> > >
> > > static: touch Text.cpp && time make:
> > >
> > > real 1m4.922s
> > > user 0m8.145s
> > > sys 0m2.404s
> > >
> > > dynamic: touch Text.cpp && time make:
> > >
> > > real 0m27.072s
> > > user 0m15.045s
> > > sys 0m1.552s
> >
> > Thank you for demonstrating that a dynamic build is slower than a
> > static one. Indeed, the real field gives the elapsed (real) time
> > between invocation of the command and its termination. This means
> > that it is a varying quantity depending on the system load.
> > The useful information comes from the user and sys fields and their
> > sum demonstrates that a static build is faster.
>
> It was an unloaded system in both cases. I know what "real" and "user"
> means in theory and I couldn't care less. In practice, one takes 65 seconds
> to finish, and the other takes 27. I clearly prefer the 27 second
> version even if somebody thinks that the 65 seconds are "really only 8".
>
> > > So it reduces simple roundtrip times byh more than a factor of 2.
> > >
> > > Try again.
> >
> > Hrmpf.
>
> Try again.
Hrmpf.
I have results different from yours. I ran three times the commands
and took the lowest "real" value. Note that "user" and "sys" values
practically didn't change between runs.
$ uname -prismo
SunOS 5.10 sun4u sparc SUNW,Sun-Blade-1000 Solaris
$ psrinfo -v
Status of virtual processor 0 as of: 09/25/2007 10:24:13
on-line since 07/04/2007 09:37:45.
The sparcv9 processor operates at 750 MHz,
and has a sparcv9 floating point processor.
static: touch Text.cpp && time make:
real 1:03.7
user 41.9
sys 3.3
dynamic: touch Text.cpp && time make:
real 1:02.8
user 37.4
sys 2.8
$ uname -prismo
SunOS 5.10 sun4u sparc SUNW,Sun-Blade-1500 Solaris
$ psrinfo -v
Status of virtual processor 0 as of: 09/25/2007 10:45:32
on-line since 07/04/2007 11:07:08.
The sparcv9 processor operates at 1503 MHz,
and has a sparcv9 floating point processor.
static: touch Text.cpp && time make:
real 53.1
user 21.6
sys 2.2
dynamic: touch Text.cpp && time make:
real 42.7
user 18.9
sys 1.9
$ uname -prismo
Linux 2.6.21-2-686 i686 unknown unknown GNU/Linux
$ cat /proc/cpuinfo | egrep "processor|model name|MHz"
processor : 0
model name : Intel(R) Core(TM)2 CPU T5200 @ 1.60GHz
cpu MHz : 1600.000
processor : 1
model name : Intel(R) Core(TM)2 CPU T5200 @ 1.60GHz
cpu MHz : 1600.000
static: touch Text.cpp && time make:
real 0m18.692s
user 0m16.849s
sys 0m1.628s
dynamic: touch Text.cpp && time make:
real 0m16.662s
user 0m15.245s
sys 0m1.316s
I'll leave to you the calculation of the reduction factor for the
roundtrip times.
--
Enrico
