[Debian-ppc64-devel] Re: ppc64 archive bloating alioth disk ...

[Andreas Jochens]

On 05-Sep-21 08:44, Sven Luther wrote:
> On Wed, Sep 21, 2005 at 08:18:57AM +0200, Andreas Jochens wrote:
> > Here is the output from ldd on i386:
> > 
> >         linux-gate.so.1 =>  (0xffffe000)
> >         libX11.so.6 => /usr/X11R6/lib/libX11.so.6 (0x5556d000)
> >         libXpm.so.4 => /usr/X11R6/lib/libXpm.so.4 (0x55638000)
> >         libz.so.1 => /usr/lib/libz.so.1 (0x41862000)
> >         libcrypt.so.1 => /lib/tls/libcrypt.so.1 (0x5564f000)
> >         libssl.so.0.9.7 => /usr/lib/i686/cmov/libssl.so.0.9.7 (0x5567d000)
> >         libstdc++.so.6 => /usr/lib/libstdc++.so.6 (0x556ae000)
> >         libm.so.6 => /lib/tls/libm.so.6 (0x41153000)
> >         libgcc_s.so.1 => /lib/libgcc_s.so.1 (0x55793000)
> >         libc.so.6 => /lib/tls/libc.so.6 (0x41019000)
> >         libdl.so.2 => /lib/tls/libdl.so.2 (0x4118f000)
> >         libXext.so.6 => /usr/X11R6/lib/libXext.so.6 (0x5579f000)
> >         libcrypto.so.0.9.7 => /usr/lib/i686/cmov/libcrypto.so.0.9.7 (0x557ad000)
> >         /lib/ld-linux.so.2 (0x41000000)
> > 
> > openssl and xorg should be enough for this as far as I can tell.
> 
> This is on a amd64 box though, does it make some difference ? 

The only differences are the '/i686/cmov' and the '/tls' prefixes in 
the library path.

> > The performance hit is roughly 10% for current hardware and it mainly
> > affects compile time - not run time. Applications which really use
> > 64 bit wide calculations and/or a 64 bit address space of course run
> > faster than 32 bit ones.
> 
> On an amd64 box or on ppc64 ?

This is on ppc64. On amd64 most things run ~30% faster in 64 bit mode
even if they do not really use any 64 bit calculations or 64 bit address
space.

> 
> > Basically the Debian standard *nix tools seem to be optimized for the 32 
> > bit case. I guess that this will change in the future. 
> > Additionally, with future new processor variants the performance hit 
> > will probably become much smaller anyway.
> 
> And in full disregard of what the IBM linux guys say about this on the PPC970
> ? Don't you know they know best, since they work closely with the folk
> building the cpu ?

I had some communications with people from IBM and they seemed to share
my view. But YMMV.

> > I am convinced that people will switch to clean 64 bit environments
> > in the end. For serious server environments, 32 bit setups will soon be 
> > a thing of the past.
> 
> Well, but not in etch timeframe anyway.

That is true. And I certainly have nothing against creating the 
possibility of a biarch environment - as long this does not block 
the development of a clean 64 bit environment.

Regards
Andreas Jochens