Reducing Fedora memory footprint?

[Panu Matilainen]

On Fri, 24 Nov 2006, Gilboa Davara wrote:
>
> Solution wise - yes, yum and apt are different - but target-wise, they
> both designed to serve the same purpose.
> Question is - can yum be optimized (E.g. by replacing the XML parser to
> a faster/leaner one) - bringing it to a point where the performance
> difference between Debian's apt and Fedora's yum is less staggering?
> (Especially in query tasks)

The XML is already parsed by a module written in C and is fast enough not 
to be a bottleneck I think. The XML files are only read during importing 
the data into sqlite database, which is then used for all operations 
(depsolving, querying etc).

FWIW, I don't think 'yum search foo' and such are slow at all these days, 
only the depsolve stage remains a bit of a sore point speedwise. Where 
exactly the time is spent and how to improve things .. profiling needed :)

BTW, Debian packages have far less dependency data in them than an average 
rpm package does because automatic dependencies and provides are not used 
there. Just an example:

$ dpkg -f libc6_2.3.6.ds1-8_amd64.deb |grep Provides
Provides: glibc-2.3.6.ds1-1, glibc-2.3.6-2

Contras that with FC6 glibc:
$ rpm -qp --provides glibc-2.5-3.x86_64.rpm |wc -l
     300

Looking through 300 provides for matches is more expensive than going 
through 2 provides :) glibc is a bit of an extreme case of course, not 
every package is *that* bloated with soname-provides but it adds up pretty 
fast and doesn't help speeding things. OTOH Debian has a larger 
repository... my point is that there are lots of differences between deb 
and rpm systems, comparing them in any sane way is not trivial.

 	- Panu -