Software Management call for RFEs
fweimer at redhat.com
Wed May 29 18:33:16 UTC 2013
On 05/27/2013 10:24 AM, Jan Zelený wrote:
>> As far as I can tell, the main difference is that apt-get and apt-cache
>> read very few, relatively large files at the beginning, so they don't
>> block on disk reads early.
>> dpkg, on the other hand, uses a database scatter across many small files
>> on disk, so you get the delay only when you actually install or remove
>> any packages. At the beginning, this is quite fast, but eventually, the
>> files will be scattered quite badly, and there is a considerable delay
>> at this step.
> This part is about disk read-write but that was not what I was writing about.
> From my experience users mostly complain about the metadata download which is
> explained above.
Users sometimes misdiagnose issues, *especially* when complaining. 8-)
I did some tests and cold cache performance tests on an old Debian
installation. Performance with cold caches is more than adequate.
Full-text searches take about two seconds. Package installation reaches
the confirmation prompt (after dependency resolution) in less than three
seconds, even for ridiculously complex tasks such as installing the
entire KDE desktop (365 additional packages on my test system).
In contrast, on Fedora or RHEL systems, 30 seconds for dependency
resolution with a cold cache are common, plus around 6 seconds for
loading all the Python code for yum. /usr/bin/time reports much more
I/O than it does non Debian (about ten times as much, as reflected in
the wall clock time).
Regarding network traffic, in addition to the explicit "apt-get update"
step (which avoids time-consuming downloads in the first place), Debian
also forces users to pick a single mirror close to them. The Fedora
master mirror list instructs yum to pick a mirror from a larger list,
which causes much greater variance in performance.
Florian Weimer / Red Hat Product Security Team
More information about the devel