Red Hat Bugzilla – Bug 150777
RPM performance bad
Last modified: 2014-06-14 23:54:46 EDT
+++ This bug was initially created as a clone of Bug #118187 +++
From Bugzilla Helper:
User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.6) Gecko/20040124
Description of problem:
I have noticed that rpm seems very slow and uses a lot of cpu.
# time rpm -qa |wc -l
Granted, this is on a 550MHz P-III machine, but a similarly sized machine
running a different rpm based distro does the same thing in less than 20% of
Version-Release number of selected component (if applicable):
rpm-4.3.2-21 (other versioner are affected too)
Steps to Reproduce:
1.run time rpm -qa >/dev/null or something similar
2.notice cpu usage hits the roof and query takes longer than it should
Actual Results: rpm slow to complete transaction
Expected Results: better performance (comparable to other rpm based distros),
less CPU usage
You can run rpm --stats -qa for a better breakdown.
If you are going to compare performance with another rpm version it's helpful to
have the version, vendor, stats, etc. Also testing against the same rpmdb would
Bad compared to what?
Well, i guess i have to admit that I was a little eager on the trigger with this
investigating the performance a little further brings the followinf info.
The rpm version i'm comparing to is a version 3 from SuSE.
It seems like this version is a little lighter on the system because it either
lack the digest/signature functionality of later rpm versions or at least
doesn't exercise those functios quite as often.
My system spends most of the time doing signature work
# rpm --stats -qa >/dev/null
total: 1 0.000000 MB 9.571010 secs
digest: 970 14.196385 MB 0.564451 secs
signature: 468 0.000000 MB 7.223979 secs
dbget: 541 14.529588 MB 1.120793 secs
Guessing that this is just not done in the SuSE version, disabling the
functionality in my FC3 version, changes the picture completely
# rpm --stats --nosignature -qa >/dev/null
total: 1 0.000000 MB 1.374332 secs
digest: 970 14.194000 MB 0.562437 secs
dbget: 490 14.455532 MB 0.278045 secs
This pretty much closes the entire discussion for me (should have done all this
before even filing the bug, sorry about that) and the only thing i would like to
know is, are there a valid reason for checking signature of each and every
package in a simple "list" operation like -qa?
The Principle of Least Surprise says check header
signatures everywhere and always.
Meanwhile, signature checking on queries can be globally
disabled by adding to /etc/rpm/macros