ports performance

Volker volker at vwsoft.com
Wed May 23 21:12:59 UTC 2007


I've lightly followed the ongoing discussion of enhancing the
performance of the ports system. But I'm unable to come to any
conclusion as one gives a patch, another one says it may break

Just to throw in a few numbers into this discussion:

2 weeks ago I took the chance to upgrade a test machine
(Ahtlon64-3400) and the machine was building ports for one or two
days (it was mostly unattended so I'm unable to give correct
values). Starting last sunday, I wanted to update my notebook (the
machine I'm using every day) to the latest ports - including Xorg 7.2.

I've had to abort port building today. After csup'ing the ports
tree, ~360 (out of ~640) ports needed to be updated.

As the frustration raised, I watched the machine performance and
figured out, port registration takes up to 12 minutes. Overall port
compilation is not an issue but the management of the ports database
is way too slow. Imagine: around 360 ports, each port takes between
6 and 12 minutes for registration. Let's calculate each registration
with only 6 minutes (most take more) that's 36 hours just for port
registration (w/o compilation, directory cleaning).

My notebook is a P4m-1.8GHz, 512 MByte system - shouldn't be too slow.

So, what's the fast solution for the average desktop to this problem?

Using top, I've seen ruby18 and pkg_create eat up a lot of cpu time.
Will 'downgrade' portupgrade-devel to portupgrade help on this?

To fix my notebook, I'm currently building packages on another
system, will delete all Gnome/X7.2 ports and start from scratch. I'm
afraid of the next major update (gnome comes to mind) if ports still
will take that long.

I will happily try beta testing fixes if they don't break anything.
BTW, `pkgdb -L' also takes too long and aborts without a message or
a core dump.



More information about the freebsd-x11 mailing list