cups does not scale very well

[Anonymous]

Hi

we are still using a old fashioned system V spooler design to run abount 1900 printers on some old SUN Netra t1 with 64 MB RAM.
Do to some minor design problems with this old stuff i tried to get the
work done with cups but:

The creation of all those printers on the test cups box (Ultra Enterprise V240) took about 8 hours. At the beginning about 200 printers where created within 2 minutes, we are not using PPD's just plain System V interface scripts. At the End the performance of creating printers slowed down to 2 to 3 printers a minute :-(.

A full reload of the configuration took about 90 minutes until cups did
respond again. The creation of printcap is already disabled, i assume otherwise it would take several minutes longer.
To slurp in a configuration file for 1900 printers shouldn't take that
long, but a truss shows about 2 to 4 seconds per printer. Just a thought: we use nagios with several thousands of services to be monitored and several hundred hosts with a complex configuration which is being read within seconds.

10 little print jobs took 4 minutes to be scheduled, just /etc/hosts or similar, the printing after the scheduler hat taken them was quick.
The CPU was during print job reception at about 100%. Using a second client to feed print jobs doubled the time to handle the jobs.

Any hints to get cups any faster?