On 03/09/15 15:24, Henning Brauer wrote:
* Federico Giannici <[email protected]> [2015-02-04 01:11]:I have done an experiment: I replaced in every rule the "set queue XXX" with "tag XXX" ("XXX" is always different so the PF optimizer doesn't collapse multiple rules in tables). In this way we found that, leaving the some amount of filter rules and only removing the queue, the CPU used in interrupts decreased from about 55% to 15% (traffic was not full in that moment).something is fishy here, since "queue foo" just tags, which coincidently is very much like "tag foo" - really almost identical codewise.
OK, but only for the rules evaluation. Then, in the case of queues, all the bandwidths (maximum, granted, etc) must be evaluated. I think here is the different and slow code.
since you're running 5.5, I'll assume ALTQ and thus the problem being gone :)
Yes, 5.5 but using the new queues definition (not "oldqueue"). Are you saying that the queues code has been replaced AFTER 5.5? Thanks.
