[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: PF, ALTQ, Priq & RED



On Mon, Jan 03, 2005 at 11:32:35AM +1100, Matt Pearce wrote:
> 
> If I am running 2 rules for udp packets to be prioritized and I want a 
> specific rule for prioritizing dns udp out to take preference over the 
> generic udp altq out rule, do I need this rule to be above the generic 
> rule and have the quick keyword or doesnt it matter as the dns rule has 
> been given a higher queue number anyways ??
  order of actual filter rules wherein you assign packets to queues
  in pf.conf doesn't matter.
  higher altq priority is what matters.
> Secondly, if I dont apply the (red) tag to some of my queue's that I 
> want the highest priority to, and I apply (red) to queue's with lower 
> priorities, does this mean traffic on the lowests queue's could be 
> dropped while none would be dropped on queue's where (red) is not 
> specified ??
  i think it means that in a situation where queued outbound packet
  bandwidth consumption is nearing/reaching (in your case) 220Kb/s, 
  queues with red on them will have packets dropped proportional
  to how close to 220 you are getting.  my understanding from
  reading and watching queues in action is that a queue with red
  shall not ever actually _reach_ its fully saturated level in 
  a cbq situation, but will asymptotically (or something?  math word)
  fall *towards* 0 as traffic increases to the limit of the declaration.
  in priq, it might be considered differently, such as if a saturation
  situation exists, packets with a lower priority will be dropped
  first?  or maybe last?  
  for your question directly, if you have one queue with red and
  one without, the red one will have packets dropped out near
  saturation, and the not-red one will have packets queued up in
  the 'queue' thing (as seen in pfctl -vvsq) as much as they
  can before being dropped, afaict.
  red is something i have steered away from preferring currently to
  let packets fill up in queuelen rather than be dropped, but my
  understanding of altq is no where near the level of those who
  have implemented it in code.
  jared
-- 
[ openbsd 3.6 GENERIC ( dec 11 ) // i386 ]