conntrack performance

[Alexander Demenshin]

On Fri, Jul 21, 2000 at 10:47:06PM +1000, Ingo Luetkebohle wrote:

> How exact is that measurement? I'd say that 5% performance loss *is* a
> problem. Not that much in itself, but when coupled with other stuff (e.g.,
> normal packet filter rules), it quickly adds up.

  I would guess that you will never reach 100% CPU load in case if you are
  handling even 100Mbit ethernet on SMP like mine (dual PIII-550).
  
  My tests were performed on loopback interface with average packet rate
  of 4000 pkts/sec, CPU load was almost 100% (in case of croos-testing,
  i.e. from anothet stations over 100M ether it is near 10%).
  
  So I do not think that it does matter, extereme conditions are extreme
  conditions and may only be used as a measurement of stability of system
  anyway (what is exactly was the reason of my test).
  
/Al