On 10/30/05, Peter Zaitsev <[EMAIL PROTECTED]> wrote: > Hi, > > > Tested this feature to see if it helps me with apache benchmark problem > - no it does not . > > Also it looks like it is firewall related issue as if firewall is > totally disabled (pf fails to load rules) everything works as > expected.
If its failing to load the rules you would have a alert on the GUI. Try running pfctl -f /tmp/rules.debug to see the errors. > Speaking about Network pooling - in my case it increased packet round > trip (2 Gbit Nicks) from 0.4 ms to 2ms. At the same time it well > decreased CPU usage during the tests so this is something to consider if > CPU performance ever becomes the problem. > > On other hand I was a bit surprised - according to vmstat number of > interrupts even on idle box jumped to some 30.000/sec (from some 150 > without this option set) > > I guess these are timer interrupts are used for pooling, so why they are > pooled about 1000 times per second if we get so many timer interrupts ? > > One more thing to note: system needs to be restarted for this option > to take an affect, however it does not say so anywhere. Try "playing around" with these sysctl's: kern.polling.each_burst: 80 kern.polling.burst_max: 1000 Also note these sysctl's (some are read only): http://www.pfsense.com/pastebin/276 Scott --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
