We've solved the problem increasing net.inet.ip.ifq.maxlen from the default of our version (50) to the default of the more recent versions (250). Does it make sens to you? How far do you think we can go with that value considering that we've 3 physical interfaces (int 100mbit, ext 100mbit and pfsync 10mbit) and that the servers have only 512Mb of RAM? Something like "Henning's rule" with 256*3 (number of physical interfaces) would be a good and safe choice with our hardware (of course we're planning an upgrade of both servers and openbsd version)? Thanks for your help Alessandro
On Tue, Nov 29, 2011 at 7:49 PM, rik <rikc...@gmail.com> wrote: > Thanks for the suggestion, I'll try with the GENERIC kernel > Is that possibile that this problem is due to hardware limitation (it's > quite an old server)? Apparently when the traffic decrease the packet loss > decrease as well and disappear just like the odd ping's result > Thanks! > Alessandro > > > > On Tue, Nov 29, 2011 at 12:15 AM, Stuart Henderson > <s...@spacehopper.org>wrote: > >> On 2011-11-28, James Shupe <jsh...@osre.org> wrote: >> > Your dmesg doesn't show the version you're running. Can you provide >> > that, >> >> Yep, seconded. If people ask for a dmesg, they mean a complete one. >> I would also try a GENERIC kernel (not GENERIC.MP). >> >> > along with ifconfig output from both machines? You may want to >> > check the physical connectivity (cable/ NIC/ switch) for the internal >> > interface of the carp master... Or just fail over to the secondary box >> > to see if the issue goes away. >> >> Well there appears to be something very odd going on with timers there >> so who knows what else might follow from that. >> >> >>>>> 64 bytes from xx.xx.xx.12: icmp_seq=4 ttl=64 time=-3.-656 ms >> >>>>> 64 bytes from xx.xx.xx.12: icmp_seq=5 ttl=64 time=0.794 ms >> >>>>> 64 bytes from xx.xx.xx.12: icmp_seq=6 ttl=64 time=0.-491 ms >> >>>>> ping: sendto: No route to host >> >>>>> ping: wrote xx.xx.xx.12 64 chars, ret=-1 >> >>>>> ping: sendto: No route to host >> >>>>> ping: wrote xx.xx.xx.12 64 chars, ret=-1 >> >>>>> 64 bytes from xx.xx.xx.12: icmp_seq=9 ttl=64 time=0.526 ms >> >>>>> 64 bytes from xx.xx.xx.12: icmp_seq=10 ttl=64 time=1.415 ms