On Wed, 19 Sep 2012 19:18:06 +0400 Gleb Smirnoff wrote:
That's strange. Per-minute read shouldn't induce large CPU load. What snmp daemon do you use? I remember that several years ago net-snmp daemon from ports used a single linked list for all ARP entries, and thus it consumed a lot of CPU when receiving a single ARP change from routing socket.
It was net-snmp-5.3.. port. Just checked on net-snmp-5.7.1 running on FreeBSD-9.0 and I see that things have changed significantly. And even # snmpbulkwalk -v2c -On -c <community> <routerip> 1.3.6.1.2.1.4.22.1.2 running in an endless loop doesn't seem to affect the number of context switches. Only the page faults count grow up to 50k per second. Don't know if this metric has some relation with router performance. # vmstat a 1 procs memory page faults cpu r b w avm fre flt re pi po fr sr in sy cs us sy id 1 0 0 1034M 108M 99 0 0 0 0 0 17175 3814 30714 2 12 86 0 0 0 1034M 108M 0 0 0 0 180 0 17230 945 31461 1 15 84 1 0 0 1034M 108M 28627 0 0 0 0 0 17268 1734 30860 2 25 73 1 0 0 1034M 108M 54835 0 0 0 0 0 17192 7084 31114 5 36 59 0 0 0 1034M 108M 52911 0 0 0 0 0 17300 3508 31402 3 32 65 ... Thanks for the helpful message. So my issue seems to be resolved. I've been too naive not to recheck an old observation on a contemporary versions. Even "/usr/sbin/arp -an" some time ago (pre 8.0?) affected the cpu load somewhat. Now even arp -an running continuosly seems to have little impact on it. Thanks, Ivan _______________________________________________ [email protected] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-net To unsubscribe, send any mail to "[email protected]"
