Gentz, an update. it's not bird related. even a: linux# time dd if=/dev/zero of=/dev/zero count=10000000 has a big difference in time on both machines, even when bird is not running. so don't know yet what it is , but not bird at all. Sorry for disturbing you. Regards, Arjan Filius On Thu, 3 Jun 2010, Arjan Filius wrote:
Hello,
today i noticed huge difference between our 2 identical bgp ubuntu 10.04 LTS with burf 1.2.3 (ppa) x64 routers. (hardware DL380G6 12G ram X5560 @ 2.80GHz) systems are still idle in having no ip traffic yet, except for the bgp sessions and "random" other traffic.
root@snxbgpprd01:~# time netstat -rn |wc -l 318259
real 0m1.296s user 0m0.950s sys 0m0.390s
root@snxbgpprd02:~# time netstat -rn |wc -l 318263
real 0m9.079s user 0m6.720s sys 0m2.500s
with linux ip route: root@snxbgpprd01:~# time ip route show |wc -l 318267
real 0m1.152s user 0m0.490s sys 0m1.040s
root@snxbgpprd02:~# time netstat -rn |wc -l 318263
real 0m9.079s user 0m6.720s sys 0m2.500s
and with bird: root@snxbgpprd01:~# time echo "show route" | birdc | wc -l 1229972
real 0m3.315s user 0m0.510s sys 0m0.230s
root@snxbgpprd02:~# time echo "show route" | birdc | wc -l 896174
real 0m20.018s user 0m2.050s sys 0m0.920s
so asking the route table takes at least 7 times more on the second rtr, while being exact the same hardware, OS and software. only other bird config due to IX connections. with "debug protocols { states, routes , filters};" i didn't notice anything special, and both routers seem more or less equal.
with these times, i noticed that it seems that the cpu usage of bird rizes to 100% as i do a netstat -rn on the linux prompt.
It looks like a route table access contention
At one time, this morning, the netstat command exceded the 3 minutes, and i finally killed it.
Has anyone same experinces, or knows perhaps what is going on? i want to start with oprofile to have a more closer look what is going on, and where that cpu time is consumed.
Regards,
-- Arjan Filius mailto:iafilius@xs4all.nl