linux# netstat -rn or bird> route show strangeness

[Arjan Filius]

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 at snxbgpprd01:~# time netstat -rn |wc -l
318259

real	0m1.296s
user	0m0.950s
sys	0m0.390s

root at snxbgpprd02:~# time netstat -rn |wc -l
318263

real	0m9.079s
user	0m6.720s
sys	0m2.500s


with linux ip route:
root at snxbgpprd01:~# time ip route show |wc -l
318267

real	0m1.152s
user	0m0.490s
sys	0m1.040s


root at snxbgpprd02:~# time netstat -rn |wc -l
318263

real	0m9.079s
user	0m6.720s
sys	0m2.500s


and with bird:
root at snxbgpprd01:~# time echo "show route"  | birdc | wc -l
1229972

real	0m3.315s
user	0m0.510s
sys	0m0.230s

root at 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 at xs4all.nl