Hi bird users
Some time ago 've experiencd some seggfaults issues with bird comiled
without -fno-strict-aliasing, -fno-strict-overflow
After addition of -fno-strict-aliasing, -fno-strict-overflow i'm
observing strange loadavg peaks on my router:
http://www.sq9mev.info/bird_cflags_loadavg_change.png
The upgrade was done on wednesday and varied loadavg is clearly visible
in above graph.
I've used 1.5.0 provided by upstream
(ftp://bird.network.cz/pub/bird/redhat/bird-1.5.0-1.x86_64.rpm) with the
same result (varied loadavg).
Next thing i've tried was to cherry pick 54bb032d and 665b8e52 on 1.5.0
and compiling without -fno-strict-aliasing, -fno-strict-overflow
This binary causes varied loadavg as well.
I've noticed that values of peaks is correlated with kernel scan time
parameter. I'm using smaller then default (20s) value of 1s.
With scan time set to quite high value of 300s load 1minute load have
some peaks, but overall loadavg looks quite constant.
5 minutes scan time seems ok for me - i' not expecting a big amonut of
alien routes, and a few of them i have will probably be propagated
successfuly with async netlink notifications.
My question is - is such a varied loadavg normal with 10s or even 20s
scan time? I have several ful view bgp sessions:
birdc show route count
BIRD 1.5.0 ready.
1263256 of 1263256 routes for 582924 networks
I've compared stable and unstable load bird processes with perf tools:
http://www.sq9mev.info/stable_unstable.diff.txt
I've compared unstable load bird processes during low and high load periods:
http://www.sq9mev.info/unstable_near1_near0_perf.diff.txt
In "unstable load" versions there is noticable amount of
if_find_by_index samples, while it's not at all in "stable" version.
Is the varied load with lower kernel scan time price of saver list node
removal?
Is it more bird issue or should i just try to find the cause in my
enviroment?
--
A: Because it messes up the order in which people normally read text.
Q: Why is top-posting such a bad thing?
Regards
Bartosz Radwan
