[j-nsp] iBGP convergence time

[Richard A Steenbergen]

On Thu, Feb 01, 2007 at 09:07:05AM +0100, Gniewko wrote:
> It takes about 3 minutes for m7i_2 to receive 200k of prefixes. But, as
> soon as I change AS to my public number (both sides of course) have to
> wait almost 20 minutes. During this long period, CPU on m7i_2 is quite
> busy.
> Could You please give me a hint what could be the source of this
> behaviour?

You know I haven't had any free time to run real tests or anything, but I 
noticed a significant increase in BGP convergence time when upgrading from 
JUNOS 7.2/7.3 (and I think some 7.4) to 7.6. When you make a policy 
change, the routes take several minutes (from 2 to 7) to install. If you 
do a show route you can see the new routes sitting in + and the old routes 
sitting in - for minutes, RPD is spinning its heels at 100% cpu, and the 
packets continue to forward over the old path while it is processing.

Yes I'll admit that these are all RE-2.0's that have long outlived their 
design lifetime, and have horrible memory states like:

Mem: 489M Active, 125M Inact, 87M Wired, 22M Cache, 69M Buf, 25M Free
Swap: 768M Total, 28M Used, 739M Free, 3% Inuse

(Not actively swapping in/out of course), but again failing real 
scientific testing all the new code seems to be significantly slower. At 
first I had blames this on a particularly nasty build of 7.6 with other 
bugs, but the other day I upgraded a similarly loaded box to 8.1 and 
noticed the same issues. Anyone else seeing the same thing? Short of 
someone slipping in some for(;;) screw_customer(); code to motivate RE 
upgrades I can't explain it. :)

-- 
Richard A Steenbergen <ras at e-gerbil.net>       http://www.e-gerbil.net/ras
GPG Key ID: 0xF8B12CBC (7535 7F59 8204 ED1F CC1C 53AF 4C41 5ECA F8B1 2CBC)