[j-nsp] MX RE how fast is slow

[Jared Mauch]

On Thu, Sep 08, 2011 at 05:48:41PM +0300, Saku Ytti wrote:
> On (2011-09-08 15:30 +0100), Mike Williams wrote:
> 
> > Sorry, the commit time is fine, couple seconds tops.
> > 
> > # show | display set | count
> > Count: 1057 lines
> > 
> > After that the RE goes off and and runs the 700k routes it has already got in 
> > inet.0 through the policy, "repopulating" the 2 other RIBs. It's this which 
> > takes the time.
> > 
> > The fact it takes several minutes to run is only a moderate annoyance.
> > Especially now others have confirmed that the MX80 really isn't that fast at 
> > all.
> 
> Hmm, could you during this operations run 'top' in CLIa and couple times 'show
> chassis routing-engine' (and maybe even 'show chassis tfeb' just in case) just
> to see if what you're doing is CPU bound. 
> 
> I'm not entirely convinced that all blame can be put on RE, maybe JNPR has less
> optimized code for PPC than Intel, some timing issues etc.
> RSP720 has less beefy CPU but feels lot snappier with RIB.

	When you said repopulating above, is this shown as a slowly updating
or flushing of the krt queue?

	- Jared

-- 
Jared Mauch  | pgp key available via finger from jared at puck.nether.net
clue++;      | http://puck.nether.net/~jared/  My statements are only mine.