[j-nsp] MX RE how fast is slow

[Saku Ytti]

On (2011-09-08 16:36 +0100), Mike Williams wrote:
 
> I've done "show chassis routing-engine" many many times.
> It spikes User to >90% for a short while (a minute or so), then stays >90% 
> kernel until it's finished. There are occasional spikes in User load, but 
> only very brief, I figure that's it dealing with a peer/snmp/jflow/etc.

Yeah, looks like nothing you can do to make it better. You could while doing it
do as Jared suggested and check 'show krt queue', but I think this is fully
expected result today in MX80.
No clue what is the root cause, is pq3 RE inherently so much slower than intel
RE or is there some implementation differences in JunOS slowing ppc
implementation.

Another datapoint, 500BGP sessions, each sending 1000 prefixes, takes bit over
10min to reach 0% packet loss, (.5M routes, receiving another .5M from RR for
aggregate of 1M)

-- 
  ++ytti