[c-nsp] LAC/LNS Routers - 7200 EOL

Phil Mayers p.mayers at imperial.ac.uk
Thu Dec 1 18:06:58 EST 2011

On 12/01/2011 10:48 PM, Phil Mayers wrote:
> On 12/01/2011 09:10 PM, Chris Adams wrote:
>> Once upon a time, sthaug at nethelp.no<sthaug at nethelp.no> said:
>>>>> While I agree that it's not optimal, but is it atypical? Isn't
>>>>> JunOS the
>>>>> same? All the important things running in single flat process,
>>>>> which has
>>>>> its own scheduling and memory management. Unix in the background
>>>>> being just
>>>>> an afterthought, really a way to bootstrap it all up.
>>>> No, there are a bunch of separate Unix processes on JUNOS handling
>>>> different things.
>>> rpd. 'Nuff said.
>> That's one process that handles a bunch of stuff (but far from
>> everything); that's hardly a "single flat process, which has its own
>> scheduling and memory management".
> http://www.juniper.net/us/en/local/pdf/whitepapers/2000264-en.pdf
> """
> The routing protocol process daemon (RPD) is the most complex process in
> a Junos OS system. It not only contains
> much of the actual code for routing protocols, but also has its own
> scheduler and memory manager.
> """

Grumble Ctrl+Enter Grumble...

To continue: certainly rpd doesn't contain everything, and the JunOS 
architecture is an improvement on classic IOS (in so, so many ways) but 
it is fair to say that rpd does operate to a very large extent as a 
micro-OS, and does indeed have its own scheduler/memory management - as 
above, Juniper describe it this way.

The paper is worth reading, along with the Junpier argument in favour of 
this model (tl;dr - performance). In my (fairly limited) experience it 
works pretty well, better than classic IOS. But I've never had an "rpd" 
crash, and I imagine claims of "modularity" ring hollow for those who 
have, and have suffered outages as a result.

More information about the cisco-nsp mailing list