[AusNOG] Routing in NX-OS vs IOS

[Lincoln Dale]

g'day,

I was off the list for a while (change of email address) but just saw this
now & thought i could add some clarity on the rationale now that i'm
resubscribed.

by way of background, I had a thing or two to do with NX-OS and Nexus.


Ben Byer wrote:
 > I've been picking through the reference docs, and the useful comparison
notes on the Cisco wiki,
 > but - I just don't understand the thinking behind the following:
 > "When you change a route map, Cisco NX-OS hold all the changes until you
exit
 > from the route- map configuration submode. Cisco NX-OS then sends all
the
 > changes to the protocol clients to take effect."

It is predominantly because the code is modular, not monolithic and because
we figured that if you were making changes to a series of route-maps that
you would want them to be atomic either the 'before' or 'after' stage, not
an indeterminite stage somewhere inbetween.

 > ....but wait! I have multiple protocol clients using the same route-map
- and maybe
 > I *don't* want the change to be propagated to all of them at once as
soon as I
 >  type "exit". The way IOS allows you to change a route-map, then cause
it to
 >  take effect on a per-peer basis via a soft clear, is operationally
*useful*.

You can still use soft reconfiguration inbound if you still with. its still
there and NX-OS is (intentionally) no different to IOS in this regard.

Note that soft reconfiguration inbound is disabled by default, because it
basically means you must keep a shadow copy of every route/prefix received
from a peer just so you can iterate over them again (re)applying route
policies.
For some people with large numbers of peers and prefixes thats a ton of
RAM, even on systems with 8GB RAM on the Supervisor.

 > To make it worse, the reference docs for the 7k series say: "If you make
changes to
 >  a route map that is used by a client, you must exit the route-map
configuration
 > submode before the changes take effect in the client. The route-map
changes
 > are not propagated to its clients until you exit from the route-map
configuration
 > submode or 60 seconds expires since entering the submode."
 >
http://www.cisco.com/en/US/docs/switches/datacenter/sw/5_x/nx-os/unicast/command/reference/l3_cmds_r.html#wp1542709
 > ...great - so now whether or not I type "exit", 60 seconds after I typed
a command
 > it will take effect...?

Its sort of as above - we (historically) had people whinging about changing
route-maps and having some prefixes leak through in the timeframe while
they were making changes, so a determination was made to exhibit either the
'before' or 'after' and make the operation appear atomic.
What we also saw was some people then still sat the
'switch(config-route-map)#' prompt and wondered why their policy wasn't
being applied - so did the 60 second timer thing.

We made sure it was documented.
Just goes to show you can't please all people.

 > And this behaviour is documented only for the 7k, not the 5k?

N5K behaves the same. Its the same NX-OS code.


Glen Turner wrote:
 > Does that behaviour still occur if you say "configure session ..."? In
any sane world nothing would change until the "commit".

Alas, 'config session' still only applies to ACLs and QoS. Intent is that
it will apply to more over time, but AFAIK hasn't happened yet.


cheers,

lincoln.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ausnog.net/pipermail/ausnog/attachments/20111205/18046382/attachment.html>