Hi Roman, Alexy,

Thank you for your comments.

> ----------------------------------------------------------------------
> DISCUSS:
> ----------------------------------------------------------------------
> 
> ** Section 4.3.  Do all the candidates need the Area Proxy System Identifier
> TLVs need the same system identifier?


I’m not sure I understand the question. Let me prattle on a bit and please let 
me know
if I do not address the question.

Vanilla IS-IS requires that each node in the routing domain have a unique 
system identifier.
This is unchanged.

Area Proxy extends this by requiring an additional system identifier for the 
Area Proxy. This is 
the Area Proxy System Identifier. This is normally advertised by the Area 
Leader so that all 
Interior Nodes know the value and it’s used by Interior Edge Nodes.  It’s a 
good idea
to have multiple candidates for Area Leader for resiliency, and having them all 
configured
with the same Area Proxy System Identifier is strongly preferred out of 
consistency. However,
it is NOT required. It is not an error for there to be multiple different 
candidate Area Proxy System
Identifiers.  In fact, this situation might happen if someone has decided to 
change the
Area Proxy System Identifier and is in the midst of reconfiguring part of the 
routing domain.

This does not cause confusion because Area Leader election is going to elect a 
single leader
and all systems will use the Area Proxy System Identifier that the leader is 
advertising.

Yes, if there is a change in leader, there may be a transient change in the 
Area Proxy System
Identifier. This would cause a set of adjacency flaps, just as changing any 
regular system identifier 
would. 


> -- Section 4.2 says “For consistency, all Area Leader candidates SHOULD be
> configured with the same Proxy System ID, Proxy Hostname, and any other
> information that may be inserted into the Proxy LSP.”


The rationale is similar to the above. Is there a separate question?


> -- Section 4.3.1 says: “All candidates advertising the Area Proxy System
> Identifier TLV MUST be advertising the same system identifier.”


I will relax this to a SHOULD.


> The first statement suggests that consistency might not always be required, 
> but
> the second statement is clear that it always must be the same identifier.
> 
> ** Section 8.  The following statement doesn’t appear to be accurate.
> 
> 8.  Security Considerations
> 
>   This document introduces no new security issues.  Security of routing
>   within a domain is already addressed as part of the routing protocols
>   themselves.  This document proposes no changes to those security
>   architectures.
> 
> -- Aren’t a few the filtering activities described in Section 5.2
> security-related?


No. These are key for ensuring the benefits of Area Proxy, most especially 
scalability, 
but if they are violated, it is not necessarily catastrophic.  

Some examples:

- If the L1 LSP of an Inside Router is leaked outside of the area, then it 
would be a 
protocol error, but other routers should recognize that they are not part of 
that area and 
ignore the LSP.

- If the L2 LSP of an Inside Router is leaked outside of the area, then it 
would be accepted
by other routers, but it would have no two-way adjacencies with anything else 
in L2 and
effectively be disconnected from the topology.

- Incorrect PSNP or CSNPs would prompt the receiving system to send PSNPs for 
Inside LSPs, but this would not per se cause problems.


> -- What are the relevant pointers to IS-IS security considerations?


AFAIK, there is no overall document for IS-IS’ security architecture. The only 
pointers I can
suggest are to RFC 5304 and 5310.  I will happily add references to these if 
folks feel that’s helpful.


> ----------------------------------------------------------------------
> COMMENT:
> ----------------------------------------------------------------------
> 
> Thank you to Alexey Melnikov for the SECDIR review.
> 
> ** Section 4.3.1
>   However, before withdrawing the Area Proxy
>   System Identifier, an implementation SHOULD protect against
>   unnecessary churn from transients by delaying the withdrawal.  The
>   amount of delay is implementation-dependent.
> 
> Are there any guidelines on how the delay period should be computed?


I don’t have any specific suggestions. My implementation practice is to
apply binary exponential backoff, with some ridiculous upper bound, but the
specifics are well outside of the boundaries of a protocol spec.


> ** Section 4.4.4.
>   An entry for a neighbor in both the IS Neighbors TLV and the Extended
>   IS Neighbors would be functionally redundant, so the Area Leader
>   SHOULD NOT do this.
> 
> Under what circumstances would this advice be ignored (i.e., why not a MUST)?


This is not a MUST because it’s a redundancy, not an error.


> ** Section 4.4.5 and 4.4.6 describe various behavior where fields “SHOULD” be
> copied.  What governs the choice of not copying these fields?  Would operators
> be impacted in troubleshooting or even operations if different implementations
> applied this advice differently?  Would it be up to local configuration? Later
> sections in 4.4.* also have “SHOULD” behavior as well where the reason for not
> following the behavior is not clear.


We intentionally left this liberal to allow for a combination of intelligent 
defaults and
local configuration. For example, if the Area Leader is configured for SRv6 and 
it
sees other nodes advertising SRv6 information, then it might be smart to 
automatically
propagate SRv6 information. But we don’t want to say ‘MUST' because not all 
areas
may be using SRv6 and the network operator might want to filter out SRv6.  Yes,
obviously different implementations that made different choices would have 
different
effects on L2.

The rationale for other information is similar. For scalability reasons, we 
want to
default to copying as little as possible.

Regards,
Tony

_______________________________________________
Lsr mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/lsr

Reply via email to