Re: [Gen-art] Gen-ART LC review of draft-ietf-pim-group-rp-mapping-07.txt

["Andy Kessler (kessler)" <[e-mail]>]

Ok, we have made progress. 

These are the suggested changes we need to make: 

1. In Section 3, Existing Algorithm. Add a new sentence to
   the end: 

   The full benefits of the new algorithm will not be realized 
   until it is widely deployed. 

2. In section 5, Common use cases, first section "Default static
   Group-to-RP mapping with dynamically learned entries", add
   a sentence to the end: 

   This is not an issue for IPv6 Multicast address ranges. 

3. In section 8, "Clarification for MIB Objects", I'm not 
   sure what we have to say to be sure its understood that this
   is a normative reference. Suggestions ? 

4. Section 5, Common use cases, remove the subsection "More use 
   cases". 

Bharat, please spin a new version with these changes when we 
resolve number 3 above. 

Thanks, 

Andy

-----Original Message-----
From: Brian E Carpenter [mailto:[e-mail]] 
Sent: Saturday, December 18, 2010 11:54 AM
To: Andy Kessler (kessler)
Cc: [e-mail]; General Area Review Team; Bharat Joshi; Stig Venaas; Mike McBride (mmcbride)
Subject: Re: Gen-ART LC review of draft-ietf-pim-group-rp-mapping-07.txt

Hi Andy, thanks for responding so promptly. A few comments in line:


On 2010-12-18 18:36, Andy Kessler (kessler) wrote:
> Hi, 
> 
> Thank you for the review. 
> My comments are below annotated with Andy>
> 
> Andy
> 
> 
> I am the assigned Gen-ART reviewer for this draft. For background on
> Gen-ART, please see the FAQ at
> <http://wiki.tools.ietf.org/area/gen/trac/wiki/GenArtfaq>.
> 
> Please resolve these comments along with any other Last Call comments
> you may receive.
> 
> Document: draft-ietf-pim-group-rp-mapping-07.txt
> Reviewer: Brian Carpenter
> Review Date: 2010-12-18
> IETF LC End Date: 2010-12-23
> IESG Telechat date: 
> 
> Summary: In good shape, a few issues.
> --------
> 
> Major issues:
> -------------
> 
> 3.  Existing algorithm
> ...
>    The algorithm defined in this document updates algorithm defined in
>    PIM-SM ( Section 4.7.1 in [RFC4601]).  The new algorithm is backward
>    compatible and will produce the same result only if the Group-to-RP
>    mappings are learned from a single mapping source.
> 
> A logical implication of the last sentence is that if some of the Group-to-RP
> mappings are learned from more than one mapping source, the new algorithm might 
> not produce the same result as the existing algorithm. Is that indeed the case,
> and if so, don't we have a problem until all PIM-SM routers have been updated?
> 
> Andy> Most vendors today do not officially support deployments with multiple
> Andy> dynamic routing protocols. They probably have inconsistent results now. 
> Andy> This algorithm will standardize the results for those vendors. It will
> Andy> be more consistent when all vendors do support the new algorithm. 

OK. Maybe you could add a sentence to make it clear that adding the new
algorithm doesn't fix the problem completely until it's universally deployed.

> 5.  Common use cases
> 
>    o  Default static Group-to-RP mappings with dynamically learned
>       entries
> 
>    Many network operators will have a dedicated infrastructure for the
>    standard multicast group range (224/4) and so might be using
>    statically configured Group-to-RP mappings for this range.  
> 
> Please indicate whether this issue also applies to IPv6.
> 
> Andy> There is no issue with IPv6. IPv4 has some older baggage because
> Andy> the 224/4 addressing scheme has been around since before there was
> Andy> SSM and BSR. IPv6 Multicast does not have that history. 

OK. For someone ignorant of the history, it might be worth adding
"This issue does not arise for IPv6."

> 
> 
> 
> 6.  Proposed algorithm
> ...
>    10.  From the remaining set of Group-to-RP Mappings we will select
>         the RP with the highest IP address.  This will serve as a final
>         tiebreaker.
> 
> Please define "highest". Does it mean "greatest when considered as an
> unsigned integer"?
> 
> Andy> This should be the same method that is used to pick 
> Andy> the highest IP address in many protocols. One I know well is
> Andy> the way that a designated router is selected in RFC 4601 in
> Andy> section 4.3.2. The main part of the function says: 
> Andy>      a.primary_ip_address > b.primary_ip_address
> Andy> To me that has always meant you compare the first byte,  
> Andy> then the 2nd, the 3rd and then the 4th. 

and the 5th, etc, for IPv6 ;-)

> Andy> I don't know where that would be defined. Maybe someone else knows. 
> Andy> It is not defined in RFC 4601. 

I have no idea either. Maybe it's a term of art in routing protocols,
so I won't insist; I was just looking at it from the viewpoint
of a naive implementor.

> 8.  Clarification for MIB Objects
> 
>    When a Group-to-RP mapping entry is created in the
>    pimGroupMappingTable in the PIM-STD MIB[RFC5060], it would be
>    acceptable to have an entry with an RP with address type 'unknown'
>    and a PimMode of Dense Mode or SSM.  These entries would represent
>    group ranges for Dense mode or SSM.
> 
>    Also all the entries which are already included in the SSM Range
>    table in the IP Mcast MIB would be copied over to
>    pimGroupMappingTable.  They would have a type of configSSM and an RP
>    with address type 'unknown' as described above.
> 
> Is this section intended to be a normative update of RFC 5060, or is
> it just commentary? The phrase "it would be acceptable" almost sounds
> like a normative MAY.  
> 
> Also, shouldn't it be 'unknown(0)' in the context of the MIB? (This
> applies in three places in the draft.)
> 
> Andy> At some point we received comments and requests to include all
> Andy> the group mappings in this table - even for Dense Mode or SSM
> Andy> which do not have an RP. We are merely stating that it OK to 
> Andy> do that and use the RP address and type (of the RP) of unknown. 
> 

Yes, it makes sense, but please make it very clear whether you
are writing normatively or not.

> 
> Nits:
> -----
> 
> 5.  Common use cases
> ...
>    o  More use cases
> 
>    By no means, the above list is complete.  Please drop a mail to
>    'authors' if you see any other use case for this.
> 
> This seems a bit strange to appear in the final RFC; at the least it needs rewriting.
> 
> Andy> I agree. That was for the draft and should be removed. Should we
> Andy> just remove the whole section ? Or change the wording to somehow
> Andy> say there are more use cases that will benefit from this algorithm ? 

Either way is fine for me.

> 
> The checker says:
> 
> == There are 1 instance of lines with multicast IPv4 addresses in the
>      document.  If these are generic example addresses, they should be changed
>      to use the 233.252.0.x range defined in RFC 5771
> 
> Andy> We have discussed this before. We need to use the addresses that in the
> Andy> examples. 239/8 is the Administratively Scoped Block as defined in RFC 5771.
> Andy> Its the same thing for 239.100.0.0/16 in section 5. We were told to add
> Andy> the comment that says its an administratively scoped multicast address
> Andy> range. Using GLOP will not make the same point we are trying to explain. 

Fair enough. Don't be surprised if someone in the IESG asks the same question.

   Brian