Re: [Isis-wg] Requesting WG adoption of draft-baker-ipv6-isis-dst-src-routing-06

[David Lamparter <[e-mail]>]

Hi Chris,

On Thu, Oct 27, 2016 at 12:51:19PM +0000, Chris Bowers wrote:
> I have two main comments/questions on this.

Useful feedback, Thanks!

> 1)  If I understand correctly, one reason to use existing multi-topology routing mechanisms for 
> this is the requirement to support deployments where only a partial subset of 
> routers support source address dependent routing.  Ideally, this would allow an enterprise site to
> start by upgrading a subset of routers to support SADR, starting at the site egress routers.  

Yes, that's the intent.

> However, this statement from section 2.3 seems to imply that all of the routers in the enterprise network
> would need to support multi-topology even if they don't support SADR.  

No, non-MT routers would be ignorant of the MT information and would not
see any of the SADR routes -- which is exactly the intended goal.  I'll
update the draft to say this very clearly.

>    As this compatibility mechanism is not considered optional, M-ISIS
>    MUST therefore be implemented for supporting the protocol outlined in
>    this document.  Even installations that previously used only MTID 0
>    (i.e.  no M-ISIS) would need to start using MTID TBD-MT0.

Ah, I see where I misworded that.  It needs to say "need to start using
MTID TBD-MT0 *for transporting SADR routes*".

Sidenote:  at some point, the draft allowed non-usage of MT for
greenfield setups (all routers with SADR support and everything in zero
MTID / no M-ISIS used at all), with homenet in mind for that.  I'm now
thinking that was a stupid idea.  However, there might be leftover weird
wording in the draft from that.

> It would be good to clarify this statement.  I interpret it to mean that all prefix advertisements
> to use TLV#237, so that even routers that are currently using TLV#236 to advertise IPv6 prefixes
> would need to start using TLV#237 instead.  But it is not clear what MT-ID a router not supporting
> SADR should use when advertising a prefix in TLV#237, since it can't use MT-ID=0.

No -- all non-SADR routing information stays in MTID 0/TLV 236 (or
2/237, if you're using separate IPv4/IPv6 MT topologies).  There is no
change to non-SADR routes, and by using a separate MTID for SADR, they
will be routed "around" as non-participants in the SADR topology to
ensure proper SADR operation.  (That's the core idea why MT is used
here.)


As a nice side-effect, the operator also has some good control here in
feeding information to non-SADR systems.  For example, they can
advertise non-SADR default routes at one or more selected SADR routers,
so traffic towards the internet flows towards the "SADR subdomain",
where it can then be SADR-routed to the proper exit.

(MT makes sure SADR-routed traffic, once in the "SADR subdomain", will
never be routed back towards non-SADR routers.  They're not in the MT
topology, thus not in SPF, thus not in the shortest path used for SADR
routes.)

> In any case, it would be good to clarify overall if and how this solution achieves the goal of 
> requiring only a subset of routers to be upgraded from a basic non-MT deployment of ISIS.  
> As part of this, it would be good to explain how TLV#222 and TLV#229 (the other multi-topology
> TLVs) are used in the partial deployment case.
> 
> 2)  For addressing PA multi-homing for IPv6 w/o NAT in existing enterprises,  I would 
> guess that >90% of enterprises running a link-state routing protocol are running 
> OSPF as opposed to ISIS.  Is the expectation that enterprises will switch to ISIS 
> in order to address this problem?   I would like to better understand what use cases this work 
> is targeted at from a practical point of view.

There is/was a companion OSPF draft:
https://datatracker.ietf.org/doc/draft-baker-ipv6-ospf-dst-src-routing/
which we need to revive I guess.  The OSPF version has the additional
constraint that it is built on another draft in progress:
https://datatracker.ietf.org/doc/draft-ietf-ospf-ospfv3-lsa-extend/
which led to (my personal) decision to focus on the IS-IS draft and "fix
OSPF later."  Sorry, OSPF ;)


Updating the doc -- also planning to explain some protocol-agnostic MT
considerations in the rtgwg document -- expect another mail from me
soon (with a diff attached),


-David