Minutes IETF99: lisp

Meeting Minutes Locator/ID Separation Protocol (lisp) WG
Title Minutes IETF99: lisp
State Active
Other versions plain text
Last updated 2017-08-09

Meeting Minutes

   LISP WG Minutes
Monday 17 July 2017 1:30pm Ð 3:30pm

- Administration
- Blue Sheets
- Agenda Bashing
- Status reports for WG drafts

Update on Draft-ietf-lisp-rfc6830bis & Draft-ietf-lisp-rfc6833bis
Ð Albert Cabellos

Comment regarding the order of documents need to make advancement in these docs.
Joel: Have a timer running on that one, because we have the code points
therefore we need to get that one moving. Luigi: Invite everybody to review the
code data points and put out all the comments so we can move these forward.
Dino: Do you have recommendations on the ones which are currently about to be
published to have pointers on these document . They currently point to 6830 and
6833? Joel: No. Let them move. Dino: Ok one of the problems IÕve had is that
any new internet draft should point to these drafts? Joel: New work we do
should be pointing to these drafts not to the existing.

LISP YANG Model - draft-ietf-lisp-yang-05
Ð Fabio Maino

No comment or questions

LISP Predictive RLOCs - draft-ietf-lisp-predictive-rlocs-00
Ð D. Farinacci/ P. Pillay-Esnault

Question from Fred Templin on whether the packets get silently discarded
Explanation on the pre-fetching of the source by using the packets that were
dropped on the floor so making use of this information Victor: There is
directionality in the examples.  How about if you are driving in a car
east-west and now you go back home in opposite direction. Any thoughts how you
can actually express that with your map request and get the right one? Dino:
Yes you can two mappings, one for each direction or you can use the circular
one or the back and forth and replicate to all of them. There is all kind of
things you could do Padma: We can think of these RLOCS being added by a third
party that has more information. It is not necessarily the EID end-point itself
registering all this info. Dino: Absolutely. I didnÕt make it clear who is
registering the RLOC. You may have assumed if you know lisp how LISP works that
the endpoints or the XTRs are registering the RLOC but we very much thought
that it would be a third party. Victor: That makes sense. It could be for
consideration that this would be extremely useful in the 3d space rather than a
linear space. Have you thought so? Dino: I could say that my intersection
example was 3D in a 2D slide. You can have a plane that moves in 3 D and you
would have more nested RLEs. People have asked f we can use this in airplane
applications where you donÕt know exactly where you are going. If you know the
geospace where the plane is you can just replicate to everything in that space.
Padma: Regionality has something we can use. For example if you pass and get
detected by radar then may be something like for the next 5 km you will not be
changing intersection. Dino: Air traffic control could be the SDN controller.
Victor: So that interface is going to be key Dino: Yes. Fred do you have any
comments wrt to not knowing your direction precisely? Fred Templin: I do not
see why it would not work for a plane in flight of course civil aviation type
Luigi: There is always a flight path that the pilots head through and before
the information is known before. Dino: It is when they deviate time to time ..
Padma: Another example that we used was trains and they will not deviate from
the tracksÉ Dino: Yes they are nailed to the ground. Luigi: Just a comment on
Slide 14 where you say that Random is when the EID is É Dino: What we meant is
that we replicate from a through D and we donÕt know the direction, in which
the EID is going in, Padma: Another interesting case is that we say that even
thought you have a long list and you can replicate to all. But in cell phone
you are handing over from one cell tower to another, so at one point you could
restrict that only to the interesting points. All this implies that we have
more machinery to update. This is very doable as they have the information
anyway. Dino: The way I was going to implement it is to not only send it to B.
Discussion on how list pruning  É. That what is really  important.. Victor: 
IsnÕt this the case where PadmaÕs comment where the information on where it
actually is and that is why you sent to B? Dino: If it is not going to move
then you just use regular mechanisms. May be we should prune Random and
unpredictable from the list. There is a requirement from somebody else and they
are not here to defend it.

Dino/Wolfgang Reidel: Discussion on use of geo-coordinate and how the third
party may be used to update the list of rlocs. Use of GPS.  The map request
will take to much time. Dino/Wolfgang/Padma: When a physical move occurs and if
we use the normal process we are saying it takes too much time. Here we are
bypassing all this. It is an IP problem not a spacial. The closer we keep the
replication to the destination. The missing piece is that the moving entity
must directly tell the replicator. Dino/Padma: Discussion on late and
pre-fetched binding possibility .Use of replication closer to the destination
Wolfgang: how may encapsulation on top of this one. I do not care if it is a
bit larger. Dino: So you are for making the header largerÉ you need to convince
millions of people but agree it might be worth trying out.

LISP EID Anonymity - draft-farinacci-lisp-eid-anonymity-02
Ð D. Farinacci/ P. Pillay-Esnault
Victor Ð What triggers the change of the EID to next?
Dino: It is the policy in the source, whenever it wants to. May be the source
will use an EID per TCP session. Albert Ð question regarding the encryption how
this can be done with the anon-eid. The signaling should be done without of out
of band from lisp Dino- we are working on a document to show  how to achieve
this in lisp. Luigi Ð Concern that fast change of eid can fill up the cache
Dino: There is nothing in the text but what we can do is rate-limiting. May be
we should look at scale issues. We need to figure out where to document it.
Fabio Ð Consideration on the inference on what the mapping system can do on
anonymity. I guess you Consideration for the security section Request for wg
doc Ð support in room. No objections

Vendor Specific LCAF - draft-rodrigueznatal-lisp-vendor-lcaf-00
- F. Maino

Dino Ð OUI should be only from the IEEE list as there are some virtualization
companies who assign OUI on their own

Request for wg Adoption Ð Support in the room. No objections.

A simple BGP-based Mobile Routing System for the Aeronautical Telecommunication
Network Ð F. Templin

Fred: The data plane is encapsulation of ipv6 over ipv6. We could use the LISP
encapsulation. Arrow uses the Generic UDP encapsulation or GUE but arrow could
use the LISP encapsulation. Padma: I know you said you do not want to drop any
packets. How do you optimize your routes and forwarding, or you do not care
about it? By not sending it on an optimize route you are not sure that you will
receive these packets. Fred:  In stable state your bgp peering between these
core ASBR are bring maintained by TCP keepalives. Rely on TCP to maintain the
link up. We are worried about initial packet loss as this may be the last
packet that plane ever sends. From what I understand about lisp is that the ITR
holds on the packet until it gets a reply back and then sends the packet. First
of all when it gets the map reply back, it canÕt know without testing it first
that the path the ITR to the ETR is a working path. It cannot know that the
forwarding path works. Dino. You said that the applications use TCP connections
will be up ahead of time and therefore those packet drops happened on syn
packets thatÕs going to happen way before the data is going to need to go over
it. Fred: The TCP connections do not go to airplane. The TCP is only between
the BGP core routers. Dino: Are they set up ahead of time? Fred: Yes same as
BGP Peering. Padma:  So you are relying on the BGP keepalive to ensure that the
packet makes it. The moment you sent the packet down from the plane you have no
guarantee that there is anything up. Fred: The stability of BGP is going to be
based on the fact that the ASBR are not moving. When an EID prefix come to the
xtr the first time and it needs to inject that EID prefix in the BGP routing
system and then get propagated to the core ASBR but they do not go further. So
very few nodes need to be updated. Padma: Yes this reminds me of OSPF. But you
are have no control whether these connections can go down. Well there is no
guarantee. Fred: you are right; this is the same thing that happens in a BGP
network. Fred/..: Discussion on neighbor discovery in arrow relying ipv6
neighbor discovery. Albert : Regarding the first packet loss,  before we
populate the map-cache is that we send it to the proxy ITR that is equivalent
to the router in slide. Fred:.. router has full knowledge of the IDS topology.
ICAO 9896 doc describes the ATN. has a sub group mobility and we meet on 3
times per year. They want to work on a solution for this year. FAA, European
and other aviation orgÉ Fred: Lisp does not run on the airplanes but only on
the ground. Wolfgang/Fred: Some discussion on a simple solution on the airplanes

Victor: Are you  adding mobility semantics to BGP?
Fred: It is like layer 2 mobility and it does not get propagated into BGP. But
it go from one stub AS to another Stub AS then it gets propagation into BGP. 
Macromobility and micromobility if you will. Victor: Are you proposing to use
community attributes for the macro level? Joel: So nothing special for mobility
Fred: discussion on slides for plane having 2 bgp routes. <É> With the
ground base lisp will have to update the MS. Discussion on BGP scale on million
to ten thousands of planes Padma: the question is really about the converging
time of bgp. Fred: The number of routers on the core of is limited to the order
of ten. Joel: They have a very constrained environment and they are designing
it for a very specific purpose. Victor: Do you see the pink cloud as 10 routers
over the world? Fred: Good question. The pink cloud is what we call the inter
network. It is subdivided into regions. Whether it would be partitioned or not
is all up in the air. Concerns regarding regional partition depending on
countries USA, Europe or China. Victor: a very interesting use case Luigi: How
to make the different mapping system talk to each other? They have the
possibility to handle specific EID requests. Wolfgang: Discussion on live-live