ANIMA
WG, IETF-93 (Prague)
Autonomic Networking Integrated
Model and Approach
Chairs: Toerless Eckert &
Sheng Jiang
There
are two sessions (150+90 minutes) for the ANIMA WG:
Monday, Morning
Session I 0900-1130
Thursday,
Afternoon Session III 1740-1910
For
both sessions, the chartered work items have the priority.
-------------------------------------------------------------------------
Monday
(July 20th, 2015) 2.5-hour session:
9:00-11:30 Monday Morning session I, Karlin I/II
1. WG
Dash - 5 min
9:00
- 9:05, by co-chairs
No comment from the room.
2. ANIMA
Generic Signaling (Design team) - 35 min
9:05
- 9:40, by Brian Carpenter, draft-carpenter-anima-gdn-protocol
Brian C.: we suggest to rename the protocol name from GDNP
(Generic Discovery and
Negotiation Protocol) to GRASP
(GeneRic
Autonomic Signaling Protocol)? Two considerations:
easier to say than GDNP;
leaves scope for future extensions.
WG showed no objection on renaming it to GRASP.
Henk Birkholz: Regarding to
CBOR, you should look at CDDL whether
it's suitable for your
problem space.
Brian C.: Not sure if CDDL is quite finished yet. For JSON-like
notations, CBOR is a very
powerful tool. Whether people think this
efficiency loss is worth
paying the other advantages moving to
object-oriented approach?
Markus Stenberg: In general, I don't care about the end coding or even
transmitting, because I
assume these would not be very chatty things.
One observation of
efficiency is that it becomes a problem if you are
sending multicast because
you can't fragment them.
Brian C.: In our model, multicast only comes up in the context of
discovery. If we could find
a way that the discovery packets never
exceed 1500 bytes, I think
it will be OK.
Toerless: one question is whether we should work in datagram
environment or it's good
enough to work in a stream environment.
Even for unicast, the stream
environment like TCP/SCTP would make
life for larger objects like
intent a lot easier.
Markus S.: then I guess it means you're essentially not using multicast for
anything except for the
three primitives. You can do multicast stream?
But I can't.
Toerless: the idea is I think GRASP will do hop-by-hop application layer
multicast. One the layer-2
link, I do TCP/SCTP, I can still multicast
across multi-hops at the
application layer.
Who thinks we should explore the Object-Oriented approach in much detail?
Lots of hum.
Who thinks that will be a bad idea if we stick into the traditional TLV design?
No hum.
WG Chair solicited hum support for the signaling documents:
lots of hum supported; non
opposed
3. ANIMA
Auto Bootstrapping (Design team) - 35 min
9:40
- 10:15, by Max Pritikin, draft-pritikin-anima-bootstrapping-keyinfra
Michael B: a clarification question: this is not saying like for a constrained
node "I cannot do
security"; but it is saying "I do not want to do security".
(Max P.: correct.) So the
constrained node considerations are still there,
maybe there are devices that
cannot support for example the certificates.
Toerless: what's happen if there is neither a nonce nor data information,
what's the security then?
Max P.: you have a permanent ownership claim of the device.
Sheng: what's the security mechanism between domain bootstrapping
server and cloud server? Is
that solved in your draft?
Max P.: Yes. It depends on the PKI model of authentication.
Brian C.: you're using url as the redirection option,
and you're assuming
one of the pros is to run
over http. If you go this way, and we go
JSON/CBOR way, we're stating
the renders stuff is big enough to
process JSON and http. Just
want to check that assumption is one of
what the WG would actually
buy.
Max P.: we did make assumption http could be used by nodes. There is
some texts regarding to
JSON, we can revise it to TLV, based on
feedbacks in the WG.
Toerless: a suggestion on redirect versus full-config download.
Maybe a way to
get to the middle line is to
have a most simple YANG model for a redirect to
specify what we'd like to
see in Anima.
Max P.: that is to say, here is profile we think proper for bootstrapping
WG Chair solicited hum support for the secure bootstrapping draft:
lots of hum supported; non
opposed
4. Autonomic
Control Plane & Addressing - 30 min
10:15
- 10:45, by Michael Behringer, draft-behringer-anima-autonomic-control-plane
& draft-behringer-anima-autonomic-addressing
Toerless: (discussing slides showing goals): maybe have separate goal for 'robust'
independent of 'must
provide IPv6 connectivity'
Michael Richardson(via Jabber): Can ACP make network so complex that even
ACP can't guarantee (robustness)
we will shoot ourself in the foot (keep ACP
simple to be robust).
Michael B.: the ACP will add some complexity to the network for sure. We have
to
make sure it as simple as
possible. But there is added value; and if you add
value that usually comes with some
additional complexity.
Sheng: you depend on IPv6 (ULA), currently doesn't support IPv4; Generic should
mean IP version independent?
Michael B.: No. We should stick to one protocol for complexity reason.
Toerless: Generic is a very generic word (eg: suggest
to reconsider wording to be
more specific). The answer
to Sheng, it's not basically anybody who wants to
use the ACP could expect
exactly get his own personal choice of layer2/3
protocols (in the ACP), but
he will get the ability to for robust connectivity that
he (hopefully) doesn't have
to worry about addressing.
Juliusz Chroboczek: some
rumor that one of the simple routing protocols in
consideration is RPL. I
strongly recommend you to take a weekend off,
implement RPL from scratch
before commit using RPL as a "simple" routing
protocol.
Michael B.: I can only comment in our implementation that we're running RPL.
Henning Rogge: most routing protocols I think use multicast to talk with each
other, how does this work
get a concept of secure tunnels with most time the
tunnels are point-to-point.
Do you plan if the ACP detects 20 neighbors in the
link-local zone to build 20
tunnels and duplicate every multicast 20 times?
Michael B.: Indeed. The fundamental observation is short of secure multicast
which
is a complex thing. If you
want to know who you are talking to, you have to have
a one-of-one security
association with your neighbors on the link.
Toerless: if we say the signaling through ASAs depends on the security of the
ACP,
then we need to figure out
what is the minimal requirement to ensure that a
message from one ASA to
another is protected all along the path by the ACP.
This either means
potentially a full mesh of p2p security associations between all
AN nodes,
or
-
and that's what we propose in ACP - to have
hop-by-hop security.
We hope this is most simple.
Michael R.: I don't understand the data plane based ACP, does it mean that
GRASP
has to provide application
layer objecting routing function?
Michael B.:
No. In that case, it means you rely on the routing instance that has been
configured.
Michael B.: who had read the draft? (Some hands raised).
Markus S.: some might only read the previous version.
Brian C.: Suggest people read all three documents and comment, also for the
GRASP and Bootstrapping. For
GRASP, hope people to tell whether there is
something wrong.
Michael R.: very concerned about the ACP draft readability.
Michael R.: still not understand data plane based ACP, will layer3 GRASP
packets,
or GRASP before things hop-by-hop?
Michael B.: It depends on the operational model of GRASP.
Brian C.: normally, the GRASP discovery result should be ACP routable address.
Michael B.: GRASP is actually one way to use ACP.
Toerless: GRASP in/out ACP, need describe in the reference model.
Michael R.: GRASP is the protocol to get the addresses for the data plane to
work,
so before it exist, how do
things work?
Michael B.: For the independent model, it needs the data plane routable before
GRASP can do anything beyond
link local.
WG Chair solicited hum support for the ACP document:
lots of hum supported; non
opposed.
5. Reference
model - 30 min
10:45
- 11:15, by Michael Behringer, draft-behringer-anima-reference-model
Sheng: there are some contents not in current charter yet. With Chair hat on,
the
current draft should focus
on the charted content.
Juliusz C.: it's not clear for me about the concrete
devices, some examples
that clearly in and clearly
out of the gray zone?
Michael B.: ...I really don't know, ideally we want a model, for example a
light bulb
autonomically registered to
the light bulb switch, it supports some basic form
of GRASP or whatever, is
that feasible? I cannot tell right now. My wish is that
a light bulb can join an
autonomic network.
Brian C.: one test case: is the lighting controller for the M floor in Prague
Hilton,
is that powerful enough to
run an ASA?
Toerless: two limitations for a full-autonomic node, one is the size; the other
one is
it must routing. Anything
that cannot route, I'd to put it as constrained and talk
about it later.
Sheng: there is one important section missing in Section 4, an overview of the
infrastructure, the
reference model should describe how those components
working together and how
they fit each other, like some components using
GRASP in some places.
Toerless: what's the minimum for naming and addressing?
Michael B.: we have to figure out what's in the certificate;
Toerless: unless someone brings a good reason why we need naming, we will
use random names.
Sheng: I don't think even random names are needed here. What we actually need
here is some statement
saying we need a name here, whatever the name is.
That's good enough for other
components. Whether it's random or some cool
mechanism to get out some
human-readable names, that's something else,
out of scope for now.
Michael
B.: We need to define naming at least to the extent to what we put in the
subject name in the certificate
John Strassner: I'm really nervous about overloading names for semantics. It's
really a bad idea. IMO, I
think instead we should look at Metadata that can be
used to be augmented. Nodes
augment their features to be used in GRASP,
or could be used in other
things.
Teco Boot: ULA, is it for the site or for the whole
domain? What happened if
multiple domains...
Michael B.: Nodes get addresses, not ASAs.
Toerless: I still like the idea, autonomic nodes get prefix, so that each ASA
could
get an address. But at this
point, I wouldn't see it as necessary minimum.
Markus S.: why don't you just stick in the type of the hash, because
technically
there is no reason to
reserve 3 bits for that. Given if you only have one hash
method, it is the same as 3
bits other methods that 40bit is very unlikely to
collide.
6. (Postponed
to Thursday Session)
Autonomic Prefix Management in Large-scale Networks - 15 min
11:15
- 11:30, by Brian Carpenter, draft-jiang-anima-prefix-management
-------------------------------------------------------------------------
Thursday
(July 23rd, 2015) 1.5-hour session:
17:40-19:10 Thursday
Afternoon session III, Congress Hall I
1. Updates
from Design Teams - 15 min
17:40
- 17:55, by Brian Carpenter & Max Pritikin
Bootstrap Design Team Update -
Max
Toerless: Ensuring the
protocols in the bootstrap are inclusive of the protocols
wanted to be used by ANIMA.
Signaling Design Team -
Brian
CBOR Mapping questions
started.
Goal to determine if CBOR is
reasonable to move forward.
Sheng: The reference model shows how the components work together. The
design teams
should work with the reference document team to ensure
congruent work efforts.
2. Autonomic
Prefix Management in Large-scale Networks - 15 min
17:55
- 18:10, by Brian Carpenter, draft-jiang-anima-prefix-management
Slides are examples to show the use case, not real numbers to be used for this
example.
3. Using
Autonomic Control Plane for Stable Connectivity of Network OAM
-
15 min 8:10 - 18:25, by Toerless Eckert, draft-eckert-anima-stable-connectivity
Sheng: Clarification question: solutions defined are built for v6 ACP.
Recv'd
v4 ACP draft should review and may apply to this draft as needed
and if applicable.
Lopez: Does CA need to be online.
Toerless: Brought in because they are part of the bootstrap draft.
Lopez: will continue to review and bring up online if required.
4. *
Autonomic Network Intent Concept and Format - 10 min
18:25
- 18:35, by Jeferson C Nobre, draft-du-anima-an-intent
Sheng: clarifying as WG chair: this is not a chartered item, although mentioned
intent in the charter text.
It may become chartered item when ANIMA is
re-chartered, or
another group may define and ANIMA will evaluate.
Toerless: Interested in what version or application of Intent will be applied
to
ANIMA.
Brian: Some objects may be large and likely over 1500 bytes
Don: SUPA BOF - may not mean the same thing on the word Intent between
ANIMA and SUPA
John: Draft in SUPA that compares Intent efforts from other sources. Summary
of draft: main diff is
amount of specifics that is conveyed in Intent, some
definitions are vague and others want
details. SUPA appears in the middle of
the two endpoints.
Michael B: focus on what we do inside of ANIMA. Every function can determine
without Intent should be
ASA specific, could increase conflict.
5. *
Autonomic Network Intent Distribution - 10 min
18:35
- 18:45, by Bing Liu, draft-liu-anima-intent-distribution
Michael B: - when would you not flood intent? Is there a use case for point to
point
intent?
Brian: you may need solicited distribution of intent if a node has been added
or was
removed and added again.
Sheng: This appears to be generic distribution function, what is specific to
Intent
here? This
could be moved into or handled by GRASP.
Brian - Yes, this could be a part of GRASP. GRASP needs to be able to extend to
handle Intent, or generic
distribution ability.
John - Nodes will send out Intent, but how do they first get that Intent. Starting
to
design and ESP. Is this required?
6. *
Self-Managed Networks with Fault Management Hierarchy - 10 min
18:45
- 18:55, by Dan Romascanu, draft-mtoy-anima-self-faultmang-framework
Author not here, please read and provide feedback
Jefferson: it would be good to understand how the model fits into ANIMA model.
7. *
Coordination of autonomic functions - 10 min
18:55
- 19:05, by Laurent Ciavaglia (remotely), draft-ciavaglia-anima-coordination
No question from the room
8.
Closing by Co-Chairs - 5 min
19:05
- 19:10, by co-chairs
*
mean the non-chartered work items.