Dear NETMOD and YANG experts,
that's food for thought, but probably already a frequently asked or discussed
question in the group.
I'm wondering about the long-term evolution of YANG as (management) data
modelling language.
Like to using following acronyms according to "NETMODs generation principle":
- current generation: YANG = Yet-Another-Next-Generation
- subsequent generation: YANBOG = Yet-Another-Next-But-One-Generation
The rationale behind my question is related ontology-based network or system
management, mainly drivien by two tightly coupled trends (T):
T1: increasing automation of operations and management (see e.g., ETSI GS ZSM
002 Reference Architecture, ETSI GANA, TM Forum IG1251 Autonomous Networks
Reference Architecture);
T2: increasing intelligence of management manager and/or management agent
associated management applications (using more algorithmic intelligence or/and
artificial intelligence, with increasing levels of intelligence).
(We might replace the term ontology-based management also by term
partially/fully machine-based management (in my view).)
Let's challenge YANG: above trends imply (at least) two requirements (R),
R1 ("killer app requirement"): human agents as operators and managers
completely replaced by machine agents (with algorithimic or/and artificial
intelligence, and an intelligence level of …) = fully automated operations and
management;
R2 ("language requirement"): Semantic capabilities of machine agents shall
support an unambiguous reasoning/inference/interpretation/… model in context of
uncertainities (e.g., probabilisitic reasong).
Question: does YANG satisfy requirements R1 and R2?
Answer: no (not yet), because ..
Status quo (YANG V1.1): very rough characterization by
a) YANG = data modeling language (but not an ontology modeling language)
b) but YANG is more: actually an information modeling language, characterized
by specification/description elements for semantics, simplified outlined by
b1) support of formal "data types";
b2) support of formal "relations" (the inter-node links in a tree structure
of YANG graphs) (constrained by binary relations only (with exception of
explicit interlinkage definitions; constrained by basic semantic only);
b3) support of informal "description" of a leaf (due to natural language)
only
b4) ...
Conclusion: semantic capabilities (b1) and (b2) provide already a formal
baseline, but (b3) violates requirements R1 & R2.
NOTE to (b3): any natural language based (= informal) specification would
require the usage of NLP-capable AI (natural language processing), which looks
pretty odd when defining semantics associated to an explicit data/information
model, the chance in defining model-theoretic semantics. Thus, (b3) should
already allow machine-based processing using algorithmic intelligence only,
which again implies a formal language (i.e., a dedicated logic as mathematical
language).
My hypothesis, in brief:
H1: the Yet-Another-Next-Generation YANG doesn't need to be replaced a
Yet-Another-Next-But-One-Generation YANBOG (on a first glance).
H2: rather, YANG Version 1.x may likely evolve towards YANG Version 2 by
explicit support of formal semantic specification elements.
(NOTE, similar what actually happened with SysML Version 1.x towards
SysML Version 2 and KerML-formal anchored semantics in the area of system
modeling languages).
H3: YANG Version 2 would correspond to basic Ontology Specification Language
capabilities (with semantic constraints of …).
That's an opptimistic hypothesis from YANG perspective (gven the history from
SMI to YANG).
However, may imagine other views in the NETMOD/NETCONF community.
Sorry for that long text but would appreciate any comment or feedback!
Regards,
Albrecht
Appendix I - YANG Version 2 semantic specification capabilities
I think that the vision should be a YANG V2 aligned to OWL 2 Full as
orientation.
I.e., YANG V2 would provide semantic capabilities according to description
logic SROIQ(D) as the most expressive-yet-decidable description logic (in my
understanding).
Whether such expressiveness would be required in the end for YANG V2 is for
further studies.
On the other side, it would facilitate optimal transformation of YANG models in
OWL ontology models.
I'm neither an expert on languages nor logics, that's why I didn't thought yet
how SROIQ(D) elements could be mapped or integrated in YANG (apart from some
ideas).
Appendix II - Intelligent actors
We are focusing on the final actor in the information path/loop as semantic
endpoint. The existing dominating, human-type of endpoint will be more and more
replaced by machine-type of endpoints.
What endpoint in detail?
The manager when looking at a managed system by a manageing system.
The controller when looking at a controlled system by a controlling system.
The operator when looking at an operated system by an operating system.
Or whatever information plane YANG is able to model data.
The spectrum of managers, controllers, operators, etc as actors has to cover
the entire range, from
A_H: humans as natural agents with natural intelligence and varying degrees
of intelligences (from newbies till sysadmins as experts);
to
A_M: machines as artficial agents with basic algorithmic intelligence till
artificial intelligence at varying degress of intelligence (according to
required levels of management automation).
Appendix III - RFC 3444 - Information vs Data Models, - or what information
theory?
That RFC from 2003 doesn't really help here, the RFC was obviously written from
a human manager/operator perspective.
That RFC needs to be revised in accordance to modern information theories in my
understanding, which would then allow a very precise distinction between data
models, information models and knowledge models.
We know that there isn't yet any Unified Theory of Information (UTI), but we
also don't need an UTI here. I'm rather pretty convinced that e.g. the Theory
of Strongly Semantic Information (TSSI) would satisfy all data engineering,
information engineering, communication engineering and knowledge engineering
requirements as in scope of IETF information, computation and communication
systems ("my subjective hypothesis!").
Dr. Albrecht Schwarz
Systems Engineer
[email protected]
www.etas.com
ETAS – Empowering Tomorrow’s Automotive Software
_______________________________________________
netmod mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/netmod
