ontological middleware enabling semantically-aware networking ben tagger and dirk trossen...
TRANSCRIPT
ONTOLOGICAL MIDDLEWAREEnabling Semantically-aware Networking
Ben Tagger and Dirk Trossen (presenting)
2
The Role of Middleware (simplified!)
User/App
Core Architecture
pub
lish
Sco
pe
pub
lish
Dat
a
subs
crib
eSco
pe
subs
crib
eIte
m
and
so
on …Middleware
current proposed
3
What we hope to achieve…• Provide middleware that matches and builds on our goal
of information centricity.• Enhance the usability of the core architecture.• Exploit the functionality of the core architecture.• Separate user-space and network-level data/metadata
requirements.• Incorporate formal modelling of metadata.
• Promote collaboration and discussion within the group!
4
The Principal Design Challenge• Understand the (meta)data flow through our system (the
blue bit).• What data/metadata can be provided to the middleware?
• i.e., what does the user know?
• What data/metadata is required at the network level?• What does the network need to know?
Core
Architecture
Meta(data)
5
Exploring the Design Space• At the extreme left, we have a flat
mapping of metamodels.• (+) Distinct separation of middleware/
network metadata.• (-) Removes the possibility of network
optimisations.
• At the extreme right, we have an exactmapping of metamodels.
• (+) All metadata is preserved at the network level – enables optimisations.
• (-) Replicating metadata that may not be required. Must compromise on expressiveness of metamodel.
✔ We are looking for the optimum design.
6
Separation of Metadata• The metadata of publications exists
between layers.• User-interesting metadata exists at
the middleware level.• Network-interesting metadata exists
at the network level.• Some metadata will be of interest to
both parties.
Questions:
• What metadata should be placedat which level?
• How is metadata modelled withinthe middleware?
• How is metadata modelled withinthe network?
• We have some answers…
ontologies!
SIds, RIds, etc.
7
Design Overview
videopublish
Content QoS Security Other…
Stock Ontologies
Annotations
User/App
Network
Ontological Metamodel
RendezvousMetamodel
Presentations
annotate
Key Points:• publications are
annotated with ontological concepts.
• ontologies can be built on any relevanttopic and easily integrated.
• If we wish to include a new feature, we need only build a new ontology.
8
Benefits of the Design 1• Formal Metamodelling
• All metadata is placed within a structure.• We can demonstrate the correctness of a metamodel. And we can
demonstrate the incorrectness!• Consistency is passed down to the Network level (Network metamodel
is built by the middleware).
• Formal Publication• Every publication has a formal definition.• Every bit of data in our network has a meaning. “The what is more
important than ….”
• Advanced Subscriptions• subscribe to publications using expressive queries.
• “give me all the photos uploaded in the last 2 days within ½ mile of me”• “give me videos that feature Madonna and in high definition.”
• The capabilities of the queries depend on the expressiveness of the publications. The more you put in, the more you get out…
9
Benefits of the Design 2• Metadata Management
• Thorough analysis of metadata; what metadata exists?• Separation of metadata; what metadata goes where?
• Leverages the Current Metamodel• Puts only the right semantics at the Network layer.• Opens the possibility to optimise operations while enhancing
functionality.
• Decouples Data and Metadata• Modularity of transport/signalling mechanisms.• Middleware contains no data. It’s all in the network layer (where it
belongs!)
10
Challenges• Expressiveness vs Complexity tradeoff
• We do not advocate an entirely ontology-based design!• Strength of the architecture allows for choosing different level of
realisations, depending on the desired tradeoff• For each area, we have to determine this tradeoff, e.g., in QoS
• Demonstrate the strength• Simplify network-level operations is only one area• Other, more user-level areas are candidates
• But these are mainly not in the scope of PURSUIT
• Demonstrate the inherent architecture advantage• Prove our ‘impedance mismatch’ claim• Need metrics to determine efforts for middleware development
• More than just complexity
11
Thanks, Questions and Discussion