Table des matières
Publish / Subscribe Systems
Tutorial reference
Tutorial web page(s)
Motivation
Diapositive PPT
Diapositive PPT
Scenarios
Diapositive PPT
What we need ?
Diapositive PPT
Conceptually speaking:
Solution: A software engineer’s point of view
Diapositive PPT
Also known as Publish / Subscribe Paradigm
Objectives of tutorial
Tutorial overview
I. Overview & introduction
Terminology
Diapositive PPT
The publish and subscribe paradigm: Characteristics
Benefits of the P/S paradigm
Other paradigms
other paradigm cont.’d
other paradigm cont.’d
A comparison of the above
Well, but … (some exceptions ?).
II. Application domains of P/S systems
Application examples & domains
Alerting services in digital libraries
Diapositive PPT
Enterprise application integration
Requirements imposed by EAI
Network and distributed system management
Requirements imposed by NW mgmt.
Selective information dissemination
Diapositive PPT
Diapositive PPT
Question ?
Part III: P/S systems
P/S system design space
Event occurrence and notification semantic
Event delivery semantic
Event propagation semantic
Other characteristics
Subscription and publication languages
Diapositive PPT
Example: LDAP-like publication model (semi-structured)
Example cont.’d
Example cont.’d (a publication / an event instance)
Example cont.’d (subscription language)
Diapositive PPT
Subscription languages
Subscription language: examples
Further subscription language design options
How about XML as sub. & pub. language ?
XML as publication language
“XML” as subscription language
Architectures and prototypes
Open Specifications & Standards
CORBA Event service
Diapositive PPT
CORBA Event service cont.’d
CORBA Notification Service
Overview of research prototypes
The LeSubscribe P/S system
Diapositive PPT
Diapositive PPT
Diapositive PPT
LeSubscribe system architecture
Overview of industrial strength systems
WebCaster - Microsoft’s Web push technology
PointCast Inc.
End of first part
Second Part Algorithms for large scale event filtering
The general matching problem
Two key problems to solve...
Challenges for local matching problem
Challenges for broker network design
Outline
Roadmap
Event Matching
The local event matching problem
Subscriptions vs. database queries
Diapositive PPT
Diapositive PPT
Problem specificity: flexibility of publication & subscription
Changes of Events and subscriptions patterns
Problem definition: Summary
Roadmap
Roadmap
R-trees
R-trees performances
Two approaches
Roadmap
Multidimensionnal indexing when events are range queries
Balanced partitioning does not work
Multidimensional indexing: Summary
Roadmap
Multiple one-dimension indexes
Roadmap
Counting algorithm
Data structures for the counting algorithm
Counting algorithm
Counting algorithm: search cost
Counting algorithm: cost of data structures
Counting algorithm: Summary
Roadmap
Hanson algorithm
Data Structures for the Hanson algorithm
Hanson algorithm: Filtering cost
Hanson algorithm: cost of data structures
Hanson vs. Counting in main memory
Roadmap
Propagation Algorithm: An Efficient main memory implementation of Hanson Algo.
Cache miss
Avoiding cache misses
Propagation algorithm
Cluster matching with prefetch
Counting versus propagation algorithm: Filtering time
Counting versus propagation algorithm: Subscription insertion time
Multiple one dimension indexes: summary
Roadmap
Filtering based on multiple multi-dimensional indexes
Schema-based Clustering
Cost analysis of propagation algorithm
Dividing a cluster
Le Subscribe filtering algorithm
Filtering performance
Insertion of subscriptions
Dynamic algorithm
Changes in subscription patterns
Local matching algorithms: Summary
Outline
Problem to solve
Limitations of network solutions
Outline
Outline
Scalability Principle 1: dowstream replication of notifications
Broadcast of subscriptions (Gryphon):
Broadcast of subscriptions: summary
Oder-based subscription forwarding (SIENA):
Use of publication advertisements (SIENA[])
Broker network design: Summary
Going further …
Conclusion-Summary
Roadmap
Continous Queries
Triggers
Subscription matching vs. Triggers & continous queries
Subscription matching vs. Triggers & continous queries
Bibliography (selected resources)
Bibliography (selected papers)
Bibliography
Bibliography
Bibliography
Unbalanced partionning: Pyramid trees [Berchtold and al.]
But not adapted for incomplete subscriptions
Redistributing a cluster
Dividing a cluster
Cost approximations
Dynamic clustering algorithm
Correlated changes in events and subscriptions
Roadmap
Tree based matching algorithm
and subscriptions are incomplete...
And events may have multi-valued attributes...
A simple analysis
|