Design and Evaluation of a WideArea Event Notification Service

1
Design and Evaluation of a Wide-Area Event
Notification Service

• ANTONIO CARZANIGA, University of Colorado at
  Boulder
• DAVID S. ROSENBLUM, University of California,
  Irvine
• ALEXANDER L. WOLF, University of Colorado at
  Boulder
• Presenter
• Gunjan Kakani

2
Outline

• Introduction
• Semantics of SIENA
• Architecture of SIENA
• Simulation results
• Critique
• Discussion

3
Introduction

• Event Notification A broad spectrum
• Alternatives to Event based systems
• Components of a General Event Notification System
• Expressiveness
• Scalability
• A generic framework Contd.

4
A generic framework
Object of Interest
Interested Party
Advertise
Subscribe
Notify
Publish
Access Point
Servers
Event Service
5
Semantics of SIENA

• Basic Protocol publish/subscribe
• Sample API subscribe(string identity, pattern
  expression)
• Notification Structure
• Filters
• Common operators
• Covering relationship
• f covers n

6
Elements of SIENA

• Patterns
• Sequence of filters
• Matched as AND
• Sequencing of evaluation
• Advertisement
• Subscription based semantics
• Advertisement based semantics
• Unsubscription and Unadvertisement
• Timing issues

7
Architecture of SIENA

• Topologies and Protocols
• Hierarchical Client/Server Architecture

clients
servers
Client/server protocol
8
Architecture of SIENA

• Acyclic peer-to-peer Vs. General peer-to-peer

Client/server protocol
Server/Server Protocol
9
Routing and Processing

• Routing strategies
• Downstream replication
• Upstream evaluation
• Routing Algorithms
• Subscription forwarding
• Advertisement forwarding
• Filter Poset

Xany
X lt 8
X gt 0
X 3
X gt 5
10
Hierarchical Architecture

• Subscription
• Notification

a
11
Hierarchical ArchitectureContd

• Unsubscription

12
Acyclic peer-to-peer Architecture

• Subscription

sub
fwd
sub
fwd
13
Pattern Matching

• Distributed monitoring following the upstream
  evaluation principle
• Requires advertisement based semantics
• Available patterns table Advertisements
  processed patterns
• Pattern Factoring
• Pattern Delegation

14
Simulation

• Expressiveness
• Data structures close to widely used database
  query language
• Covering relations are well behaved, reasonably
  computational complexity
• Linearly complex patterns
• Framework
• Server/Client topology
• Subscribe, advertise, publish behaviors
• Cost Vs. various application behaviors and load
• Simplistic network characteristics

15
Simulation Results Summary

• Acyclic peer-to-peer architecture with
  subscription forwarding scales well and
  predictably under all circumstances
• Hierarchical client/server architecture has a
  lower per-subscription cost than the acyclic
  peer-to-peer. This cost does not depend on the
  behavior of objects of interest or interested
  parties
• In both architectures, the subscription cost is
  amortized for increased densities of interested
  parties. The difference is less for high number
  of interested parties
• Cost of delivering notification is same for both
  architecture, though acyclic architecture has no
  cost for ignored notification which is fixed in
  hierarchical architecture

16
Critique

• Strengths
• Comprehensive details of SIENA
• Good rationale for design choices
• Leverages on existing results and design
  principles where ever possible
• The system it self qualifies to be called a
  general purpose wide-area event notification
  service
• Weaknesses
• Ignores timing issues
• Ignores security, reliability and fault tolerance

17
Relevance to Embedded Systems

• Relevance to Embedded Systems
• Too far from the framework of embedded systems
• Real time requirements
• Ad-hoc behavior
• Limited processing power
• Poor network connectivity
• A customized solution would be more desirable for
  embedded applications
• To limit the description language and
  expressiveness
• To limit the network traffic
• To define a topology and algorithm optimized for
  the kind of requirement
• Can be used a base since it avoids many
  computational and network complexities by
  implementing simpler data model and efficient
  algorithms
• Should support dynamic network topology

18
Thank You

• Discussion QA