Chained Negotiation for Distributed Notification Services

1
Chained Negotiation for Distributed Notification
Services

• Richard Lawley, Michael Luck and Luc Moreau
• Intelligence, Agents, Multimedia Group
• School of Electronics Computer Science
• University of Southampton
• ral01r_at_ecs.soton.ac.uk

2
Outline

• Introduction
• Notification Services
• Negotiation
• Chained Negotiation
• Evaluation of Negotiation Engine
• Future Work Conclusion

3
Introduction

• Notification services (NSs) provide asynchronous
  communication between services in distributed
  environment
• Consumers and providers may conflict over
  delivery preferences (QoS)
• Negotiation enables mutually acceptable QoS
  levels to be found

4
Introduction

• Previous work introduced automated
  service-oriented negotiation engine suitable for
  use with notification service
• Distributed notification services require
  different kind of negotiation
• Chained negotiation supports reselling or
  redistribution of existing services

5
Notification Services (1)

• Message-oriented middleware layer using
  publish-subscribe model
• Publishers publish information on a topic
• Consumers subscribe to that topic
• Notifications are delivered by NS to consumer
• Can process notifications before sending
• Filtered notifications
• Notifications combined into a digest

6
Notification Services (2)

• Example notification topics
• Changes in database content
• Release of new tools or services
• Change in workflow execution state
• Many existing notification services
• MSMQ, MQSeries, Grid Monitoring Architecture,
  NaradaBrokering
• Negotiation not supported

7
Distributed Notification Services

• Centralised NSs leads to scaling problem
• Overloaded with many consumers
• Single point of failure
• Distributed NS allows loads to be spread and
  effect of failure to be reduced

8
Distributed Notification Services

• Consumers publishers connect to different NS
• Enables notifications to be redistributed
• Less bandwidth used
• Consumers use local NS instead of remote

NS1
NS2
P2
C1
Publisher
Notification Service
C2
NS3
P1
Consumer
NS4
9
myGrid

• e-Science project aimed at Bioinformatics
• Allows workflow-based in-silico experiments
• Automated workflow management through
  personalisation, notification of change and
  publication of experiments
• Increasing focus on data-intensive bioinformatics

10
myGrid Notification Service

• JMS-based notification service
• Persistent message delivery model
• Push/Pull subscribers
• Supports federated topics
• Topics at each NS are part of a virtual topic
• Consumer subscribing to virtual topic receives
  notifications from all member topics
• Uses service directory for virtual topic data

11
QoS in Distributed NS

• Consumers still want to request QoS levels in a
  distributed NS
• If QoS requests are simply granted, cant
  redistribute notifications
• Some form of negotiation needed to find
  acceptable levels of QoS

12
Negotiation

• The process by which two or more parties
  exchange proposals in order to reach a mutually
  acceptable agreement on a particular matter
• Components of negotiation
• Item the object under negotiation
• Terms attributes of item under negotiation
• Proposals set of values suggested by one party
• Preferences values acceptable to a party
• Utility functions a method of evaluating the
  utility of a particular proposal

13
Negotiation

• Negotiation formally described as
• Negotiation Protocols Rules governing a
  negotiation
• Types of participants
• Valid negotiation states actions
• Negotiation Strategies How participants behave
  within protocol
• Proposal Generation
• When to accept a proposal
• Specification of negotiation enables automated
  negotiation to take place

14
Chained Negotiation

• Conventional negotiation (direct) involves
  negotiation between client and supplier
• Chained negotiation involves middlemen between
  supplier and client
• Enables reselling or redistribution
• Made up of multiple subnegotiations
• Item can be supplied by supplier or middlemen

15
Chained Negotiation

• Middlemen choose which sub-negotiation to proceed
  with at each step
• Upstream away from client
• Downstream towards client

16
Chained Negotiation Protocol

• Rule Proposals may be sent upstream only if
  there is time for the reply to get back to the
  client
• Messages contain distance of sender from client
• Rule NC cannot confirm downstream without
  having a commitment in place upstream
• Prevents consumers committing to something
  unavailable
• Propose, Accept, Confirm-Accept messages
• Enables getting everyone prepared to accept

17
Chained Negotiation

• Proximity function determines whether one
  proposal is satisfiable by another
• Used to find suitable existing commitments
• Actions created each time proposal received
• Selecting existing commitments
• Passing across proposal
• Scoring functions choose best action

18
Chained Negotiation

• Scoring functions used
• Proximity how close is the proposal in each
  action to the last received proposal?
• Acceptable? would this action directly lead to
  an acceptable state?
• Favour least used if the previous few rounds
  have resulted in upstream negotiation, this
  favours downstream more

19
Chained Negotiation Evaluation

• Evaluate different types of negotiation
• Direct (client supplier)
• Forwarded (middleman forwards proposals)
• Chained (middlemen uses existing commitments)
• Experiment setup similar to evaluation of
  previous negotiation engine
• Common interval-based time model
• Only time-dependent tactics used
• Single tactics used for proposal generation
• Only client utility considered

20
Variable Deadline

• What effect does varying the deadline have on
  utility and success rate?
• Vary deadline (1-100 messages).
• Measure utility and success rate for different
  types of negotiation

21
(No Transcript)
22
(No Transcript)
23
(No Transcript)
24
Variable Deadline

• Problem with spikes
• Client doesnt know distance from supplier
• Final concession not reaching destination
• ?Failed negotiation
• Predictable troughs
• 2n(m 1) 1

25
Variable Negotiation Terms

• Easy to match to existing commitment when using a
  single term
• All terms in a proposal need to match commitment
  in order to reuse it
• Increase number of terms and measure success rate
  for each type of negotiation

26
(No Transcript)
27
Supplier Capacity

• Multiple connected middlemen
• Clients spread between them
• Compared to negotiating with publishers NS

M
M
M
M
M
S
28
Supplier Capacity

• Most chained negotiations satisfied by existing
  commitment (lt5 direct)
• Load on supplier greatly reduced
• Negotiations more successful and better utility
  when using chained negotiation

29
Future Work

• Reduce impact of deadline problem
• Integrate with myGrid Notification Service
• Use myGrid Notification Service to provide better
  performance in some test cases
• Alternate interaction patterns between NSs?

30
Conclusion

• Distributed negotiation services need a new type
  of negotiation to select QoS levels
• Better success rate client utility can be
  obtained by reusing existing commitments
• Benefits of chained negotiation decrease with
  larger numbers of terms
• Chained negotiation suffers from a periodic poor
  result with certain values for the deadline

31
Questions?

• E-mail ral01r_at_ecs.soton.ac.uk