End-To-End Qos Contract Negotiation: Hélia Pouyllau

End-to-End QoS Contract
Negotiation
Hélia Pouyllau
Outline
 Context and motivations
 A framework for QoS contracts
 1st problem: Individual request negotiation: DP
 2nd problem: Multi-request negotiation:
Symmetric DP
 Other problems
 Conclusion
Outline
Symmetric DP
 Other problems
 Conclusion
Internet: multi-service networks to
multi-network service
Networks
•DiffServ Protocols Services
•MPLS •Routing (BGP ...) •Multimedia
•IntServ •IP •VPN
Domain
Domain
Heterogeneous No QoS information Need of resources

Independent Reservation
•QoSR
•Distributed Self-Management •Contract approach (SLA, SLS)
•Extended-RSVP
•WS architectures •Service Level negotiation
•Resource Reservation
(MESCAL, SWAN) (COPS-SLS)
(RSVP)
Service provider's challenges: client satisfaction, cost control

Service establishment
Domain
Domain66
Domain
Domain00 Domain
Domain11
 Heterogeneous domains
 Different QoS classes per trail
 Different QoS parameters Domain
Domain22
 Contract privacy
 Routing : Main route + P2P view (domain
Domain 44
knows
Domain target and neighbouring domains)
Domain Domain
Domain33
Domain55
QoS contract Negotiation
Contracts
Contractson
onMiddle-to-End
Middle-to-Endpaths
paths
QoS budget
End- User
(via EUM)
State of the art
 Distributive QoS management
 [TIPHON,ETSI],[EUROSCOM],[MESCAL]: distributed model
 [Kamienski03]: centralized negotiation, SLA
 [Lima05],[Lima03], [Nguyen03]: COPS-SLS
 [Nguyen06]: qos composition fuzzy distributed CSP
 Service selection driven by QoS

 [YuLin05]: centralized , knapsack problem
 [GuNahrstedt03], [GuNahrstedt04]: distributed service compostion
 [SEO05]: centralized, SLA
 [ZengBenatallah03], [Berbner06]: centralized
 QoS and cost control

 [LiuSquillante01]: queueing model
 [Johari,Tsitsiklis04],[Shakkottai05]: game theory
Outline
Symmetric DP
 Other problems
 Conclusion
SLOs: QoS Parameters
Notations
QoS vector composition
Example
Outline
Symmetric DP
 Other problems
 Conclusion
Individual request negotiation
 Given a target, an end-to-end QoS
requirement,
 Negotiate QoS contracts of each participant
 Minimizing a global cost,
 Subject to an individual QoS budget
Problem formulation
DP formulation
Distributed algorithm in DP
 1st sweep: middle2end QoS budget + cost to that point
 2nd sweep: best local commitment selection
Commitment remaining QoS select best Best cost for

+ cost budget + cost commit. choice QoS request
QN-1
Q1,C1 Q2,C2 CN-1 QN,CN
Q’1,C’1 Q’2,C’2 Q’N,C’N
Q0
domain1 domain2 domainN-1 domainN

Global QoS Message from Select commitment Select commitment
request domain 1 in domain N-1 in domain N
Outline
Symmetric DP
 Other problems
 Conclusion
Resource conflict
Domain
Domain66
Domain
Domain00 Domain
Domain11
Domain
Domain22
Domain
Domain44
Domain Domain
Domain33
Domain55
30 60
Example 5 15
40 50
5 15
30 90
30 120 15 20 30 55
10 35 5 20
35 85
10 25 40 45
delay 5 20
SP3
150 SP1 SP2
45 30 30
5 10
30 60
jitter 15 15
60 90 40 20
15 30 5 10
35 55
10 20
30 25
5 15
40 15
5 15
Example
30
5
40
10
30 45
30 75 15 5 30 10
5 20 5 5
35 40
10 10 40 SP6
10
SP5 SP2
105
25 30
5
30
15
55 50 40
10 15 10
35 15
10 5
Multi-request negotiation
 In a finite set of domains, given a target, end-
to-end QoS requirements,
 Negotiate QoS contracts of each participant
 Minimizing a global cost,
 Subject to an individual QoS budget and
QoS capacities of participants
Problem formulation
Number N of domains is finite
Forward-Backward DP
 Suited to intra-domain optimization of QoS commitments
 Starts at both extremities
 Each “peer” decides its own best commitment
Best local
QoS commitment
Left QoS commit. matching QoS request Right QoS commit.
+ best cost + best cost
QL,CL QR,CR
Local optimization
 Branch and Bound, Interior point, Branch

and Cut
Local optimization
 Allows to renegotiate a first QoS commitment to allow extra
users
 Distributes QoS bottlenecks to less committed/constrained
nodes (so requires cooperating domains!)
Joint QoS commitment

Back propagation of
For both connections
QoS/cost constraints Q’+,p’+
Q+,p+ R-,p-
Forward propagation of
No more possible QoS/cost constraints
Q’-,p’-
QoS commit.
for second connection
Generalization (on-going)
Request dependence
r1
r3 a b
s t
c d
r2
Pipe (set of paths) negotiation (on-
going)
 Path wise vs. Pipe wise
 Forward-Backward DP
 Local optimization: LP(?) => choice of a path
QN-1
Q1,C1 Q2,C2 CN-1 QN,CN
Q’1,C’1 Q’2,C’2 Q’N,C’N
Q0
domain1 domain2 domainN-1 domainN

Negotiation in an open system
 Finite set of domains are cooperating to
achieve negotiation
 Is negotiation realistic in an open system ? 
Can we compel every domain to cooperate ?
 A solution: Modelizing QoS request from
« the rest of the world » as disruptive flows:
 We respect their constraints
 We do not search to optimize their prices
 So any candidate for cooperation ?
Outline
Symmetric DP
 Conclusion
Conclusion
 Description of a framework to handle
composition and decomposition of QoS vectors
Description Implementation
Individual request Done MatLab, WS
Multi-request with single path Done Matlab (centralized and

distributed), WS
Multi-request in a graph; On-Going On-Going
negotiation in an open
system; negotiation of pipes
 Re-negotiation, failure cases
Future work
 Multi-request negotiation in a graph: mechanism to detect request
dependencies, termination
 Pipe negotiation, Negotiation in an open system
 Comparison of the different solution performances
 Margin
 Use of failure cases, Definition of re-negotiation,
negotiation policies ...etc.
 Negotiation learning and QoS routing
 QoS classes definition based on service –> based on
demand (game theory)
End-to-End QoS Contract
Negotiation
-?-
Performances

End-To-End Qos Contract Negotiation: Hélia Pouyllau

Uploaded by

Copyright:

Available Formats

End-To-End Qos Contract Negotiation: Hélia Pouyllau

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

End-To-End Qos Contract Negotiation: Hélia Pouyllau

Uploaded by

Copyright:

Available Formats

End-to-End QoS Contract

Heterogeneous No QoS information Need of resources

Service provider's challenges: client satisfaction, cost control

 Service selection driven by QoS

 QoS and cost control

Commitment remaining QoS select best Best cost for

domain1 domain2 domainN-1 domainN

 Branch and Bound, Interior point, Branch

Joint QoS commitment

domain1 domain2 domainN-1 domainN

Multi-request with single path Done Matlab (centralized and

You might also like