"Subclassing and Composition – A Pythonic Tour of Trade-Offs", Hynek Schlawack
S-CUBE LP: A Soft-Constraint Based Approach to QoS-Aware Service Selection
1. S-Cube Learning Package
A Soft-Constraint Based Approach to QoS-Aware Service
Selection
Université Paris-DESCARTES
Mohamed-Anis ZEMNI, Salima BENBERNOU, Manuel CARRO
www.s-cube-network.eu
2. Learning Package Categorization
S-Cube
Quality Definition,
Negotiation and Assurance
Quality Management and Prediction
Analysis Operations on SLAs:
Detecting and Explaining Conflicting SLAs
3. Service Selection and QoS
Service selection is the first step to improve service
composition within Service-Oriented-Architecture (SOA):
• Searches for services fitting users’ requirements
• Explores services’ properties
• Aims at putting together several elementary services
• Generates new value-added service
Quality of Service (QoS) for selection often critically important:
• Software services expose not only functional characteristics, but also
non-functional attributes describing their QoS
• Defines the service level (Key Performance Indicator)
• A service fulfilling all the functionality but with low QoS is not
interesting
4. Learning Package Overview
Problem Description
Extending SCSP with Penalties & new SLA Model
Conclusions
5. Problem Description:
Service Selection Scenario
Select only one service
among the available
services that have the
same functionalities but
with different QoS
Functionalities
+
QoS
User request (criteria)
1
2
Used Approach at Design-time
6. Problem Description:
Service Selection Techniques in the Literature 1
Constraint Satisfaction Problem (CSP):
• Classical formulation of constraints
• Quite expressive to represent several real life problems
• Defines a set of variables, each of them ranging on a finite domain,
and a set of constraints restricting the values that these variables can
take simultaneously
• All the constraints must be satisfied simultaneously
!
Lack of built-in capabilities to express preferences among constraints
and the lack of possibility of giving approximate solutions for problems
which are overconstrained
7. Problem Description:
Service Selection Techniques in the Literature 1
Soft Constraint Satisfaction Problem (SCSP)
• Include the concept of preferences into every constraint in order to
obtain a suitable solution which can be optimal or, in general, a
reasonable estimation, maybe at the expense of not fulfilling all
constraints
• Relies on composing the constraints in order to obtain the optimal
solution
• Applied to the requirements (in terms of preferences) of the users
! Only one solution returned that is optimal
* Stefano Bistarelli, Ugo Montanari, and Francesca Rossi. Semiring-
based constraint satisfaction and optimization. J. ACM, 44(2):201–
236, 1997
8. Problem Description:
Service Selection Techniques in the Literature 1
C-semi-ring : Algebraic structure
Only one domain for
all variables
Example : Searching for services Available at y% of the time and with reputation = z
9. Problem Description:
Problem at Design-time
2. I have to fix
new criteria
1. Required criteria
cannot match any service!!!
User request (criteria)
10. Problem Description:
Problem at Runtime
!
Some problems, encountered by the service may
lead to service malfunctions
activity interrupted,
must apply penalty!!!
Out of
service Out of
service
contract violation
11. Problem Description:
SLA
SLA - Definition:
“An XML document and a contract for…
• Advertising the quality level of the services
• Taking note about the user preferences
• …”
I want an SLA
ensuring the
performances I
am searching for
Propertie
s
Pro perties
QoS
?
13. Learning Package Overview
Problem Description
Extending SCSP with Penalties & new SLA Model
Conclusions
14. Main Objective
Automatically switch from a faulty
service to a new one
User request (preferences,
… Out of
service Out of
penalties)
service
Design-time
Runtime
15. Approach Main Points
Definition of Soft Service Level Agreement (SSLA) an SLA
model extended with preferences and penalties
Extension of Soft Constraint Solving Problem handling
penalties: Define in SSLA the penalty artifacts, such that, if a
selected service failed, another one should replace it that
fitting with the agreed QoS in the contract with penalties if
some of them are not fulfilled
SSLA to SCSP mapping
16. Kinds of penalties
Arithmetical Penalties
• In relation with measurable qualities of service
• Direct relation to service variables
• E.g. availability, the response time, the reputation, etc.
• The application of arithmetical penalties is a consequence of a
contract breach and therefore the transition to a different selection
using the choices expressed by the customer in the form of
preferences
Behavioural Penalties
• Related to the behavior of either the customer or the service provider
• The application of behavioral penalties is not always a consequence of
a contract breach and so, switching to another choice is not obligatory
and even less replacing the service
18. Soft SLA Definition:
Preferences & Penalties
I prefer to get a payment
service and delivery service
having response time < 5ms. I
also accept services with
response time between 5ms
and 20ms with preference =0,5
Etc.
Response time
Preferences
If the first Most preferred
preference is not <5ms
fulfilled during the
execution I would
apply penalty P7
[5ms,20ms[
>20ms
Less preferred
19. Soft SLA Definition
Guarantee terms are expressed in terms of preferences and
penalties
• Preferences are ranked (most preferred to less preferred)
• Penalties are applied if a preference is not fulfilled
The service broker search for service fulfilling the QoS from
the most preferred to the less preferred (at design-time)
Penalties are applied only at runtime and never at design-
time, on the faulty service
SSLA document
QoS Variable Preference Preferences Penalties Preferences/Penalties
variables doamins degree association
20. Extending SCSP Using Penalties
SCSP
Constraint
System
Constraints
Operations
Solution
21. Extending Constraint System
SCSP
Constraint
CS = <S; D{}; V>
System
S = algebraic structure
including preference
Constraints values
V = QoS variables
D{} = Variable domains
Operations
Penalties into S
Solution
22. Extending Constraints Using Penalties
SCSP
Constraint
Def = Definition of the
System
constraint in terms of
preference value
Constraints Type = in terms of
variable intervening in
the constraint
Operations
Penalties into Def
Solution
23. Rewrite operations Logic
SCSP
Constraint
System Combination =
combination of the
constraints (pref)
Constraints Projection = generates
the optimal solution
Operations Rank generated
solutions and
keep them all
Combination of penalties
Solution
24. Extending SCSP Using Penalties
SCSP
Constraint
System
Global Preferences
Constraints
Most preferred
+
Operations
Less preferred -
Solution
25. Penalty based SCSP
Case Study
Penalty based SCSP
Constraint
System
Constraints = Penalty values
= Preference values
Operations
Solutions
26. Penalty based SCSP
Case Study
Penalty based SCSP
Constraint
System
Constraints
Operations
Solutions
27. Penalty based SCSP
Case Study
Penalty based SCSP
Constraint
System
Constraints
Operations
Solutions
28. Penalty based SCSP
Case Study
Penalty based SCSP
Constraint
System
Constraints
Operations
Solutions
29. Proposed Approach Logic
Input: Constraints, penalties, table of constraint definitions
Output: Choices with their possible alternatives ordered
Begin
For each selection alternative do
Combine all the constraints together (apply the min operator);
End for;
Order the results according to preference values into groups;
For each preference value group do
Order the elements corresponding to the penalty value;
End for;
End;
31. Learning Package Overview
Problem Description
Extending SCSP with Penalties & new SLA Model
Conclusions
32. Conclusions
1. Soft constraint-based framework
2. Express QoS properties reflecting both customer
preferences and penalties applied to unfitting situations
3. Solution for overconstrained problems
– The application of soft constraints makes it possible to work around
overconstrained problems and offer a feasible solution
4. Provide ranked choice to offer more flexibility at design-time
to find required services, and at runtime to ensure users’
rights
5. Concept of penalties in SCSP
We plan to extend this framework to also deal with
behavioral penalties
34. Further S-Cube Reading
[ZBC10] Mohamed Anis Zemni, Salima Benbernou, and
Manuel Carro
A Soft Constraint-Based Approach to QoS-Aware
Service Selection
In proceeding of the Service-Oriented Computing - 8th
International Conference (ICSOC 2010), volume 6470
of Lecture Notes in Computer Science, pages 596-602
San Francisco, CA, USA, December 7-10, 2010
35. Acknowledgements
The research leading to these results has received
funding from:
The European Community’s Seventh Framework
Programme [FP7/2007-2013] under grant agreement
215483 (S-Cube).