1. Proactive SLA Negotiation
Khaled Mahbub and George Spanoudakis
City University London
www.s-cube-network.eu © City University London

2. SLA and SLA Negotiation
Service Level Agreement (SLA)
 A Service Level Agreement (SLA) is an explicit formal contract between the
provider and the clients of a service determining guarantees on service quality
properties and penalties in case of defaults
SLA Negotiation
 SLA negotiation is the process where service providers and clients reach an
agreement on desired levels of service that satisfy each partner involved in the
negotiation process.
 SLA negotiation typically involves the resolution of differences between the
initial positions of the two parties regarding the desired levels of service
provision
 Two types of negotiation
• Reactive: after decisions on binding with a service have been made and/or
once SLA violations has occurred
• Proactive: prior to service binding and/or SLA violation
www.s-cube-network.eu © City University London

3. Proactive SLA Negotiation – Triggering
Scenarios (SLA Violation Scenarios)
 Following forecasts for a higher level of client demand that can put
agreed SLA terms at risk.
 Following forecasts that the constituent services of a system will fail or
violate their own SLAs (this applies to both software and infrastructure
services).
 If it becomes known that competitors provide or are going to provide
better service offerings
www.s-cube-network.eu © City University London

4. Handling of SLA Violations
 Traditional Approaches
– application of penalties depending on the importance of the service
level objectives [1,2]
– runtime negotiation (re-negotiation) of SLA [3,4,5,6,7]:
- To accept a service from the existing provider with reduced service levels
- To terminate the existing SLA and negotiate new SLA with a new provider
 At runtime, these approaches have limitations as they
– May affect the quality of the delivered service, or
– fail to guarantee uninterrupted service
 Proactive SLA negotiation with alternative service providers is a means
for addressing such limitations
 Proactive SLA negotiation needs to be driven and integrated with service
discovery
www.s-cube-network.eu © City University London

5. Two Phase Proactive SLA Negotiation
 Identification of potential service providers (aka SLA negotiators): service
discovery, search in published SLA templates (i.e., service levels
advertised by providers)
 Pre-agreement: Process of agreeing SLA terms without putting the SLA
into force; it may establish a time frame within which the pre-agreement
can be automatically brought into force without further negotiation
 Final agreement: Confirmation that an SLA comes into force
 The two phases are required for proactive SLA negotiation but may also
be useful in reactive negotiation.
www.s-cube-network.eu © City University London

6. Architecture for Proactive (& Reactive) SLA
Negotiation
 The service discovery query may be
specified in order to identify candidate
replacement services that a composite
service (S) uses currently.
 The runtime service discovery tool
identifies potential alternative services.
 The negotiation broker manages the
negotiation process on behalf of a service
consumer/provider.
– Jess rule engine is used as the negotiation
engine
– Negotiation rules are expressed in XML and
translated into Jess rules by negotiation
rule translator.
www.s-cube-network.eu © City University London

7. Negotiation Rule (1)
www.s-cube-network.eu © City University London

8. Negotiation Rule (2)
<tnsr:NegotiationRule name="rule1">
<tnsr:If>
<tnsr:LogicalExpression>
<slac:Condition relation="LESS-THAN">
<slac:Arg1><slac:QualityAttribute name="RESPONSE_TIME"party="CONSUMER"/></slac:Arg1>
<slac:Arg2><slac:Constant type="NUMERICAL" unit=”ms”>10</slac:Constant></slac:Arg2>
</slac:Condition>
<slac:LogicalOperator>AND</slac:LogicalOperator>
<slac:Condition relation="EQUAL-TO">
<slac:Arg1><slac:QualityAttribute name="PRICE"party="CONSUMER"unit=”GBP”/></slac:Arg1>
<slac:Arg2><slac:Constant type="NUMERICAL">0.5</slac:Constant></slac:Arg2>
</slac:Condition>
</tnsr:LogicalExpression>
</tnsr:If>
<tnsr:Then>
<tnsr:Action>
<tnsr:Accept>
<tnsr:QualityAttribute name="PRICE“ party="CONSUMER" />
<tnsr:QualityAttribute name="RESPONSE_TIME" party="CONSUMER"/>
</tnsr:Accept>
</tnsr:Action>
</tnsr:Then>
</tnsr:NegotiationRule>
 if the consumer of a service has made an offer (or counter-offer) where the
response time of the service must be less than 10 milliseconds (ms) and the
price to be paid per service use is 0.5 pounds, the offered values will be
accepted.
www.s-cube-network.eu © City University London

9. Negotiation Rule (3)
 A rule in Jess has the form
(defrule rule-name(logical-operator (cond-1 …cond-n))
action)
 XML negotiation rules are transformed into Jess rule by applying a set of
translation patterns:
Negotiation Rule Element Jess Representation
<slac:Arg><slac:QualityAttribute A-P
name="A” party="P"/></slac:Arg>
<slac:Arg><slac:Constanttype="NUMERICAL">C C
</slac:Constant></slac:Arg>
<slac:Condition relation="REL"> {Jess-Rep(Arg1) REL Jess-
<slac:Arg1>...</slac:Arg1> Rep(Arg2)}
<slac:Arg2>...</slac:Arg2>
</slac:Condition>
<tnsr:Action><tnsr:Set> (modify 0 (Jess-Rep(Arg1)
<tnsr:LogicalExpression> (fact-slot-value 0 Jess-
<slac:Condition relation="EQUAL-TO">
Rep(Arg2))))
<slac:Arg1>...</slac:Arg1>
<slac:Arg2>...</slac:Arg2>
</slac:Condition></tnsr:LogicalExpression>
</tnsr:Set></tnsr:Action>
www.s-cube-network.eu © City University London

10. SLA Negotiation Process (1)
 potential candidate services
(RS) are identified.
 Services in RS are ranked
based on their distances to the
query.
 RS is updated regularly.
 An SLA is negotiated and a pre-
agreed SLA is established for
each service in RS.
www.s-cube-network.eu © City University London

11. SLA Negotiation Process (2)
 Time validity period is set for each
pre-agreed SLA.
 pre-agreed SLA is proactively re-
negotiated if the service is unused
and the validity period is about to
expire.
 A service in SBA is replaced by
the best available service in RS if
it is needed.
 If a service in RS is selected for
dynamic binding, its pre-agreed
SLA is automatically enforced.
www.s-cube-network.eu © City University London

12. Proactive SLA Negotiation – Evaluation (1)
 Case study and evaluation set up:
– A service based client application, called Route Planner, implemented
by a BPEL service orchestration process.
– Route Planner uses several services, including Global Positioning
Service (GPS) and electronic map service (eMapS).
– A service discovery query is specified in order to identify candidate
replacement services for GPS of Route Planner. The discovery query
expresses the structural discovery criteria as WSDL of GPS service.
– An SLA template with four QoS terms (price, availability, response time
and mean number of service request per hour) for negotiation is used.
www.s-cube-network.eu © City University London

13. Proactive SLA Negotiation – Evaluation (2)
 Setup
– A set of 15 service consumer negotiation rules (CNR set) is sued.
– 20 different sets of provider negotiation rules sets (PNR set) are used,
where each set contains between 5 and 20 rules. During experiment,
negotiation broker randomly selects one PNR to simulate the different
behaviour that different service provider might have.
– Three different service sets (registries), containing 100, 300 and 500
services, are used. Service sets are populated by geographic location
related services taken from SEEKDA.
– All experiments were carried out using a Pentium 2.33GHz with
3.23GB RAM machine.
www.s-cube-network.eu © City University London

14. Proactive SLA Negotiation – Evaluation (3)
 Measurements of the time needed to
a) Build the initial candidate service set (RS) by runtime service
discovery (SD)
b) Maintain the RS at runtime due to arrival of new service or change in
the description of an existing service in the registry
c) Select a new service for replacing a service in the service based
system
Times required for (a), (b) and (c) were measured for executions of service
discovery process
• Without SLA negotiation (SD Only)
• With proactive SLA Negotiation (SD with Proactive SLA)
• With reactive SLA negotiation (SD with Reactive SLA)
www.s-cube-network.eu © City University London

15. Proactive SLA Negotiation – Evaluation (4)
 Results (time measurements)
SD Only SD with Proactive SLA SD with Reactive SLA
100 300 500 100 300 500 100 300 500
Replacement
Service Set (RS)
160.62 408.05 686.697 157.399 435.985 715.685 159.804 408.03 688.23
Building
(Time in Sec.)
Replacement
Service Set (RS)
Maintenance. 690.5 698.3 673.5 837.2 881.2 892 637.4 668.5 677.6
avg(trsm-*)
(Time in milli-sec)
Selection of
Replacement
Service from RS. 22 28 21.8 53.2 45.8 50.2 453 459.4 443.8
avg(trep-*)
(Time in milli-sec)
www.s-cube-network.eu © City University London

16. Proactive SLA Negotiation – Evaluation (5)
 Results (time measurements)
– service discovery with proactive SLA negotiation is slightly larger than
the time required to identify a replacement without any SLA
negotiation.
– time required to select and bind a replacement service at runtime in
the case of SD with reactive SLA negotiation is significantly larger than
the service selection and binding time in the case of SD with proactive
SLA negotiation
www.s-cube-network.eu © City University London

17. Proactive SLA Negotiation – Evaluation (6)
 Results (effectiveness)
Service Agpro = avg(trep-sd-rea-sla)  RGpro=avg(trep-sd-rea-sla) / BEpro= (trsm-sd-pro-sla – trsm-
Registry avg(trep-sd-pro-sla) avg(trep-sd-pro-sla) sd-rea-sla)/(trep-sd-pro-sla – trep-
sd-rea-sla)
100 187 8.51 0.999
300 230.4 10.03 1.028
500 192.8 8.84 1.089
www.s-cube-network.eu © City University London

18. Proactive SLA Negotiation – Evaluation (7)
 Results (effectiveness)
– possibility of inefficient resource utilization.
– the efficiency of resource utilization with proactive SLA
discovery/negotiation over a period of time T can be defined as,
• SRRR: the service replacement request rate;
• SRUS: service registry update rate;
• tmatch : average time required to match a query with a service
• tSLA-Neg : average time required to negotiate an SLA with a service
• tinit-RS : the time needed to build the initial copy of RS
– to have efficient resource utilization, there must be that
SRRR ≥ SRUR + Rinit/T i.e. SRRR ≥ SRUR
www.s-cube-network.eu © City University London

19. Conclusion & Future Works
 Although preliminary, our experiments have shown that proactive SLA
negotiation leads to significant reduction of the time required to perform
service replacement at runtime.
 This approach has opened new possible lines for future investigation,
– Support for proactive negotiation for hierarchical SLA.
– Heuristics for tuning the triggering of proactive SLA negotiation
process so as to reduce the number of cases where pre-agreed SLAs
never get used
– Support for dynamic adaptation of negotiation rules.
www.s-cube-network.eu © City University London

20. S-Cube Publications
 Check the following S-Cube papers for a detail description of Proactive
SLA Negotiation framework.
 Khaled Mahbub and George Spanoudakis, "A Framework for Proactive SLA
Negotiation", 5th International Conference on Software and Data Technologies
(ICSOFT 2010), 22 - 24 July, Athens, Greece.
 Khaled Mahbub and George Spanoudakis, "Proactive SLA Negotiation for Service
Based Systems", Proceedings of the 2010 IEEE World Congress on Services
(SERVICES 2010), 5-10 July 2010, Miami, Florida, USA.
 Khaled Mahbub and George Spanoudakis "Proactive SLA Negotiation for Service
Based Systems: Initial Implementation and Evaluation Experience", 8th IEEE
International Conference on Services Computing, July 4 - 9, 2011, Washington DC,
USA.
www.s-cube-network.eu © City University London

21. References
1. Service Level Agreements”, the Proceedings of the Usage of Service Level Agreements in Grids
Workshop, 2007
2. Q. He, J. Yan, R. Kowalczyk, H. Jin, Y. Yang, "Lifetime Service Level Agreement Management
with Autonomous Agents for Services Provision". Information Sciences, Elsevier, 2009
3. “SLAM – Service Level Agreement Management”, Whitepaper, Whitestein Technology, August
2005.
4. Dominic Greenwood, Giosue Vitaglione, Lukas Keller and Monique Calisti, “Service Level
Agreement Management with Adaptive Coordination”, International Conference on Networking
and Services (ICNS’06), July 19-21, 2006, Silicon valley, USA.
5. Di Modica G.,Tomarhio O. and Lorenzo V., "A framework for the management of dynamic SLAs in
composite service scenarios", Service-Oriented Computing - ICSOC 2007 Workshops: ICSOC
2007, International Workshops, Vienna, Austria, September 17, 2007
6. Michael Parkin, Peer Hasselmeyer, Bastian Koller, Philipp Wieder: An SLA Re-negotiation
Protocol. In proceedings of the 2nd Workshop on Non Functional Properties and Service Level
Agreements in Service Oriented Computing at ECOWS 2008. Dublin, Ireland. 12 November
2008.
7. Rizos Sakellariou and Viktor Yarmolenko, "On the Flexibility of WS-Agreement for Job
Submission", Proceedings of the 3rd International Workshop on Middleware for Grid Computing
MGC '05, vol. 117, 1-6 (Nov. 2005)
www.s-cube-network.eu © City University London