A FRAMEWORK FOR PROACTIVE SLA NEGOTIATION

Khaled Mahbub and George Spanoudakis

Department of Computing, City University London, Northampton Square, London EC1V 0HB, U.K.

Keywords: Service Discovery, Service Level Agreements, Proactive SLA Negotiation, Service Monitoring.

Abstract: In this position paper we propose a framework for proactive SLA negotiation that integrates this process

with dynamic service discovery and, hence, can provide integrated runtime support for both these key

activities which are necessary in order to achieve the runtime operation of service based systems with

minimised interruptions. More specifically, our framework discovers candidate constituent services for a

composite service, establishes an agreed but not enforced SLA and a period during which this pre-

agreement can be activated should this become necessary.

1 INTRODUCTION

A service level agreement (SLA) is an explicit

contract between the provider and the consumers of

a service that defines the quality and, sometimes,

functional properties which should be guaranteed

during the provision of the service, as well as the

penalties that should be applied in case of defaulting

(Wieder, et al., 2008). An SLA is set through a

negotiation process between the provider and the

consumer of a service. This process is particularly

complex in the case of composite services since, in

order to ensure that the provision of a composite

service S is in line with the SLAs required by its

clients, the provider of S should also negotiate and

establish subordinate SLAs with the providers of the

constituent services of S. Furthermore, when a

constituent service of S becomes unavailable at

runtime or fails to perform according to its SLA, the

provider of S should be able to discover alternative

replacement services for it and negotiate SLAs with

them at runtime.

As it has been suggested by Zisman, et al.

(2008), to minimise the runtime interruption in the

provision of composite services, the discovery of

back up replacement services for their constituents

should be proactive, i.e., it should be performed

before a constituent service of S becomes

unavailable or fails to perform according to its

established SLA. Proactiveness is important since

service discovery is a time consuming activity and,

therefore, carrying it in a reactive mode, is likely to

cause significant interruption in the provision of the

composite service and violations of its own SLAs.

SLA negotiation should also be proactive as it will

be necessary to have adequate SLAs for the potential

replacement services that have been identified by

proactive discovery attempting SLA negotiation just

prior to binding to an alternative service is likely to

cause significant delay.

Existing work on service level agreements has

focused on SLA specification (Kritikos and Pernici,

2009), negotiation (Dumitrescu and Foster, 2005)

and monitoring (Mahbub and Spanoudakis, 2007;

Raimondi, et al., 2007). The need for runtime SLA

negotiation or re-negotiation has been realised by Di

Modica, et al. (2007); Parkin, et al. (2008);

Sakellariou and Yarmolenko (2005) and He, et al.

(2009), where either the terms of an SLA are revised

to accept service from an existing provider (Di

Modica, et al., 2007; Sakellariou and Yarmolenko,

2005) or a new SLA is negotiated with a new service

provider and an existing SLA is terminated (Parkin,

et al., 2008). All these approaches, however, are

reactive as they support corrective actions only after

an SLA has been violated. Thus, they may fail to

guarantee uninterrupted runtime provision of

composite services.

To address this shortcoming, in this position

paper we introduce an approach for proactive

runtime SLA negotiation. Our approach is based on

an extension of a tool for proactive runtime service

discovery which is described by Zisman, et al.

(2008). Our approach weaves SLA negotiation into

runtime service discovery and provides a clear

process model for carrying these two activities in a

251

Mahbub K. and Spanoudakis G. (2010).

A FRAMEWORK FOR PROACTIVE SLA NEGOTIATION.

In Proceedings of the 5th International Conference on Software and Data Technologies, pages 251-256

DOI: 10.5220/0003040902510256

 SciTePress

coordinated manner. It also extends a language for

expressing runtime service discovery queries that

has been developed by Zisman, et al. (2008) to

enable the specification of SLA negotiation criteria

using it. Thus, our approach provides integrated

runtime support for both proactive service discovery

and SLA negotiation that enables provision of

composite service with minimised interruptions.

Proactive SLA negotiation is weaved into the

discovery process and is performed after the

execution of service discovery queries to ensure that

adequate SLAs can be set for the discovered

services. The objective of proactive negotiation is to

establish an agreed but not enforced SLA and a

period during which the consumer of the service will

be able to activate the pre-agreement should this

become necessary. The negotiation process is also

repeated when a pre-agreed SLA comes close to

expiry and, therefore, it has to be renegotiated.

The rest of this paper is structured as follows. In

Section 2, we discuss the architecture of the

framework for proactive runtime service discovery

and SLA negotiation. In Section 3, we describe the

negotiation process. In Section 4, we briefly present

the negotaiton mechanism used in the farmewrok. In

Section 5, we review related work and finally in

Section 6, we provide some concluding remarks and

outline directions for future work.

2 PROACTIVE SERVICE

DISCOVERY AND SLA

NEGOTIATION FRAMEWORK

The architecture of our integrated service discovery

and SLA negotiation framework is shown in Figure

1. The framework consists of a runtime service

discovery tool, a service listener, a proxy negotiation

broker and a monitor, and interacts with external

service registries and event captors.

The runtime service discovery tool is used to

identify potential alternative services for the services

that a composite service uses currently. The

discovery process is driven by service discovery

queries. These queries are associated with each of

the constituent services S

of the composite service S

and specify the conditions that should be satisfied by

any service that could replace S

’s in the

composition. These conditions can refer to the

structural (interface), behavioural, contextual, and

quality characteristics that services should have in

order to be acceptable replacements for S

. Service

discovery queries can be executed in two modes: (a)

in a reactive mode where the query is executed when

the constituent service S

that it is associated with

becomes unavailable or fails to satisfy an agreed

SLA and, therefore, a replacement service should be

identified, or (b) in a proactive mode where the

query is executed in parallel with the operation of

the composite service S in order to discover and

maintain a set of candidate replacement services for

it. In the proactive execution mode, the query is

executed initially to build a set of replacement

services for S (RS) and then anytime when an event

indicating that the description of some service in RS

has been changed or a new service that could be a

candidate for inclusion in RS has emerged.

Figure 1: Architecture for proactive (and reactive) SLA

negotiation.

The proxy negotiation broker is the component that

manages the negotiation process on behalf of a

service consumer (i.e., the composite service) or a

service provider. Our architecture assumes that a

separate instance of this component is associated

with each of the two sides (the service provider and

consumer). The structure of proxy negotiation

broker is described in more detail in Section 2.1. The

negotiation process can be either reactive or

proactive. In proactive negotiation, the negotiation

process is carried out according to a two-phase

protocol that may result in a provisionally agreed

SLA but not activated SLA (see “Pre-agreed SLA”

in Figure 1) or negotiation failure. In reactive

negotiation, the negotiation process is executed

according to a single phase protocol that can result

in an agreed and activated SLA (see “Active SLA”

in Figure 1) or negotiation failure. In the framework,

a pre-agreed SLA describes a service level

ICSOFT 2010 - 5th International Conference on Software and Data Technologies

252

agreement that has been reached but not activated

yet. A pre-agreed SLA has an expiry period within

which it becomes active, if the consumer of the

service decides to activate it.

The service registry contains descriptions of

services. These should include at least a

specification of the interface of the service (WSDL)

and SLA templates indicating the terms (e.g. service

quality levels, costs etc) under which the provider of

service is typically willing to provide it. Additional

types of service descriptions that are supported by

the framework are models of service behavior

(expressed in BPEL) and further quality

characteristics that complement the SLA templates

by specifying the entire range of values for a given

characteristic.

The service listener contacts service registries

regularly to identify changes in existing service

descriptions or new services that might have become

available.

The monitor in the architecture of Figure 1 is

responsible for monitoring the provision of a service

by a given provider and the use of it by a set of

service consumers. The monitor is used to detect if

the SLA guarantee terms which should apply to the

provision of the service are satisfied, and whether

the conditions of the negotiation triggering rules of

the relevant party are satisfied in order to generate

signals for triggering negotiation.

If monitor detects (or forecasts) that the

conditions of negotiation triggering rules in the

negotiation policy of a service provider or consumer

are (or will be) violated, it informs the relevant

negotiation broker to initiate a negotiation or

renegotiation. The checks performed by the monitors

take into account events that are intercepted during

the use of services (e.g. service invocations and

responses). These events are notified to the

framework by different types of event captors that

may be associated with different services (e.g.

SOAP message captors). These events are notified to

the monitor for verifying the adherence of services

to different SLA guarantee terms and checking

whether some SLA negotiation activity should be

initiated.

The circumstances under which the negotiation

of new SLAs should start are determined by

negotiation triggering rules (e.g., when a

provisionally agreed SLA is about to expire).

Service providers and consumers may specify

separate sets of such rules, which come to force and

get monitored after the SLA is established.

2.1 Proxy Negotiation Broker

Figure 2, shows the structure of the proxy

negotiation broker. The broker deploys suitable

Negotiation Engines that are responsible for

negotiating and agreeing the guarantee terms of an

SLA. The broker wraps different types of

negotiation engines (e.g. rule driven negotiation

engine or linear programming based negotiation

engine) and provides a common interface to access

its functionalities. The Specification Translator is

responsible for generating specifications that are

necessary for the negotiation engine to conduct the

negotiation process.

Figure 2: Proxy negotiation broker and its components.

Figure 3 shows the interface of the proxy negotiation

broker. This interface offers the following methods,

 SetupNegotiation: This method allows the

service discovery process to initialize the

negotiation broker for the negotiation process.

This method accepts four parameters. The first

parameter is the string representation of an

XML document that contains the negotiation

rules. The XML document is written according

to the negotiation rule specification schema of

our framework (Mahbub and Spanoudakis,

2010). The specification translator transforms

these negotiation rules into a form that is

understood by the deployed negotiation engine.

The second parameter specifies the type of

negotiation engine that should be used in the

negotiation process (e.g. rule based or linear

programming based). The third parameter

specifies the ID of the service consumer that the

negotiation broker is negotiating for and the

fourth parameter specifies the ID of the

negotiation broker that this negotiation broker is

negotiating with. This method returns a unique

ID for the negotiation process.

 StartNegotiation: This method is used by the

service discovery process to start the negotiation

process. This method accepts one parameter that

specifies the unique ID of the negotiation

process.

 SetOffers: The Proxy Negotiation Broker uses

this method to set the offers produced by the

negotiation engine to its counterpart, i.e., the

A FRAMEWORK FOR PROACTIVE SLA NEGOTIATION

253

negotiation broker on the other side that this

negotiation broker is negotiating with. This

method accepts one parameter, which is the

string representation of an XML document that

contains an SLA template. The XML document

is written according to the SLA specification

schema of our framework (Mahbub and

Spanoudakis, 2010).

Figure 3: Interface of proxy negotiation broker.

3 SLA NEGOTIATION PROCESS

The UML activity diagram shown in Figure 4

presents the service discovery process of the

framework with the activity of SLA negotiation

embedded within it. The process starts with the

submission of a service discovery query by the

composite service (i.e., the consumer of constituent

services). As discussed in Section 2, this query can

specify different service discovery criteria, namely:

(a) structural criteria describing the interface of

required services, (b) behavioural criteria describing

the functionality of required services, and (c)

constraints describing quality characteristics of

required service. The initial execution of the service

discovery query (see the action state Execute Query

in Figure 4) results in a list of potential candidate

services (RS). The candidate services are identified

by evaluating the structural, behavioural and quality

characteristics specified in a query against the

structural, behavioural and quality of service

specifications in service registries. The execution of

the discovery query also computes distances

between a query and candidate services based on the

query criteria and ranks the candidate services based

on their distances to the query. The list of potential

candidate services is updated by executing the

service discovery query when the framework is

informed via the service listener that a new service

has become available in a registry or the description

of an existing service has been modified (see the

signal accept state New/Amended Service

Description in Figure 4). This ensures that new or

updated services are considered by the process.

After an initial set of candidate services has been

built or updated (see the action state Create/Update

Candidate Service Set), the framework selects a

service that does not have a negotiated SLA from RS

for negotiation (see the transition guarded by the

condition Exists Service in RS without Negotiated

SLA).

In the negotiation phase (i.e., the action state

Negotiate SLA), the desired level of service is

negotiated with the selected candidate service. In

this phase, the QoS characteristics of each candidate

service are negotiated in order to achieve the best

possible SLA for the services. Negotiation during

this phase may fail and, if this happens, for a

selected candidate service then the service is

removed from RS and a new negotiation will start

with another candidate service in RS which does not

have a negotiated SLA. If the negotiation with a

selected service succeeds, a provisional SLA is

established and the selected service in RS is updated

to flag the existence of the pre-agreed SLA.

Figure 4: SLA negotiation process.

It should be noted that, the negotiated SLAs for the

services in RS do not come into force immediately.

For each pre-agreed SLA, the negotiation process

establishes a time period over which the pre-agreed

SLA can be automatically brought into force without

further negotiation. This will happen if the relevant

service is selected for binding to the composite

service. If the validity period of a pre-agreed SLA

comes close to expiry without the candidate service

being bound to the composite service, the

framework will proactively re-negotiate the SLA

(see the transition guarded by the condition Pre-

agreed SLA about to expire, from the action state

Create/Update Candidate Service Set to the action

state Select Service RS for Negotiation).

Following the selection of a service in RS for

binding at runtime, its SLA is automatically

enforced (see the action state Activate SLA in Figure

4). When an SLA comes into force, its guarantee

ICSOFT 2010 - 5th International Conference on Software and Data Technologies

254

terms become subject of monitoring (see the action

states Receive SLA and Service & SLA Monitoring in

Figure 4). If the monitoring process detects violation

of the SLA or the deployed service becomes

unavailable then the service is replaced by the best

available service in RS (see the transition from the

action state Service & SLA Monitoring to the action

state Select & Bind Best Service in RS). The

detection of violation of the conditions in

negotiation triggering rules (e.g. active SLA about to

expire) triggers the negotiation phase to establish a

new SLA.

4 NEGOTIATION MECHANISM

Although, as discussed in Section 2, our framework

supports the deployment of different types of

negotiation engines, we use a linear programming

based negotiation engine as the default. This is

because linear programming enables relatively quick

optimal solution in decision making (Megiddo

1987). Despite this advantage, the use of linear

programming techniques in the context of SLA

negotiation has received limited attention, and the

few approaches that apply linear programming for

SLA management have certain limitations. For

example, Hung (2003) assumed that participating

parties expose their preferences over the negotiation

issues in order to formulate the linear program. In

reality, however, participants might not be willing to

disclose information about their own preferences.

Cradellini, et al. (2007) used linear programming to

set and tune the provision of an agreed SLA rather

than in SLA negotiation.

In our framework, each participant expresses its

preferences over negotiation issues by a set of

negotiation rules. The proxy negotiation broker

transforms the negotiation rules into a linear

program and solves it to generate SLA offers during

the negotiation process. A negotiation rule in our

framework has the generic structure:

IF (condition) THEN (action) ELSE (action)

Conditions in these rules are expressed as atomic

conditions over quality attributes of services or

logical combinations of atomic conditions. Rule

actions can be of three types: (i) accept actions

which enable the acceptance of the value of one or

more attributes in a given SLA offer, (ii) reject

actions which enable the rejection of the value of

one or more QoS attributes in a given SLA offer,

and (iii) set actions which allow to define a new

value or range of values for one or more QoS

attribute.

Negotiation rules are transformed into linear

inequalities applying the techniques discussed by

Nielsen, et al. (2000); Costa and Monteiro (2003).

For example, a negotiation rule of the form,

IF (p > 0) THEN (q >=0)

can be transformed into the following set of

inequalities,

where, B is a binary variable and C1, C2, C3 are

constants whose values should be large enough.

5 RELATED WORK

There is a substantial amount of work related to SLA

negotiation. The parts of it that are more relevant to

our approach relate to runtime negotiation and

management of compensations for SLA violations.

The provision of compensation in case of

violation of SLA is argued by Rana, et al. (2007)

and He, et al. (2009). These approaches claim that

the penalty clauses in the SLA should not only

specify the monetary penalties or impact on potential

future agreements between the parties but also

several other issues such the law that will be applied

in cases where a conflict between the provider and

the client arise, and the impact of the penalty clauses

on the choice of service level objectives.

Runtime renegotiation is suggested by Wieder, et

al. (2008); Sakellariou and Yarmolenko (2005); Di

Modica, et al. (2007) and Parkin, et al. (2008), to

manage SLA violations. Service level objectives are

revised and renegotiated at runtime and the deployed

service is adjusted to the newly agreed service level

objectives by Di Modica, et al. (2007). A similar

approach which allows changes in service level

objectives whilst keeping an existing SLA is

described by Sakellariou and Yarmolenko (2005).

Parkin, et al. (2008) described a renegotiation

protocol that allows the service consumer or service

provider to initiate renegotiation while the existing

SLA is still in forced. In this protocol any party may

initiate the renegotiation and after a successful

renegotiation the existing SLA is superseded by a

new contract.

The approaches discussed above are reactive in

nature, i.e., renegotiation starts only after an existing

SLA is violated. The outcome of renegotiation is

either a revised set of service level objectives

0 < P + C1*B

0 <= Q + C2*B

P <= 0 + C3*(1-B)

A FRAMEWORK FOR PROACTIVE SLA NEGOTIATION

255

allowing the acceptance of a service from an

existing provider or a new SLA for a new service

provider terminating the existing SLA. Hence, all

these approaches either affect the quality of the

delivered service or fail to guarantee uninterrupted

service. Unlike them, our approach weaves proactive

SLA negotiation into dynamic service discovery to

enable the runtime operation of service based

applications with minimised interruptions.

6 CONCLUSIONS AND FUTURE

WORK

This position paper proposes a framework that

integrates service discovery, monitoring and

proactive SLA negotiation. The service discovery

process is used by service consumers (i.e., service

based applications) in order to identify potential

alternative services for the services that they

currently use. The framework supports the proactive

negotiation of SLAs with each alternative service

prior to its deployment. The negotiation process is

carried out according to a two-phase protocol that

may result in a provisionally agreed SLA but not

activated SLA or negotiation failure. A provisional

SLA is a service level agreement that has been

agreed by a service provider and a service consumer

but has not been activated yet. Such an SLA has an

expiry date by which it will either be activated or

cease to exist.

The presented framework has also opened broad

scope of future investigations. For example the

framework can be extended to support proactive

negotiation for hierarchical SLA i.e. a complex SLA

can be decomposed into several SLAs and

negotiated separately to come to a final agreement.

Also the framework can be extended to support

dynamic adaptation of the negotiation rules, i.e. the

participants will be able to dynamically change the

negotiation rules during the negotiation process.

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).

REFERENCES

Cardellini V., Casalicchio E., Grassi V., Lo Presti F.,

2007. Efficient provisioning of service level agreements

for service oriented applications. In Proceedings of

IW-SOSWE at ESEC/FSE.

Costa J. C., Moneiro J. C., 2003. Input Generation for Path

Coverage in Software Testing. IEEE Workshop on

Compilers and Tools for Constrained Embedded

Systems

Di Modica G., Tomarhio O. and Lorenzo V., 2007. A

framework for the management of dynamic SLAs in

composite service scenarios. ICSOC Int. Workshops,

Dumitrescu C. L., Foster I., 2005. GRUBER: A Grid

Resource Usage SLA Broker. Int. Euro-Par

conference.

He Q., Yan, J., Kowalczyk R., Hai J., Yang Y., 2009.

Lifetime Service Level Agreement Management with

Autonomous Agents for Services Provision.

Information Sciences, Elsevier.

Hung P. C. K., 2003. A Primitive Study of Logrolling in e-

Negotiation. 36thHawaii Int. Conf. on System Sciences

Kritikos K., Pernici B., (editors), 2009. Initial Concepts

for Specifying End-to-End Quality Characteristics and

Negotiating SLAs. S-Cube project deliverable: CD-

JRA-1.3.3. http://www.s-cube-

network.eu/achievements-results/s-cube-deliverables.

Mahbub K., Spanoudakis G., 2007. Monitoring WS

Agreements: An Event Calculus Based Approach, Test

and Analysis of Service Oriented Systems, (eds) L.

Baresi, E. diNitto, Springer.

Megiddo N., 1987. On the complexity of linear

programming. In Advances in economic theory Fifth

world congress, Cambridge University Press

Nielsen S. R., Pisinger D., Marquardsen P., 2000.

Automatic Transformation of Constraint Satisfaction

Problems to Integer Linear Form - an Experimental

Study. In Proc. of Techniques foR Implementing

Constraint programming Systems,

Parkin M., Hasselmeyer P., Koller B., Wieder P., 2008.

An SLA Re-negotiation Protocol. In proceedings of

the 2nd Workshop on Non Functional Properties and

SLAs in Service Oriented Computing

Raimondi, F., Skene, J., Chen L., Emmerich, W., 2007.

Efficient monitoring of web service SLAs. Technical

report. Research Notes (RN/07/01). UCL, London,.

Rana O., Warnier M., Quillinan T. B., Brazier F. and

Cojocarasu D., 2007. Managing Violations in Service

Level Agreements. In Proceedings of the Usage of

Service Level Agreements in Grids Workshop.

Sakellariou R., Yarmolenko V., 2005. On the Flexibility

of WS-Agreement for Job Submission. Proc. of 3rd

Int. Workshop on Middleware for Grid Computing,

Mahbub K, Spanoudakis G., 2010. SLA Specifications.

http://www.soi.city.ac.uk/~am697/sla/SLA-Spec.zip

Wieder P., Seidel J., Wäldrich O., Ziegler W., Yahyapour

R., 2008. Using SLA for Resource Management and

Scheduling - A Survey. Grid Middleware and Services

Challenges and Solutions, Springer.

Zisman A., Spanoudakis G., Dooley J., 2008. A

Framework for Dynamic Service Discovery. 23rd Int.

Conf. on Automated Software Engineering.

ICSOFT 2010 - 5th International Conference on Software and Data Technologies

256