A MAS TO MANAGE AND MONITOR SLA
FOR CLOUD COMPUTING
A Draft Position Paper
Benjamin G
ˆ
ateau
Centre de Recherche Public Henri Tudor, SISE Unit - Service Science & Innovation Dept
29 Av. John F. Kennedy, L1855 Luxembourg, Luxembourg
Keywords:
Dynamic infrastructure, Normative organization, SLA management.
Abstract:
The More than a technological solution, the Cloud Computing is also an economical advantage and will play
an important roles in next years. However, in order to ensure a QoS commitment between a provider and a
customer, Service Level Agreements (SLA) describe a set of non-functional requirements of the service the
customer is buying. SLA is the best way to ensure QoS. In this paper, we use a MAS to manage SLA by
monitoring the respect of service in the context of Cloud Computing.
1 INTRODUCTION
Evolution of high level ICT infrastructures in Europe
brings some difficulties due by a complex manage-
ment and the need of more and more energy. One
current fashionable solution to this problem is the use
of cloud computing services. Cloud computing is a
ICT model allowing an easy access through networks
to mutualised and configurable resources able to be
quickly activated and deactivated.
Cloud computing has become a mainstream tech-
nology offering mutualisation of IT infrastructures as
services along several paths such as Software (SaaS),
Platform (PaaS), and Infrastructure (IaaS). Compa-
nies such as Amazon, Microsoft, IBM, and Google,
to name but a few, offer such services, which rely
on virtualization and pay-as-you-go business models.
Flexibility and elasticity are also important features
of Cloud Computing made possible by the concept of
Dynamic Infrastructure.
Dynamic Infrastructure is an information technol-
ogy paradigm concerning the design of data centers
so that the underlying hardware and software can re-
spond dynamically to changing levels of demand in
more fundamental and efficient ways than before. The
basic premise of Dynamic Infrastructures is to lever-
age server virtualization technology to pool comput-
ing resources wherever possible, and then to allocate
these resources on-demand. This allows for load bal-
ancing and is a more efficient approach than keeping
massive computing resources in reserve to run tasks
that take place, for example, once a month. The
potential feature benefits include enhancing perfor-
mance, scalability system availability and uptime, and
the ability to perform routine maintenance on either
physical or virtual systems all while minimizing in-
terruption to business operations and reducing cost
for IT. Dynamic Infrastructures also provide the fun-
damental business continuity and high availability re-
quirements to facilitate cloud or grid computing.
Figure 1: VMware Infrastructure example (source:
VMware).
Fig. 1 illustrates an example of dynamic infras-
tructure based on VMware solution. Physical servers
of the datacenter stores virtual machines that can be
loaded, managed and configured remotly by the cus-
tomer of the service (Iaas kind in this case). For that
the customer previously specifies his needs in term of
687
Gâteau B..
A MAS TO MANAGE AND MONITOR SLA FOR CLOUD COMPUTING - A Draft Position Paper.
DOI: 10.5220/0003558206870692
In Proceedings of the 1st International Conference on Cloud Computing and Services Science (MAS-2011), pages 687-692
ISBN: 978-989-8425-52-2
Copyright
c
2011 SCITEPRESS (Science and Technology Publications, Lda.)
distribution (OS), CPU, memory, disk space, band-
with, etc. to define the virtual machine(s) he will pay
for (pay on demand and based on the configuration he
ask for).
Figure 2: Moving VM in a VMware Infrastructure.
Dynamic infrastructures allow the use of resource
when it is needed. For instance, Fig. 2 represents the
case when clients of the service provider don’t use
their virtual machine at a certain time (night for in-
stance). It could happen that only few virtual ma-
chines are running on each physical machine. So
physical servers are not fully used and gathering vir-
tual machines on one physical server could permit
to save energy and use less resource. But by mov-
ing virtual machines, the provider must ensure that
the customers pay for the service he previously spec-
ified and that the Quality of Service (QoS) is still the
same. QoS delivery affects the value of the service for
the client and significantly depends on IaaS or PaaS
provider’s infrastructure. Therefore there is need to
divide responsibility/risk between XaaS provider and
customer. To describe the responsibilities a formal
description of non-functional requirements from the
client’s point of the view is required.
Service Level Agreements (SLA) permit to spec-
ify the needs and offers of customer and the service
provider. To summarize the have to define some SLA
in order to .
The aim of our work is to define a SLA infras-
tructure that provides the same guarantees and proofs
that we can find e-contract management, with more
flexibility. In this paper we are considering a datacen-
ter providing IaaS remotly an automatically (such as
GoGrid
1
or Cloud.com
2
for instance) through a web-
based application. This application mediates the in-
teractions between customers and providers. Agree-
ment between parties are represented as SLAs and are
specified with a set of web interfaces defining value of
KPI that the provider has to achieve (king of rules or
obligations). To this aim, we will use an e-contract
1
www.gogrid.com
2
www.cloud.com
model to enable an agent-based execution and man-
agement of them on one hand, and, on the other hand,
we will add a monitoring system to control and en-
force the SLAs between provider and customers.
This paper is organized as follows. We make a
quick survey on SLA management in the next sec-
tion, then we present our normative multi-agent sys-
tem solution before proposing a solution to monitore
SLA with an electronic institution based on multi-
agent system for the cloud computing.
2 SLA MANAGEMENT
SLA describes a set of non-functional requirements
of the service the customer is buying. The agreement
usually contains also penalties when the requirements
are not met. An example of a non-functional require-
ment would be ”RTO - Return to Operation time for
a service in case of a failure. To describe a non-
functional requirement we needed an objective to be
achieved (e.g. RTO under two minutes ) and a set of
indicators that prove the objective is met (e.g. new in-
stance bootstrap time). The objective to be achieved
is called ”Service Level Objective (SLO) and the indi-
cators are called ”Key Performance Indicators (KPIs).
Service Level Objective (SLO) is the objective of ser-
vice quality that has to be achieved. It is represented
by a set of measurable KPIs with thresholds to decide
if the objective is fulfilled or not. The fulfillment of
an SLOs describes a state of service when all of the
SLOs key performance indicators are within a spec-
ified thresholds. KPIs usually consist from one or
more raw monitored values including min, avg and
max specifying the scale. They can also represent
some aggregated measurement (e.g. average output)
within a sliding window that is combined from one or
more monitoring outputs. The provider has to be able
to measure and affect the KPIs otherwise it would not
make sense to guarantee them. The Cloud Comput-
ing infrastructures are usually large scale, therefore
SLAs need to be formally described to enable their
automated handling and protection.
Automated SLA protection is based on a set of
policy rules. Each policy rule is formed by one or
more conditions (KPI’s value matching pattern) and
one or more actions. KPIs are periodically evaluated
according to defined policies. If one or more condi-
tions are met, then appropriate actions are triggered.
An example of the policy action can be increasing
number of service instance. The action is triggered
by a server load KPI matching the policy rule. This
enables to automatically keep the load KPI under cer-
tain value specified by the SLO and avoid violating
CLOSER 2011 - International Conference on Cloud Computing and Services Science
688
the general SLA. These kinds of rules are named elas-
ticity rules.
Specification of the policies (rules and action) is a
complex task in large scale dynamic system. Analy-
sis of historical KPI data and triggered actions can be
used to specify new or modify existing policies. This
enables system adaptation and automated SLA pro-
tection evolution. System can for example learn from
historical data about periodical service load peaks
and generate specific rules to keep the system with
throughput specified in SLA . Another research chal-
lenge is the specification of models able to describe
and simulate these kinds of dynamic systems.
The SLA can also contain certain penalties if one
or more of the SLOs is broken. The amount of a
penalty usually differs with the SLO guarantee class
(e.g. Bronze, Silver, Gold) . This allows to overbook
resources with a certain risk that some of the SLOs
will be broken (e.g. Some of the Bronze SLO are not
going to be fulfilled in favor of some Gold SLOs that
are connected with higher penalties). The system can
be then optimized to minimize the penalties and max-
imize the utilization.
With the development of the Internet and Virtual
Enterprises appearance, e-Services and e-Contracting
in general have gained an increasing interest in the
business domain. While some B2B applications han-
dle electronic contracts (e-contract) and digital signa-
tures as digitalized paper contracts (Laurikkala and
Tanskanen, 2002), an increasing number of them
aim at increasing their enactment and management
(e.g. (Koetsier et al., 2000)). We consider in this paper
that the management of a SLA defining a set of SLO
which have to be respected is the same that manag-
ing an e-contract composed by a set of clauses repre-
sented by obligations which have to be also respected.
Indeed, in his PhD thesis relating to E-contract
modeling and e-enactement (Krishna, 2010), P. Radha
Krishna describe the clauses of an electronic contract
as obligations being part of a SLA. He also says:
E-contract management solutions should maintain,
monitor and manage contract rules derived from
these SLAs. Contract parties should verify QoS pa-
rameters by performing an SLA monitoring, which in-
volves monitoring the performance status of the of-
fered service. The e-contract management system
could assess the SLA requirements and apply penal-
ties if there is any deviation.
The SLA4D-Grid project (Wieder et al., 2009)
defines a SLA management layer on top of an ex-
isting infrastructure providing e-contracting capabil-
ities. The infrastructure specifies, implemnts and de-
ploys a SLA-based service stack for e-Contracting.
The authors say: The SLA4D-Grid project is design-
ing and implementing an SLA management layer. The
functions of the developments cover the complete SLA
lifecycle, including SLA design, contract establish-
ment, SLA provisioning, and SLA monitoring.”.
The SLA@SOI project (Comuzzi et al., 2010) is
a FP7 project dealing with the definition of a SLA
management framework. The consortium defined a
reference architecture, specified a SLA template and
the methodology to translate SLAs inot monitoring
specifications (for the EVEREST environment). With
the RESERVOIR project, they the defined how us-
ing cloud standards for the interoperability of cloud
frameworks.
In this global trend, multi-agent technologies have
been introduced to achieve the automation in creation,
execution and monitoring of e-contracts by agents on
behalf of users. Lots of work deal with negotiation
of the issues related to the content of contract and
to their execution(e.g. (Dignum and Sierra, 2001)).
The resulting contracts consist in digital agreement
between contractual parts where rights and duties in
terms of deliverables, costs and delays of the partic-
ipants are explicitly represented. However such con-
tracts often lead to inflexible relations between par-
ticipants. The obtained result is in contrary to the
requirements of open and dynamic system that are
stressed by the actual business paradigms aiming at
improving the competitiveness of companies like dy-
namic virtual enterprises and dynamic service out-
sourcing (Hoffner et al., 2001). Moreover, few of the
existing research works take into account the moni-
toring of contracts clauses (Padovan et al., 2002) that
bind agents together.
In (Boissier and G
ˆ
ateau, 2007) we proposed an
Electronic Institution model based on MAS to man-
age electronic contract by specifying obligations. In
this paper we aim at doing the same for the manage-
ment of SLA for Cloud Computing. In the next sec-
tion we present M OISE
Inst
and S YNAI, normative or-
ganization modelling and middleware.
3 NORMATIVE MULTI-AGENT
SYSTEM
In this work we propose a multi-agent support for the
enactment and monitoring of the different SLO speci-
fied in the SLA. SLAs specify the agreement between
customers and the provider concerning their partic-
ipation to the distributed execution of the job. A
SLA must describe both the functioning and the struc-
ture organizing this functioning. Moreover it con-
tains explicit legal dimensions bearing on the involved
participants. In order to take this into account we
A MAS TO MANAGE AND MONITOR SLA FOR CLOUD COMPUTING - A Draft Position Paper
689
propose to use a “normative organizational model”,
called M OISE
Inst
, to express SLA. This normative or-
ganizational model is accompanied by a specialized
“normative middleware”, called S YNAI
3
to monitor
and enforce legal aspects expressed in the SLAs.
3.1 Normative Organization Modelling
M OISE
Inst
(G
ˆ
ateau et al., 2007) is founded on the
M OISE
+
organizational model
4
(Hubner et al., 2002).
It is composed of the following components that are
used to specify an organisation of agents in terms of
structure, functioning, evolution and norms (OS of the
Fig. 3):
A Structural Specification (SS) defines: (i) the
roles that agents will play in the organization, (ii)
the relations between these roles in terms of au-
thority, communication or accointance, (iii) the
groups, additional structural primitives used to de-
fine and organize sets of roles;
A Functional Specification (FS) defines global
business processes that can be executed by the dif-
ferent agents participating to the organization ac-
cording to their roles and groups;
A Contextual Specification (CS) specifies, a pri-
ori, the possible evolution of the organization in
terms of a state/transition graph;
A Normative Specification (NS) defines the deon-
tic relations gluing the three independant specifi-
cation (SS, FS, CS). This NS clearly states rights
and duties of each roles/groups of SS on sets of
goals (missions) of FS, within specific states of
CS.
A BNF
5
definition of OS is available in (G
ˆ
ateau
et al., 2004) for SS and FS and in (G
ˆ
ateau et al., 2005)
for CS and NS and in (G
ˆ
ateau, 2007) for the whole
model.
3.2 Normative Organization
Middleware
In human societies, institutions define rules (North,
1990) that enclose all kinds of formal or informal
constraints that human beings use to interact. These
last years, electronic institutions have been intro-
duced in multiagent domain. They propose to model
these rules with normative systems (Jones and Carmo,
2001). Institutions are defined as a set of agents,
3
S YNAI: SYstem of Normative Agents for Institution.
4
M OISE
+
: Model of Organization for multI-agent Sys-
tem.
5
BNF: Backus-Naur Normal Form.
Figure 3: M OISE
Inst
, normative organization specification
model, and S YNAI, normative middleware.
which behave according to norms taking into account
their possible violation. The functioning of the agents
is supervised and controled with a set of institution
services (e.g. (Dellarocas, 2000)). The institution ser-
vices that support and use M OISE
Inst
are regrouped in
a specific “normative middleware” called S YNAI on
which the agents execute. This layer is in charge of :
(i) managing the life cycle of SS as entering/exiting of
agents within the organization, or requesting/leaving
of roles or groups by the agents, (ii) coordination of
the concurrent execution of FS as commitment to mis-
sions or achievement of goals, etc, (iii) dynamic and
evolution of the organization state through the CS,
(iv) the monitoring and supervision of norms of NS
activated/deactivated by the evolution of the organi-
zation.
Using both S YNAI and M OISE
Inst
, we are able to
define a SLA management system as a set of agents
whose behavior is ruled by an organization expressed
in M OISE
Inst
and controlled by an arbitration sys-
tem, implemented with S YNAI, that has the possibil-
ity to reward or to punish agents in case they respect
or not their agreements as expressed in SLAs. Using
M OISE
Inst
, agents used on the provider infrastructure
are able to reason and to take into account the differ-
ent SLAs described with M OISE
Inst
. The S YNAI plat-
form on its side take into account this specification in
order to supervise and control virtual resources. Both
layers are based on an agent execution platform.
These four specifications form the Organizational
Specification (OS). The Organizational Entity (OE) is
then built from the set of agents that have adopted a
role according to the SS of the OS. From this time the
S YNAI middleware manages and controls the func-
tioning of this OE (composed of active contexts, ac-
tive norms, current structure and current functioning)
by the way of different events corresponding to the
entry/exit of agents of the OE, adoption/leaving roles
or groups, change of context, commitment to mis-
sions, achievement of goals.
CLOSER 2011 - International Conference on Cloud Computing and Services Science
690
4 SLA MANAGEMENT
ARCHITECTURE PROPOSAL
The framework we propose is based on classic
VMware or other virtualisation architecture used by
a IaaS provider. In our solution, each physical
server is represented by an agent belonging to a
M OISE
Inst
organization. An organization specified
with M OISE
Inst
represents a SLA that a customer
could defined. For instance, with help of a web in-
terface, the customer build his virtual network com-
posed of database server, web server, storage server
and application server linked between them. Each
component has its own characteristic as seen before
(CPU, memory, etc.). By drawing its configuration,
the customer specifies the Structural Specification of
the Organization (a machine selected is a role that a
physical server will play through a virtual instance)
and the Functional and Normative Specification (each
role will have the Obligation to achieve the character-
istics defined for each virtual machine which could be
considered as the SLO of the SLA).
Figure 4: M OISE
Inst
as SLA template in Cloud Computing
architecture.
On the Fig. 4, we represents on the left the infras-
tructure coming from the datacenter. The agents (in
blue) are launched on each physical server and repre-
sent them. By creating specific virtual machines, they
instantiate the current structure of an organization (a
topology buyed by a customer, i.e. a OE instantiating
an OS).
The agent act via the S YNAI layer which monitor
the agent. The monitoring is used for:
detecting that a virtual machine violates the SLA
for any reason and triggering the agent playing the
role (i.e. the physical machine hosting the virtual
machine) have to modify the VM configuration in
order to respect a specific SLO of the SLA (a norm
of the OE)
showing to the customer the state of the OE re-
garding the OS (transparency purpose)
5 CONCLUSIONS
Service Level Agreement (SLA) describes agreement
on non-functional requirements between provider and
customer. SLA consists of service level objectives
(SLOs) that are evaluated according to measurable
Key Performance Indicators (KPIs). Automatic SLA
protection enables further increase of the system uti-
lization and system profit. In currently available sys-
tems only some basic SLAs like ”uptime over a time
period guarantee are available.
This part of cloud computing is target of intensive
research now. Formal ways of SLA description have
to be standardized. Algorithms and models for the
resources allocation, automated SLA protection and
evolution has to be found.
These areas are for example in the scope of
RESERVOIR (Resources and Services Virtualization
without Barriers) - EU FP7 funded project
6
backed
by consortium of industry and academy partners.
Future works will concern the specification of
the OS with M OISE
Inst
and its implementation with
Utopia (Schmitt et al., 2010; Schmitt et al., 2011). A
national project funded by the Fond National of Re-
search
7
in Luxembourg will permits to also work on
security and risk through SLA management in cloud
computing. We aim at gathering monitoring QoS and
security in the same way.
REFERENCES
Boissier, O. and G
ˆ
ateau, B. (2007). Normative multi-
agent organizations: Modeling, support and con-
trol, draft version. In G. Boella, L. v. d. T. and
Verhagen, H., editors, Normative Multi-Agent Sys-
tems. Dagstuhl Seminar Proceedings 07122, Interna-
tionales Begegnungs- und Forschungszentrum fuer In-
formatik (IBFI), Schloss Dagstuhl, Germany.
Comuzzi, M., Kotsokalis, C., Rathfelder, C., Theilmann,
W., Winkler, U., and Zacco, G. (2010). A framework
for multi-level sla management. In Dan, A., Gittler,
F., and Toumani, F., editors, Service-Oriented Com-
puting. ICSOC/ServiceWave 2009 Workshops, volume
6275 of Lecture Notes in Computer Science, pages
187–196. Springer Berlin / Heidelberg.
Dellarocas, C. (2000). Contractual agent societies: Ne-
gotiated shared context and social control in open
multi-agent systems. In Proceedings of the Workshop
6
http://www.reservoir-fp7.eu
7
FNR: www.fnr.lu.
A MAS TO MANAGE AND MONITOR SLA FOR CLOUD COMPUTING - A Draft Position Paper
691
on Norms and Institutions in Multi-Agent Systems,
4th International Conference on Multi-Agent Systems
(Agents-2000), Barcelona, Spain.
Dignum, F. and Sierra, C., editors (2001). Agent-Mediated
Electronic Commerce - The European AgentLink Per-
spective, volume 1991 of Lecture Notes in Artificial
Intelligence. Springer Verlag. ISBN 3-540-41671-4.
G
ˆ
ateau, B. (2007). Modlisation et Supervision
d’Institutions Multi-Agents. PhD thesis, Ecole
Nationale Superieure des Mines de Saint Etienne.
defended at CRP Henri Tudor, Luxembourg.
G
ˆ
ateau, B., Boissier, O., Khadraoui, D., and Dubois, E.
(2005). M OISE
Inst
: An organizational model for
specifying rights and duties of autonomous agents.
In van der Torre, L. and Boella, G., editors, 1st In-
ternational Workshop on Coordination and Organisa-
tion (CoOrg 2005) affiliated with the 7th International
Conference on Coordination Models and Languages,
Namur - Belgium.
G
ˆ
ateau, B., Boissier, O., Khadraoui, D., and Dubois,
E. (2007). Controlling an interactive game with a
multi-agent based normative organizational model. In
Vzquez-Salceda, J., Boella, G., Boissier, O., and Mat-
son, E., editors, ”Coordination, Organizations, Insti-
tutions, and Norms in Agent Systems II”, volume 4386
of LNCS, pages 86–100. Springer Berlin / Heidelberg.
ISBN: 978-3-540-74457-3.
G
ˆ
ateau, B., Khadraoui, D., and Dubois, E. (2004). Architec-
ture e-business s
´
ecuris
´
ee pour la gestion des contrats.
In 3
`
eme Conf
´
erence sur la S
´
ecurit
´
e et Architectures
R
´
eseaux (SAR), La Londe, Cote d’Azur - France.
Hoffner, Y., Field, S., Grefen, P., and Ludwig, H. (2001).
Contract-driven creation and operation of virtual en-
terprises. Comput. Networks, 37(2):111–136.
Hubner, J. F., Sichman, J. S., and Boissier, O. (2002).
M OISE
+
: towards a structural, functional, and de-
ontic model for mas organization. In Proceedings
of the first International Joint Conference on Au-
tonomous Agents and MultiAgent Systems, pages 501–
502, Bologna, Italy. ACM Press. ISBN 1-58113-480-
0.
Jones, A. and Carmo, J. (2001). Handbook of Philosophical
Logic, chapter Deontic logic and contrary-to-duties,
pages 203–279. Kluwer.
Koetsier, M., Grefen, P. W. P. J., and Vonk, J. (2000). Con-
tracts for cross-organizational workflow management.
In EC-WEB ’00: Proceedings of the First Interna-
tional Conference on Electronic Commerce and Web
Technologies, pages 110–121, London, UK. Springer-
Verlag. ISBN 3-540-67981-2.
Krishna, P. R. (2010). E-contract modeling and e-
enactement. PhD thesis, International Institute of In-
formation Technology, Hyderabad - 500 032, INDIA.
Laurikkala, H. and Tanskanen, K. (2002). Managing con-
tracts in virtual project supply chains. In PRO-VE
’02: Proceedings of the IFIP TC5/WG5.5 Third Work-
ing Conference on Infrastructures for Virtual Enter-
prises, pages 93–100, Deventer, The Netherlands, The
Netherlands. Kluwer, B.V. ISBN 1-4020-7020-9.
North, D. C. (1990). Institutions, Institutional Change and
Economic Performance. Political Economy of Institu-
tions and Decisions. Cambridge University Press, 1st
edition edition. ISBN 0521397340.
Padovan, B., Sackmann, S., Eymann, T., and Pippow,
I. (2002). A prototype for an agent-based secure
electronic marketplace including reputation tracking
mechanisms. International Journal of Electronic
Commerce, 6(4):93–113.
Schmitt, P., Bonhomme, C., Aubert, J., and G
ˆ
ateau, B.
(2011). Programming electronic institutions with
utopia. In Aalst, W., Mylopoulos, J., Sadeh, N. M.,
Shaw, M. J., Szyperski, C., Soffer, P., and Proper, E.,
editors, Information Systems Evolution, volume 72 of
LNBIP, pages 122–135. Springer Berlin / Heidelberg.
ISBN 978-3-642-17722-4.
Schmitt, P., Bonhomme, C., and G
ˆ
ateau, B. (2010). Easy
programming of agent based electronic institution
with utopia. In In proceedings of the 10th interna-
tional conference on New Technologies of Distributed
Systems (NOTERE 2010), Tozeur - Tunisia.
Wieder, P., Hasselmeyer, P., and Koller, B. (2009). En-
hancing a national academic computing infrastructure
with e-contracting capabilities. In Cunningham, P. and
Cunningham, M., editors, Proceedings of eChallenges
e-2009 Conference. ISBN: 978-1-905824-13-7.
CLOSER 2011 - International Conference on Cloud Computing and Services Science
692