A Multi-Agent based Architecture for Cloud Infrastructure

Auto-adaptation

Hanen Chihi

, Walid Chainbi

and Khaled Ghedira

University of Tunis, Higher Institute of Computer Science/SOIE, Tunis, Tunisia

University of Sousse, Sousse National School of Engineers/SOIE, Sousse, Tunisia

University of Tunis/SOIE, Tunis, Tunisia

Keywords: Cloud Computing, Self-adaptation, Autonomic Computing, Multi-Agent Systems.

Abstract: Cloud computing including hardware, software, communication and networks are growing towards an ever

increasing scale and heterogeneity, becoming overly complex. To manage such growing complexity,

autonomic computing focuses on self-adaptable computing systems to the maximum extent possible without

human intervention or guidance. In this paper, we design a multi-agent-based architecture for Cloud

infrastructure auto-adaptation. This model is based on distributed multi-agent systems which collaborate to

enrich the Cloud with self-* capabilities. The proposed model represents an effective method to reduce

servers’ power consumption while achieving the required performance and providing trusted Cloud.

1 INTRODUCTION

In the recent years, Cloud computing has attracted

considerable attention (Mell and Grance, 2011). The

concept of Cloud computing abstracts the runtime

infrastructure to the user. It is a type of distributed

system consisting of a set of interconnected and

virtualized computers based on service-level

agreements established through negotiation between

providers and users. In this way, users will be able to

access Cloud data and applications through Internet

anywhere, at any time, on-demand and pay-per-use

(Hurwitz et al., 2009).

To adapt the dynamic behaviours of Cloud

infrastructure and to meet performance

requirements, fault tolerance, reliability, security,

etc., without manual intervention, techniques and

methodologies are needed to be designed and

implemented. Autonomic Computing (AC) may help

providers and users to reach this goal. AC is a

suitable candidate to endow the Cloud with self-*

capability (Solomon et al., 2010).

An autonomic system is a set of autonomic

elements, which implement intelligent control loops

to monitor, analyse, plan and execute using

knowledge of the environment. Several research

efforts focused on enabling the autonomic properties

address four main areas (IBM, 2003): Self-

Configuring, Self-Optimizing, Self-Healing, Self-

Protecting.

Many researchers have addressed Cloud

autonomic behaviors at all levels, from the hardware

level to software systems and applications (IBM,

2003). At the hardware level, data-center and

operating system may be dynamically adapted and

reconfigured. At the application level, databases,

application server and web servers may be

dynamically reconfigured to satisfy needed

performance. Moreover, efforts have also focused on

autonomic middleware, programming systems and

runtime (Van et al., 2009).

Deriving from distributed artificial intelligence

technology, multi-agent technology rising recent

years is a fusion of distributed computing, artificial

intelligence and other disciplines of programming

methods. Agent-based approaches have been a

source of technologies to a number of research areas.

These include distributed planning and decision-

making, communication languages, automated

auction mechanisms, coordination mechanisms and

learning mechanisms. Moreover, agent technology

offers key advantages for the development of AC

systems as it supports autonomy, adaptability, etc.

Motivated by the fact that agent technology is

already ready for being integrated into a framework

of AC (Chainbi, 2010), Multi-Agent Systems (MAS)

can be used as basic components for implementing

Chihi H., Chainbi W. and Ghedira K..

A Multi-Agent based Architecture for Cloud Infrastructure Auto-adaptation.

DOI: 10.5220/0004374700900095

In Proceedings of the 3rd International Conference on Cloud Computing and Services Science (CLOSER-2013), pages 90-95

ISBN: 978-989-8565-52-5

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

intelligence in Clouds making them more adaptive,

flexible, autonomic and smarter.

This work focuses on the Infrastructure-as-a-

Service delivery Cloud computing class. It proposes

a general multi-agent-based architecture for Cloud

infrastructure adaptation. Agents are used to help

the intelligent provisioning of basic resources to user

applications, optimize the use of infrastructure

provided as services and the management of the

essential hardware maintaining the requested QoS,

protect and repair Cloud infrastructure.

The remainder of this paper is organized as

follows. Section 2 presents the related work. Section

3 describes the proposed MAS-based architecture

for Cloud infrastructure management. Cases studies

are detailed in Section 4. Finally, we conclude our

study in Section 5.

2 RELATED WORK

Over the last decade a number of researchers

describe the use of intelligent agents in service

oriented architecture based system (Li et al., 2008);

(Overeinder et al., 2008); (Chainbi, 2010). Agents

can effectively and automatically provide some

basic levels of computer and network defense that

could be applied to Cloud infrastructure. In addition,

intelligent agents offer various solutions to control

access and authentication, distributed trust

management, audit and intrusion detection, and

diagnostic and system restoration.

Mazur et al. (Mazur et al., 2011) propose an

autonomic monitoring defensive mechanism for

Cloud integrating intelligent agents, computational

intelligence, and ontologies. Distributed intelligent

agents collect data within the Cloud by monitoring

devices, data streams and code execution. This

information is handled in the form of ontology-based

models. Within the system, these ontologies can be

operated on individually or brought together.

Authors propose a new method that can enrich each

generation of ontology during its creation by using

related, known data to seed the process.

Collazo-Mojica et al. (Xabriel et al., 2012)

propose a SOA API, in which users provide a Cloud

application model and get back possible resource

allocations in an IaaS provider. The solution

emphasizes the assurance of quality of service (QoS)

metrics embedded in the application model. An

initial mapping is done based on heuristics, and then

the application performance is monitored to provide

scaling suggestions.

Cao et al. (Cao et al., 2009) propose a service

oriented QoS-assured Cloud architecture that

includes physical and virtual resources for Cloud

service provisioning. They propose an autonomic

strategy that assures users’ request QoS.

Kim et al. (Kim et al., 2011) propose an

intelligent multi-agent model based on virtualization

rules for Cloud resource management. This model

automatically allocates resources suitable to

optimize Cloud performance. It infers user’s request

by analyzing and learning user context information.

This multi-agent model is composed by three main

agents: User Agent, Gathering Agent, Agent

Manager and Virtualization Register that

synchronizes with distributed agent to manage log

data from the creation of VM to its destruction.

Frincu et al. (Frincu et al., 2011) propose a self-

adaptive distributed scheduling platform composed

of MAS implemented as an intelligent feedback

control loops supporting defined policies and

exposing self-healing capabilities. To face the

challenge of building an autonomous self-healing

system, an adaptive inter-provider MAS was

proposed. The scheduling module is able to offer

fully distributed storage and communication

mechanisms, support fault-tolerance using agents as

recoverable modules, support autonomy by

dynamically changing scheduling policy, and adapt

the negotiation policy.

In (Zarrabi and Zarrabi, 2012), the author

introduces a Cloud intrusion detection system which

is developed based on Cloud computing and can

make up for the deficiency of traditional intrusion

detection. They introduce intrusion detection system

as a Service in a Cloud to protect user network. It

exploits some characteristics in network traffics that

make it possible to extract the required data from the

user network for evaluation. This architecture is

intended to be scalable by allowing users to combine

the features of different intrusion detection systems’

services for more reliable solution.

Although current approaches offer significant

solutions to the problem of Cloud adaptation, their

main drawback is that they don’t clearly describe the

adaptation process of Cloud resources and how

adaptation characteristics of Cloud are addressed

(e.g. security, optimization, context change, etc.). In

addition, existing works are interested in only one

property of AC.

Our work focuses on the autonomic resource

management on the Cloud environment. However,

with an extended scope aimed at jointly considering

resource healing and protection needs and also

power consumption and energy costs.

Another characteristic of the proposed model is

AMulti-AgentbasedArchitectureforCloudInfrastructureAuto-adaptation

the use of distributed MAS that communicate locally

(intra-MAS communication) and in a global level

(intrer-MAS communication).

3 MULTI-AGENT BASED

ARCHITECTURE FOR CLOUD

AUTO-ADAPTATION

Cloud infrastructure can be represented as a tree of

components reflecting the spatial arrangement and

interconnection of Cloud resources. Each level

contains a particular category of resource. The

Cloud central manager is located in the first level.

The second level presents regions distributed over

the world. The third level contains data-centers

which are deployed in regions. The forth level

comprises PM which are located in distributed data-

centers. In each PM, at least one VM is deployed.

In this work, we propose to auto-adapt the Cloud

infrastructure based tree structure. Therefore, we

propose a distributed Cloud agents-based

architecture that uses the tree-based Cloud

infrastructure. Each node in the tree is considered as

a MAS.

Components of different levels have different

functions. The role of a component is ensured

regarding its location by a MAS composed mainly

by four agents: self-configuration agent, self-

optimisation agent, self-protection agent and self-

healing agent.

Each agent in the multi-agent-system integrates

functionality for data management. They accept

information from the next-level agents or previous-

level agent, and continuously learn and updates

according to their own environment. Cloud

adaptation evolves as a result of VMs being created,

migrated, and terminated regarding users’ request. In

addition, Cloud adaptation incorporates the

optimization and the healing of Cloud resources and

ensures Cloud trustworthiness.

Figure 1 shows a multi-agent Cloud

infrastructure architecture which includes physical

resources layer and virtual resources layers to

support autonomic Cloud infrastructure. In

particular, Figure 1 shows an example of an

autonomic Cloud tree with three regions. The region

(in the left) is composed of two data-centers. The

data-center (in the left) contains two physical

machines. The PM (in the left) hosts two VMs

which deploy services used by customers.

In the multi-agent autonomic Cloud architecture,

distributed agents interact via message exchanges

with the environment in the form of

perceptions/effects. Regarding the exchange of

messages between agents, two types of message

exchanges may take place:

- Exchange local messages between agents of a

single MAS based on the means of communication

at their disposal. For example, a self-optimization

agent invokes a self-configuration agent of the

same resource for reconfiguration purpose.

- Exchange of messages between agents located on

remote MAS. For example, to ensure the migration

of a VM, a VM-level agent can contact a DC-level

agent to add a new PM which will host the

migrated VM.

3.1 Self-configuration Agent

The proposed self-configuring module provides

abstraction with complete customization of Cloud

resources. Reporting directly to the hardware

environment, the self-configuring agent at the IaaS

comes down to managing resources. The most

common configuring operations in IaaS are:

resources’ allocation, resources’ release, resources’

migration.

If the current active resources don’t support

users’ request, the autonomic manager should update

infrastructure by adding new resources. It deploys

and startup resources in the IaaS systems. The

starting operation of VM requires prior

configuration.

If the autonomic agent detects that the VM has

terminated services execution, it saves its context

and stops it, then it updates the state of the agent of

the previous layer. The same process is repeated for

the PM and data-center level agents.

The self-configuration agent should ensure VM

migration. The VM is moved from on PM to another

while continuously running, without any remarkable

effects from the point of view of the end users.

3.2 Self-optimisation Agent

The Cloud-based virtualization offers a margin to

the administrator of the IaaS for intervention on PM.

It allows, among others, the Cloud providers to

implement different policies for the allocation or re-

allocation of resources in order to make savings or to

fulfill a contract customer. The IaaS should regularly

scan its environment and reorganizes available VMs

on machines and data-centers to free up some of

them. This release of machines reduces the energy

consumption and cost of IaaS.

CLOSER2013-3rdInternationalConferenceonCloudComputingandServicesScience

Regions

Datacenters

Physical Machines

Autonomic Cloud

Infrastructure

Manager

ion

Datacenter

ion

Figure 1: Multi-agent Architecture for Cloud Infrastructure Adaptation.

The self-optimisation agent collects and analyses

some indicators (reliability, power quality…), then,

it updates Cloud infrastructure by minimizing the

number of active PMs.

3.3 Self-protection Agent

The self-protection agent is built around the IaaS

model for providing security to any Cloud

infrastructure. A self-protection agent is composed

of these modules: detection, analysis and reaction.

Detection module is responsible for intrusion

detection. It collects information form network and

received by the resource. It selects items of interest

and forwards them to the analysis module.

Thereafter, the analysis module should take a

decision and pattern matching plan. The reaction

module is responsible for remote configuration and

control of the resource by applying the defined

protection plan.

3.4 Self-healing Agent

Self-healing agents are able to recover resource from

failures. It is characterized by an intelligent feedback

loop that supports data collection and analysing,

module of recovery plan definition and module for

applying recovery plan.

The recovery module analyses and defines

several healing plans. Then according to collected

information, it selects healing plan. Then the self-

healing agent executes the selected recovery plan.

Infrastructure failures are failures of components

that cannot be handled without interrupt services’

running on it. In the proposed architecture, we

propose, before dealing with recovery plan, to

migrate connected VMs to others PMs.

4 CLOUD INFRASTRUCTURE

ADAPTATION SCENARIOS

This section discusses the autonomic adaptation of

Cloud infrastructure. For this, it is necessary to

adapte VMs that will support the execution of cloud

applications. Once instantiated, each VM is subject

to an autonomic adaptation that ensures the

configuration, optimisation, protection and healing

of the distributed resources.

4.1 Self-configuration Scenarios

The self-configuration agent mainly focuses on the

management of distributed resources in order to

maximize the system utilization.

4.1.1 Resources Allocation Scenarios

Figure 2 describes an example representing adding

new resources. The ACIM deduces the need for two

additional VMs. A distributed agent decides that we

need to create new PM in an existing data-center and

deploys two VMs in order to satisfy users’ request.

Ré

ion

Datacenter

Ré

ion

Datacenter

…

Figure 2: Allocate new VM and PM resources.

AMulti-AgentbasedArchitectureforCloudInfrastructureAuto-adaptation

4.1.2 Resources Release Scenario

In the VM level, self-configuration agents verify the

state of active resource. If the agent indicates that

the resource (VM, PM or DC) state is idle then the

agent should save resource context, update PM agent

context and stop the resource. Figure 3 represents an

example of releasing resource task.

Ré

ion

Datacenter

Ré

ion

Datacenter

Figure 3: Release resource.

4.1.3 Resources Migration Scenario

The self-configuration agent of each VM verifies

continuously the state of the associated VM. If the

VM performance is lower than fixed threshold then

the self-configuration agent creates needed VMs

with appropriate features. Thereafter, agents of

destination VMs and source VMs communicate

information about application being running in the

VM and they migrate VM context. Then, the self-

configuration agent of the VM, PM and DC levels

update their context and they stop the resource

associated to the idle agent (see Figure 4).

Ré

ion

Datacenter

Ré

ion

Datacenter

…

Figure 4: Migration of VM.

4.2 Self-optimization Scenario

The Cloud computing technology makes the

resource as a single point of access to the client and

is implemented as pay per usage. To reduce

infrastructure cost, the proposed autonomic strategy

self-optimizes the Cloud infrastructure by

optimizing the used VM number.

Figure 5 presents an example of Cloud

infrastructure optimization scenario. The

optimization process consists of the following

process. First, distributed MAS find PMs that are

partially used, then, self-configuration agent of

selected PMs migrate the associated resources in

order to free up more resources. Then, self-

configuration agents update their context and they

stop the resource associated to the idle agent.

Ré

ion

Datacenter

Ré

ion

Datacenter

Figure 5: Self-optimization of Cloud resources.

4.3 Self-protection Scenario

At the VM and PM levels, each self-protection agent

should collect continuously security information.

Then, it analyzes them and executes the appropriate

protection plan in order to:

- Ensure the availability of the information held

within or between systems participants.

- Maintain information integrity exchanged within

Cloud nodes.

- Detect and correct intrusions.

For example, if there are failed resources, then

autonomic agents should apply a healing plan in a

transparent way and let the Cloud infrastructure

evolves in a safe manner.

4.4 Self-healing Scenario

The self-healing agent detects failed resources

(Figure 6). The correction process begins by

migrating application to other VMs then the self-

healing agent reconfigures failed resources. Once

resources are repaired they can be activated.

CLOSER2013-3rdInternationalConferenceonCloudComputingandServicesScience

Ré

ion

Datacenter

Ré

ion

Datacenter

…

Figure 8: Self-healing of Cloud resources.

5 CONCLUSIONS

By combining AC and agent technology, this paper

proposes a new multi-agent based architecture for

Cloud infrastructure adaptation which includes

physical resources (PM and data-center), virtual

resource (VM) and an autonomic Cloud

infrastructure manager. The proposed autonomic

Cloud infrastructure strategy instantiates and

coordinates the self-* capabilities for Cloud

infrastructure adaptation.It is based on MAS that are

composed mainly by four agents:

- Self-configuration agent able to allocate, release or

migrate resources.

- Self-optimization agent used to optimize the

amount of active resources in order to reduce

energy consumption.

- Self-protection agent that ensures the credibility

security of Cloud infrastructure.

- Self-healing agent that repairs failed resources in a

transparent way for the users.

Moreover, we have analysed the behaviour of the

proposed model through simulation scenarios.

For future work, we intend to expand our

solution to propose a multi-objective algorithm for

Cloud infrastructure self-optimization that

dynamically modifies the Cloud infrastructure in

order to reduce energy consumption.

REFERENCES

Cao, B. Q., Li, B., Xia, Q. M., 2009.A Service-Oriented

Qos-Assured and Multi-Agent Cloud Computing

Architecture, CloudCom '09 Proceedings of the 1st

International Conference on Cloud Computing, 644 –

649.

Chainbi, W., 2010. An Agent-Based Methodology for

Self-* Systems ", International Journal Multiagent and

Grid Systems Journal, Volume 6, Number 1, pp. 55-

69, IOS Press, 2010.

Frincu, M. E., Villegas, N. M.,Petcu, D., Muller,

H.A.,Rouvoy, R., 2011. Self-Healing Distributed

Scheduling Platform, 11th IEEE/ACM International

Symposium on Cluster, Cloud, and Grid Computing

(CCGrid), 225-234.

Hurwitz, J. Bloor, R. Kaufman, M., Halper, F., 2009.

Cloud Computing For Dummies, no.2.

IBM Group, 2003: An architectural blueprint for

autonomic computing. http://www-

03.ibm.com/autonomic/pdfs/AC.

Kim, M., Lee, H., Yoon, H., Kim, J. I., Kim, H., 2011.

IMAV: An Intelligent Multi-Agent Model Based on

Cloud Computing for Resource Virtualization,

International Conference on Information and

Electronics Engineering, IPCSIT, vol.6, 199-203.

Li, J.,Ma,D., Li L., Zhu,H., 2008. AADSS: Agent-based

Adaptive Dynamic Semantic Web Service

Selection,Int. Conf. on Next Generation Web Services

Practices NweSP 08).

Mazur, S., Blasch, E., Chen, Y., Skormin, V.,

2011.Mitigating Cloud Computing Security Risks

using a Self-Monitoring Defensive Scheme,

Aerospace and Electronics Conference (NAECON),

39-45.

Mell, P., Grance, T. 2011. The NIST Definition of Cloud

Computing (Draft), National Institute of Standards and

Technology, vol.53, no.6, 1-7.

Overeinder, B. J.,Verkaik P. D., Brazier, F. M. T.,2008.

Web service access management for integration with

agent systems, Proc.Of the 23rd Annual ACM

Symp.on Appl. Computing, Mobile Agents and Syst.

Track, Mar. 2008.

Solomon, B., Ionescu, D., Litoiu, M., Iszlai, G. 2010.

Designing autonomic management systems for Cloud

computing, International Joint Conference on

Computational Cybernetics and Technical Informatics

(ICCC-CONTI) , 631–636.

Van, H. N., Tran, F. D., Menaud, J.M. 2009. Autonomic

virtual resource management for service hosting

platforms, Workshop on Software Engineering

Challenges in Cloud Computing, 1-8.

Xabriel J., Collazo-Mojica, S., Sadjadi, M., Ejarque, J.

Rosa, Badia, M., 2012. Cloud Application Resource

Mapping and Scaling Based on Monitoring of QoS

Constraints, Knowledge Systems Institute Graduate

School, 88-93.

Zarrabi, A., Zarrabi, A., 2012. Internet Intrusion Detection

System Service in a Cloud, IJCSI International

Journal of Computer Science Issues, Vol. 9, Issue 5,

No 2, 308-315.

AMulti-AgentbasedArchitectureforCloudInfrastructureAuto-adaptation