A CBR Approach to Allocate
Computational Resources Within a Cloud
Platform
Fernando De la Prieta, Javier Bajo and Juan M. Corchado
Abstract Cloud Computing paradigm continues growing very quickly. The
underlying computational infrastructure has to cope with this increase on the
demand and the high number of end-users. To do so, platforms usually use
mathematical models to allocate the computational resource among the offered
services to the end-user. Although these mathematical models are valid and they are
widely extended, they can be improved by means of use intelligent techniques.
Thus, this study proposes an innovative approach based on an agent-based system
that integrated a case-based reasoning system. This system is able to dynamically
allocate resources over a Cloud Computing platform.
1 Introduction
The technology industry and the scientific community have taken great strides in
recent years toward implementing the Cloud Computing (CC) technological para-
digm. This has resulted in a rapid growth of both private and public platforms [12,
17, 25, 28] aimed to provide innovative solutions that can resolve the current needs
of the CC paradigm.
The marketing model used in the CC paradigm is innovative, as it is based on a
pay-as-you-go concept [2], in which users must negotiate and previously establish a
Service Level Agreement (SLA) in order to access services [1]. Once this contract
F. De la Prieta (✉) ⋅ J.M. Corchado
Department of Computer Science and Automation Control,
University of Salamanca, Plaza de la Merced s/n, 37008 Salamanca, Spain
e-mail: fer@usal.es
J. Bajo
Department of Artificial Intelligence, Technical, University of Madrid,
Bloque 2, Despacho 2101, Campus Montegancedo,
Boadilla del Monte, Madrid 28660, Spain
e-mail: jbajo@fi.upm.es
© Springer International Publishing Switzerland 2016
P. Novais et al. (eds.), Intelligent Distributed Computing IX,
Studies in Computational Intelligence 616
DOI 10.1007/978-3-319-25017-5_7
75
for computing goods has been established, both the users (through regular payments)
and the CC system (by maintaining the service) are obligated to follow through with
their agreement. In this regard, novelty is determined by the innovative spectrum of
underlying technology (virtualization, service farms, web services, etc.), which have
recently reached the point of allowing the services to be offered with the same level
of quality, regardless of existing user demand [16, 26, 31]. These new possibilities at
a technological level lead to the birth of a new concept, elasticity [9], which is based
on the just-in-time production method [13].
Existing research in the state of the art is based on methods that use centralized
algorithms based on mathematical and heuristic models [15, 19, 30], neither of
which can ensure the efficiency of the system, or even its availability, in the event of
a system failure.
Given these shortcomings, it is necessary to study new techniques that allow for
the evolution of existing models with regard to elasticity of services. This study
proposes the use of models derived from Artificial Intelligence (AI), since fron an
internal point of view, a CC is characterized by its massive distribution, hetero-
geneity, and high level of uncertainty, which is precisely where the application AI
holds great potential. The inclusion of proactive, self-adaptation and learning
capabilities, among others, is key for the evolution of these elastic management
algorithms for computational resources. As a result, agents and multiagent systems
[29] (MAS) were selected among all the available AI techniques because of their
distributed nature and ability to work in environments such as CC systems, whose
characteristics would clearly identify them as open systems.
Using this MAS-based approach, the framework of this study proposes a
dynamic and self-adapting model for the distribution of computational resources in
a CC environment. This model is based on the learning capabilities provided by a
case-based reasoning (CBR) [10] system, an approach which has not previously
been used in this type of distributed environment. These reasoning systems develop
a reasoning model similar to that of humans, using past experiences to solve a
specific problem.
This work is organized as follows: the following section provides a description
of the context of and related approaches, Sect. 3 proposes a solution based on
multiagent systems, while the evaluation and validation of these systems are pre-
sented in Sect. 4. Finally, the last section presents the conclusions of the research.
2 Resource Allocation in Cloud Computing Platforms
In a CC environment, the hardware infrastructure is virtualized [7, 8], which means
that there is an abstraction layer between the real hardware infrastructure and the
computing nodes. Each of the services is actually deployed in the computing nodes
of this abstraction layer (referred to as virtual machines). In turn, the services are
generally distributed among various computational nodes, which is why their needs
76 F. De la Prieta et al.
to be a work balance system that can distribute the requests among the various
computational nodes attending the services.
The use of virtualization greatly simplifies the management of computational
resources at the infrastructure level, making is possible to dynamically create or
eliminate virtual machines on demand or even migrate a virtual machine from one
physical server to another in execution time, without needing to stop or pause the
machine. Therefore, and given the capabilities offered by virtualization technology,
the problem, while complex, is actually simple in itself, since it is only based on the
efficient redistribution of physical (real) resources among the different computa-
tional (virtual) nodes.
In current literature, the distribution of resources is viewed from two points of
view [5]:
• QoS-aware based, or market oriented [5]. This first group is associated with a
client-oriented distribution of resources model which attempts to minimize
computational risks in order to distribute the computational resources according
to the SLA reached, and following the pay-per-use economic model. According
to this model, the management techniques for the computational resources aim
to adhere to these agreements at all time, thus providing the quality of service
that was requested and consequently expected by the end user. The state of the
art includes studies in line with this approach by means of mathematical models
[18, 23, 27].
• Energy-aware based [5]. In this second approach, the distribution of resources
takes place by taking into account both the pre-established SLA and the energy
consumption, which assumes compliance with both. There are fewer studies in
the state of the art with this approach as compared to the first, although they are
more novel. This includes a variety of techniques are also based on mathe-
matical models [3, 15, 19].
In light of these studies in the current state of the art, it is necessary to propose a
model for the distribution of computational resources which would take energy
consumption into account. The aim is to reduce the energy consumption required to
satisfy the SLA that have been established with the platform users. The present
study follows a completely different approach based on optimization techniques and
AI, which allows for the distribution of resources by following a distributed and
scalable model, thus allowing the system to learn over an extended period of time.
As noted above, the CC computational paradigm has grown strongly in recent
years; its development has led to the advancement of a large number of platforms,
both public and private. A MAS framework based on VO has been selected to deal
with these obstacles. Although one may initially consider these two distributed
systems (MAS and CC) to be incompatible, a detailed analysis demonstrates that
they are in fact not only complementary, but share considerable synergy between
them. First of all, CC environments can cover the computational needs for per-
sistence of information and the computing potential that MAS require for different
applications such as data mining, management of complex services, etc. Addi-
tionally, MAS can be used to create a much more efficient, scalable and adaptable
A CBR Approach to Allocate Computational Resources … 77
design for the CC environment than what is currently available. Finally, the use of
MAS in the framework of the design for CC systems provides this paradigm with
new characteristics such as learning or intelligence, which makes it possible to
develop much more advanced computational environments in all aspects (intelligent
services, interoperability among platforms, efficient distribution of resources, etc.).
The number of studies that can be found on the state of the art relating CC with
agent technology is actually quite low. However, this tendency is changing and it is
becoming increasingly common to find studies and applications focused on this
field. Despite the limited number of studies on the matter, Agent-based Cloud
computing, or the Agent-based Cloud platform, is becoming a common concept,
mentioned by various authors in recent years [4, 6, 14, 20–22, 24].
3 Proposed Architecture
Taking into account the needs and shortcomings detected in the review of the state
of the art, this study proposes a new model of allocating resources based on a CBR
approach and guided by a multiagent architecture especially designed for the
management of CC environments. This section will describe the key components
that allow extending the operation of the elastic algorithms for the distribution of
resources proposed within this work.
To begin, we would like to note that since the proposed MAS is a distributed
system by nature, each of the agents that work in the distribution of resources can
be located throughout the entire CC environment. That is, the CC system is
monitored and controlled in a distributed manner. This distributed monitoring
model makes it possible to instantly adapt existing resources to the CC environment
according to demand for each service, which in turn meets the dual objective of
complying with the established SLA agreements and reducing energy consumption.
Figure 1 presents an overview of the agent-based architecture The following
agents are directly related to the monitoring and control of the hardware:
• Local Monitor. In charge of gathering data related to the state of the local
resources for each physical server, including the physical machine as well as the
different virtual machine it hosts.
• Local Manager. In charge of controlling the computational resources of the
physical machine. In other words, responsible for initiating or turning off virtual
machines according to the previously configured service templates.
• Global Manager. The primary agent in charge of decision making with regard
to the distribution of computational resources. In order to perform this task, the
agent uses a CBR-BDI model, which will be explained in detail in the following
section. As a means of support for making decisions regarding the distribution
of resources, this agent uses a partial knowledge base (provided by the Local
Monitor) and the ability to modify local resources in each machine (provided by
the Local Manager).
78 F. De la Prieta et al.
The redistribution of resources at a macro level is performed by the Global
Manager agents, which have greater authority than the Local Manager agent and
can inform them of the need to start up a new machine with a specific service and
specific characteristics. At the end of this process, a new virtual machine (VM), with
specific characteristics of Memory and virtual cpus will be instantiated in order to
meet the current demand.
The Global Manager is a highly specialized agent that implements a CBR-BDI
[10, 11] deliberative architecture. As a result, the reasoning process in each physical
node is based on past experience gained from storing similar cases. The case
memory is central to the entire CC system; the system’s global knowledge can be
shared by each of its members, in this case the Global Manager agent. Given that
this memory can grow exponentially as a maintenance strategy, a high-speed
schema-less database is used to provide fast access to the stored data, based on
MongoDB.1
The Global Manager initiates the process by defining the concept of case
C = {P, S(P), E} where:
• P corresponds to the problem description, which has a matrix-matched repre-
sentation associated to the instantiation of the use of resources.
• S(P) is associated with the solution to the problem: S(P) = {M, vcpu} in terms of
memory and vcpu.
• The efficiency (E) is measured from two perspectives:
• the degree of efficiency of the proposed solution within the physical server
where the virtual machine has been instantiated. This degree of efficiency is
Global
Supervisor
Physical Resource M
. . .
Physical Resource 1
Local
Manager
Global
Regulator
Global
Regulator
Local 
Resource
Monitor
VM 1 VM i
Service Demand
Monitor
Service
Service
Supervisor
. . .
VM j
. . .
Network
Service Demand
Monitor
Service
Service
Supervisor
. . .
VM 2 VM 1 VM 2
User
Consume
Organization
Resource
organization
Management
Organization
Hardware
Manager
Hardware
Manager
Fig. 1 Agent-based architecture for a Cloud Computing Platform
1http://www.mongodb.org.
A CBR Approach to Allocate Computational Resources … 79
proposed by the Local Monitor agent according to the usage rates of the
processor and the allocated memory.
• The degree of efficiency from the point of view of the service. The degree of
efficiency measures the number of additional nodes required by the service.
The CBR (Case-Based Reasoning) process is initiated and retrieves similar cases
from the case memory. The most similar cases are selected according to the fol-
lowing steps: (i) Select the cases from the physical machines with similar charac-
teristics; (ii) a vector is configured for each case that contains the same number of
virtual machines that are in the case; and (iii) the cases selected from this subset are
those that previously used the same service that is now requesting resources, and
during a period of time similar to the current case.
A solution to the problem, which is based on the retrieved cases, will be prepared
during the reuse phase:
• If the case base does not contain a previous similar case, the solution to the
problem will be associated to the minimum resources determined at the level
where the service is instantiated.
• If, on the other hand, similar cases are retrieved, the solution to the problem will
be the closest case multiplied by the case efficiency:
• If the values assigned to the previous solution are greater than the values
assumed by the machine, due to the fact that there are not many resources
available, the result of the case will be the maximum amount of resources
available in the machine.
Once the solution to the case has been calculated, the new node will be
instantiated and its use evaluated from a micro and macro perspective, thus pro-
viding the value of efficiency for the solution. Finally, during the final state of the
proposed CBR cycle, the case and its corresponding efficiency will be stored for
future use.
4 Evaluation
The evaluation and validation of the model for this study will be done through a CC
platform developed within the scope of the research carried out by the BISITE
research group,2 and will include different computational services at the hardware
and software level. This CC platform was deployed in the HPC environment of the
BISITE research group and composed of 15 latest generation machines that support
virtualization in the hardware with the use of Intel-VT technology and the KVM
virtualization system.
During the experiment, 10 threads that query to specific methods of the service
(GetSize and GetFolderContent) are launched every three seconds, to a maximum
2http://bisite.usal.es.
80 F. De la Prieta et al.
of 40 threads. The process starts once the agent-based architecture detects a
decrease in performance, at which time it directly executes the adaptation process.
The Global Manager agent for each of the physical machines that host the service
nodes. We should recall that the Global Manager agent is a specialized agent that
uses a CBR-BDI reasoning process [10] in charge of the distribution of resources at
a macro level. Once they receive the initial alert, these agents resend the alert
message to the remaining Global Manager agents in the CC system.
Each individual Global Manager hosted by each physical machine carry out in
parallel the process described in the previous section. Thus n solutions are pro-
posed. The agent-based architecture reactively selects the node that offers the most
resources at the virtual machine level.
The results in terms of QoS can be seen in Fig. 2, which also show an increase in
the quality level after the adaptation has been completed.
The case study was repeated numerous times, which made it possible to store a
good number of past experiences in the case memory. However, as presented in
Fig. 3 when there are many cases in the memory and a number of past experiences
similar to the current problem, the adaptation results are actually better because the
QoS level is lower.
0,7
0,6
0,5
0,4
0,3
0,2
0,1Re
sp
on
se
 ti
m
e 
(se
c.)
Elapsed time (sec.) Elapsed time (sec.)
0
0,7
0,6
0,5
0,4
0,3
0,2
0,1R
es
po
ns
e 
tim
e 
(se
c.)
0
Fig. 2 Experiment 1: readjustment of the infrastructure resources for method (left no adaptation;
right adaptation)
Elapsed time (sec.) Elapsed time (sec.)
0,7
0,6
0,5
0,4
0,3
0,2
0,1R
es
po
ns
e 
tim
e 
(se
c.)
0
0,9
0,8
0,7
0,6
0,5
0,4
0,3
0,2
0,1
0
R
es
po
ns
e 
tim
e 
(se
c.)
Fig. 3 Experiment 1: readjustment of the infrastructure resources for method (left experiment 1;
right informed adaptation)
A CBR Approach to Allocate Computational Resources … 81
5 Conclusions
This study initially set forth to be one of the first MAS approaches to fall within the
framework of control and monitoring systems in a CC environment. The study
proposed a new architectural model based on a MAS with a clearly integrative
character. A series of algorithms for the distribution of computational resources in a
CC environment were developed, evaluated and validated. Its biggest innovation
centers on the system’s dynamic ability to automatically adapt according to demand
and learn from previous experiences.
This new model has demonstrated that a control and monitoring system in a CC
environment can be designed with MAS. The inherently distributed nature of MAS
makes it possible to implement elastic algorithms for services by following a dis-
tributed strategy. The distribution of responsibilities within the scope of this type of
algorithm makes it possible not only to make decisions where the problems actually
arise, but to distribute the computing capability required to reach a solution among
different instances of the CC environment.
This approach also ensures independence of the decision-making process in
software layers where the various actions are executed. There is no doubt that a
change in the capabilities offered by the underlying technology will also require
changes to be made in the proposed reasoning models, as with any approach with a
traditional design. Given the definitions of roles at a high level, if the technology
proposes new capabilities, the adaptation in the proposed architecture will consist of
modifying the individual or individuals that perform specific tasks or have a role
within the MAS.
Finally, This approach can maximize the degree of efficiency of the proposed
solutions with regard to previous solutions, which in turn progressively improves
the response and the system’s capability since it is capable of learning. Moreover,
this learning ability is important in an uncertain environment such as the CC
system. If the context or environment of the CC platform changes at any given time,
the adaptation model will evolve in turn, adapting the proposed solutions in order to
maximize the efficiency of the given solution.
Acknowledgments This work has been supported by the MICINN project
TIN2012-36586-C03-03.
References
1. Alhamad, M.; Dillon, T.S., Chang, E.: Conceptual SLA framework for cloud computing
(2010)
2. Armbrust, M., et al.: A view of cloud computing. Commun. ACM 53(4), 50–58 (2010)
3. Beloglazov, A. Abawajy, J. Buyya, R.: Energy-aware resource allocation heuristics for
efficient management of data centers for cloud computing. Future Gener. Comput. Syst. 28(5),
755–768 (2012)
82 F. De la Prieta et al.
4. Braubach, L., Jander, K. Pokahr, A.: A middleware for managing non-functional requirements
in cloud PaaS. In: IEEE International Conference on Cloud and Autonomic Computing
(ICCAC), pp. 83–92 (2014)
5. Buyya, R., Beloglazov, A., Abawajy, J.: Energy-efficient management of data center resources
for cloud computing: a vision, architectural elements, and open challenges. Preprint arXiv:
1006.0308 (2010)
6. Cao, B.-Q., Li, B. Xia, Q.-M.: A service-oriented QoS-assured and multi-agent cloud
computing architecture. In: Cloud Computing, pp. 644–649. Springer, Berlin (2009)
7. Che, J., et al.: A synthetical performance evaluation of OpenVZ, Xen and KVM. In: IEEE
2010 Asia-Pacific Services Computing Conference, pp. 587–594. IEEE (2010)
8. Chen, W., et al.: A novel hardware assisted full virtualization technique. In: The 9th
International Conference for Young Computer Scientists, pp. 1292–1297. IEEE (2008)
9. Chiu, D.: Elasticity in the cloud. Crossroads 16(3), 3–4 (2010)
10. Corchado, J.M., et al.: Replanning mechanism for deliberative agents in dynamic changing
environments. Comput. Intell. 24(2), 77–107 (2008)
11. Corchado, J.M., Laza, R.: Constructing deliberative agents with case‐based reasoning
technology. Int. J. Intell. Syst. 18(12), 1227–1241 (2003)
12. Fisher, P., Pant, R., Edberg, J.: Cloud Computing: Assessing Azure, Amazon EC2, Google
App Engine and Hadoop for it Decision Making and Developer Career Growth. Apress, New
York (2010)
13. Hutchins, D.:Just in time. Gower Publishing Ltd., London (1999)
14. Kang, J. Sim, K.M.: Cloudle: an ontology-enhanced cloud service search engine. In: Web
Information Systems Engineering–WISE 2010 Workshops. Springer, Berlin, Heidelberg,
pp. 416–427 (2011)
15. Kusic, D., et al.: Power and performance management of virtualized computing environments
via lookahead control. Cluster Comput. 12(1), 1–15 (2009)
16. Liu, F., et al.: NIST Cloud Computing Reference Architecture, vol. 500, p. 292. NIST Special
Publication (2011)
17. Luo, J.-Z., et al.: Cloud computing: architecture and key technologies. J. China Inst. Commun.
32(7), 3–21 (2011)
18. van Nguyen H., Dang Tran, F. Menaud, J.-M.: Autonomic virtual resource management for
service hosting platforms. In: Proceedings of the 2009 ICSE Workshop on Software
Engineering Challenges of Cloud Computing, pp. 1–8. IEEE Computer Society (2009)
19. Raghavendra, R., et al.: No power struggles: coordinated multi-level power management for
the data center. In: ACM SIGARCH Computer Architecture News, pp. 48–59. ACM, (2008)
20. Sim, K.M.: Agent-based cloud computing. IEEE Trans. Serv. Comput. 5(4), 564–577 (2012)
21. Talia, D.: Cloud computing and software agents: towards cloud intelligent services. In: WOA
2011, pp. 2–6
22. Talia, D.: Clouds meet agents: toward intelligent cloud services. IEEE Internet Comput. 16(2),
78–81 (2012)
23. Van Hien N., Tran, F.D., Menaud, J.-M.: SLA-aware virtual resource management for cloud
infrastructures. In: Ninth IEEE International Conference on Computer and Information
Technology, CIT’09, pp. 357–362. IEEE (2009)
24. Venticinque, S., et al.: A cloud agency for SLA negotiation and management. In:Euro-Par
2010 Parallel Processing Workshops, pp. 587–594. Springer, Berlin, Heidelberg (2011)
25. Von Laszewski, G., et al.: Comparison of multiple cloud frameworks. In: 2012 IEEE 5th
International Conference on Cloud Computing (CLOUD), pp. 734–741. IEEE (2012)
26. Wang, L., et al.: Cloud computing: a perspective study. New Gener. Comput. 28(2), 137–146
(2010)
27. Wei, G., et al.: A game-theoretic method of fair resource allocation for cloud computing
services. J. Supercomput. 54(2), 252–269 (2010)
28. Wen, X., et al.: Comparison of open-source cloud management platforms: OpenStack and
OpenNebula. In: 2012 9th International Conference on Fuzzy Systems and Knowledge
Discovery (FSKD), pp. 2457–2461. IEEE (2012)
A CBR Approach to Allocate Computational Resources … 83
29. Wooldridge, M., Jennings., N.R.: Intelligent agents: theory and practice. Knowl. Eng. Rev. 10,
(02), 115–152 (1995)
30. You, X., et al.: RAS-M: resource allocation strategy based on market mechanism in cloud
computing. In: Fourth ChinaGrid Annual Conference, ChinaGrid’09, pp. 256–263. IEEE
(2009)
31. Zhang, Q., Cheng, L., Boutaba, R.: Cloud computing: state-of-the-art and research challenges.
J. Internet Serv. Appl. 1(1), 7–18 (2010)
84 F. De la Prieta et al.