An Adaptive Control Strategy for Resource
Allocation of Service-based Systems in Cloud
Environment
Siqian Gong, Beibei Yin, Wenlong Zhu, Kaiyuan Cai
School of Automation Science and Electrical Engineering
Beijing University of Aeronautics and Astronautics
Beijing, China
{gongsiqian, yinbeibei, chinabuaazwl, kycai}@buaa.edu.cn
Abstract—Service-based systems (SBS) resource allocation in
cloud computing environment is important to guarantee Quality
of Service (QoS) to satisfy requirements of service requests as
well as reduce operating costs. In this paper, a control strategy is
presented to allocate the system resources of cloud provider to
the servers using PID self-tuning adaptive control dynamically.
Meanwhile, an adaptive mechanism using Radial Basis Function
(RBF) neural network is devised to adjust parameters of PID
controller adaptively in run time. Experimental results show that
the approach enables to satisfy the requirements of services-
requests with less adjustment of resource allocation, and has
better adaptive ability and stability compared to conventional
PID control.
Keywords—resource allocation; PID self-tuning adaptive
control; RBF neural network; quality of service; cloud
environment
I. INTRODUCTION
Service providers can accurately scale their service in pay-
per-use business mode when cloud computing providing on-
demand access to computational resources [1]. For providing
more reasonable services, it is important to use of resources
intelligently. Consequently, SBS has been adopted in cloud
computing environment and Internet has been considered as a
supercomputer. As a result, the vision of “software as a
service”, “platform as a service”, “infrastructure as a service”
arise, which require these services are in a changing and
commercial cloud environment [2].
However, though there are advantage of using SBS in cloud
computing environment to fully utilize services, allocating the
resources and managing QoS through cloud computing are
still problems when analyzing the costs [3]. For example, users
can publish service requests to an intensive agent from
anywhere, then the cloud provider provide a platform for huge
enterprises like hospitals to manage their metadata and medical
data as services with high QoS in cloud computing [4]. When a
peak workload of service requests occurres, the enterprises
charge by a large amount of resource to services to meet the
requests, and when this workload disappeared, they do not
want to pay for the idle resources [4]. As a result, it is
challenging for services, whose QoS are highly affected by
changing resources [5, 6].
Furthermore, SBS do not possess the mode of “cost-
effective”, “pay-per-use” and take much attention on less time,
simply selecting the best service to user is not a proper option
[7]. In this case, the cloud provider and its services should be
aware of how service-requests use its resource [8]. Due to the
increasing commercial value of cloud services, there exists the
need to manage the QoS in run time, while an adaptive
approach to charge the resource usage is needed.
As the services utilize resource to process service-requests,
the more allocated resource achieves high QoS services, but
leads to high costs of users. To solve the conflicting situation,
providing a proper service, whose QoS is close to the service-
request required is necessary for QoS management and
resource allocation. It means that the less error of QoS between
practical and required is better. Hence, one problem to be
solved is reducing the error less and less related to resource
allocation. In addition, different service-requests requires
various QoS, thus the QoS requirement of service-requests
fluctuate dynamically. Accordingly, accurate and quick
adjustment of resource allocation is needed to cope with the
fluctuation of requirements.
Since the SBS in cloud computing environment is a
dynamic, nonlinear and time-varying system, guarantee the
stability of system is necessary to be considered. Self-tuning
adaptive control architecture is adopted in this paper to
maintain the stability of system and to improve the accuracy of
QoS. Furthermore, it requires highly adaptive ability as the
cloud computing is a real time environment. In order to
enhance the adaptive ability, a system identification as adaptive
mechanism is devised to capture the nonlinear relationship
between QoS and resource allocation. Due to the dynamical
system, it is difficult to build precise model to describe the
control system. Thus neural network is adopted as the system
identification to cope with the complex, nonlinear and time-
varying environment [9, 10]. Therefore, the control system
2015 IEEE International Conference on Software Quality, Reliability and Security Companion
/15 $31.00 © 2015 IEEE
DOI 10.1109/QRS-C.2015.17
32
2015 IEEE International Conference on Software Quality, Reliability and Security Companion
/15 $31.00 © 2015 IEEE
DOI 10.1109/QRS-C.2015.17
32
2015 IEEE International Conference on Software Quality, Reliability and Security Companion
/15 $31.00 © 2015 IEEE
DOI 10.1109/QRS-C.2015.17
32
2015 IEEE International Conference on Software Quality, Reliability and Security - Companion
978-1-4673-9598-4/15 $31.00 © 2015 IEEE
DOI 10.1109/QRS-C.2015.17
32