Page 83 - Proceedings of the 2018 ITU Kaleidoscope

MACHINE LEARNING OPPORTUNITIES IN CLOUD COMPUTING DATA CENTER
MANAGEMENT FOR 5G SERVICES

1 2
Fabio López-Pires and Benjamín Barán
1 Itaipu Technological Park, Hernandarias, Paraguay
2 National University of the East, Ciudad del Este, Paraguay

ABSTRACT Additionally, network management challenges based on
software-defined networking (SDN) are also analyzed from
Emerging paradigms associated with cloud computing the perspective of VMP problems, where ML techniques
operations are considered to serve as a basis for integrating may result in a promising approach to support these types of
5G components and protocols. In the context of operational decisions. Finally, different open challenges are
resource management for cloud computing data centers, discussed as future directions to further advance this active
several research challenges could be addressed through research field.
state-of-the-art machine learning techniques. This paper The remainder of this work is structured as follows: Section
presents identified opportunities on improving critical 2 briefly presents the considered two-phase optimization
resource management decisions, analyzing the potential of scheme for VMP problems, while Section 3 discusses
applying machine learning to solve these relevant problems, the main opportunities for ML techniques as a promising
mainly in two-phase optimization schemes for virtual machine approach to address identified research challenges. Finally,
placement (VMP). Potencial directions for future research are conclusions and future directions are left to Section 4.
also presented.
Keywords - 5G service operations, cloud data centers, 2. TWO-PHASE OPTIMIZATION SCHEME FOR
machine learning, virtual machine placement. VMP PROBLEMS IN CLOUD COMPUTING

Recent research advances in VMP problems for cloud
1. INTRODUCTION
computing include proposals of complex infrastructure as a
According to Rost et al. [17], 5G networks and services will service (IaaS) environments for VMP problems, considering
increase exponentially in data traffic, storage and processing, both service elasticity and the overbooking of physical
considering smartphones as gateways to remotely access resources [14]. In the context of 5G services, and considering
resources through cloud computing. In this case, several smartphones as simple gateways to access remote resources
challenges should be addressed to further advance cloud as previously mentioned [17], 5G service providers should
computing in order to serve as a basis to integrate 5G associate a cloud service infrastructure with each mobile
components and protocols. customer. A cloud service infrastructure S b may be composed
In the context of resource management for cloud computing of a set of VMs according to customer preferences or
data centers, main research challenges could be addressed by requirements, where both elasticity and overbooking should
designing management solutions based on machine learning be considered, as previously proposed by the authors in
(ML) techniques. [14]. In the described context, VMP problems represent an
This work briefly discusses recent contributions on one important topic for cloud computing data center management.
of the most studied problems for resource allocation in The following sub-sections describe the highlights of a
cloud computing data centers: the process of selecting two-phase optimization scheme for VMP problems in cloud
which requested virtual machines (VMs) should be hosted computing, representing the main focus of the challenges
at each available physical machine (PM) of a cloud analyzed in this work.
computing infrastructure, commonly known as virtual
machine placement (VMP). The considered contributions 2.1 Considered VMP Formulation
focus on a two-phase optimization scheme for VMP problems
[1] (see Figure 1), which takes into account incremental An online problem formulation is considered when inputs of
VMP (iVMP) and VMP reconfiguration (VMPr) as the main the problem change over time and algorithms do not have
sub-problems with online and offline phases respectively. the entire input set available from the beginning (e.g. online
The identified challenges for considered VMP problems heuristics) [3]. On the other hand, if inputs of the problem
[11] are presented on the particular context of 5G services, do not change over time, the formulation is considered offline
and mainly take into account ML techniques for addressing (e.g. memetic algorithms (MAs) proposed in [8] and [12]).
relevant decision making on cloud computing infrastructure Online decisions made along the operation of a dynamic
operations (e.g. when a VMPr phase should be triggered?). cloud computing infrastructure negatively affects the quality

978-92-61-26921-0/CFP1868P-ART @ 2018 ITU – 67 – Kaleidoscope