Content uploaded by Carlos Becker Westphall
Author content
All content in this area was uploaded by Carlos Becker Westphall on Apr 12, 2016
Content may be subject to copyright.
Order@Cloud: A VM Organisation Framework
Based on Multi-Objectives Placement Ranking
Guilherme Arthur Geronimo
Federal University of Santa Catarina
Florian
´
opolis, Brazil
guilherme.geronimo@ufsc.br
Rafael Brundo Uriarte
IMT Institute For Advanced Studies
Lucca, Italy
rafael.uriarte@imtlucca.it
Carlos Becker Westphall
Federal University of Santa Catarina
Florian
´
opolis, Brazil
carlos.westphall@ufsc.br
Abstract—This paper proposes a flexible and extensible frame-
work, named Order@Cloud, to improve the Virtual Machine
Placement of a Cloud. It organises the Cloud VMs by relocating
them based on Multiple-Objectives of the environment. These Ob-
jectives are represented by Rules, Qualifiers, and Costs, which can
be easily added, extended and prioritised. Based on Evolutionary
Searches, Order@Cloud theoretically guarantees the adoption of
a better set of Placements. More specifically, it seeks the non-
dominated solutions (Pareto Set) and compare then considering
the implementation cost of the scenario and its benefits. In
contrast to existing solutions that address specific objectives,
our framework was devised to support any type of priority
and to be easily extensible, which enables the implementation
of new and generic prioritised elements. Moreover, we conducted
experiments using data from a real Cloud environments and show
the flexibility of our approach and its scalability.
Keywords—Virtual Machine Placement; Cloud Computing;
Multi-Objective Optimisation.
I. INTRODUCTION
Although Cloud Computing (CC) brings many benefits,
Cloud’s mismanagement usually accentuates problems related
to waste of resources. For example, VMs performance degrada-
tion due noisy neighbours [1], rise of thermal hotspots on data
center [1] and shortage of resources due constant migrations
[2]. Moreover, according to [3] approximately 30% of PMs
are actually idle, having an average usage ratio of 10%.
Many approaches were proposed to mitigate this problem,
such as Simulation-Based, Policy-Based, Bin Packing and Evo-
lutionary Algorithms (e.g. Ant Colony, Genetic Algorithms).
However, these approaches usually focus in specific objectives,
e.g. on decreasing the energy consumption, the number of Ser-
vice Level Agreement (SLA) violations and resource wastage.
To address this limitation, Xu et al. [4] propose a wider
definition for VMP (which is also used by [5], [6], [7]) as a
multiple objective (MO) optimisation problem that (simulta-
neously) tries to minimise the total resource wastage, power
consumption and thermal dissipation costs. This view extends
the previous approaches by modelling the VMP problem as
a MO optimisation issue, which also enables the managers to
consider other facets of VMP, for instance, internal policies
and Service Level Objectives (SLO).
Nevertheless, these many facets, which MOs cover, need to
be represent and implement in a standard manner. Since each
facet has its own nature, which could be quantifiable and/or
qualifiable, to the best of our knowledge, no suitable model
meet our needs to represent different types of evaluations at a
standard manner.
Moreover, to solve MOs conflicts and to guarantee fast
convergence to a good solution, proposals usually adopt Evo-
lutionary strategies with specific heuristics to build a set of
non-dominated scenarios (using Pareto). However, in these
solutions, the selection of the best scenario is another issue due
limited evaluation strategies and to considerable execution time
and cost. Moreover, none of them take into consideration, at
the same time, important characteristics of Clouds that impact
in this task, such as SLA, policies and scalability.
Considering these limitations, we propose an easily exten-
sible framework, named Order@Cloud, to address the VMP or-
ganisation problem. This framework adopts a flexible approach
that enables the assessment and comparison of from single
placements to the whole clusters, enabling evaluations with
grater precision. Moreover, it uses MO qualification functions
to select VMs which should be relocate, generates the possible
placement scenarios, filters the non-dominated results (Pareto
set) and selects the best result from this considering the costs
to reach the scenario and its benefits.
The Order@Cloud framework is easily extensible, takes
advantage of constraints and SLAs to reduce the computation
cost, enabling a fast local-optimal search. The main contribu-
tions of this framework are: (i) the support to generic multiple
objectives, (ii) objective prioritisation, (iii) generic environ-
ment rules and (iv) ability to compare different scenarios.
The rest of this paper is organized as follows: Section
II provides a background for the concepts and brings the
related works; Section III presents a Cloud model and details
the problem; Section IV presents the framework concepts,
goals and architecture; Section V provides implementation
guidelines to the Framework for Rules, Qualifiers and Cost
functions; Section VI contains the description and results
from the preliminary tests; Section VII concludes the article
addressing the future works and open issues.
II. BAC KG ROUND A ND RELATED WORKS
In the first part of this section we present some concepts
which help to understand our solution. Then, we analyse the
related works.