Computational Intelligence - May 2015 - 54

Table 1 Descriptions and order of service models and the
related resource management problems.
Service Model

deScripTion

Business
Process-as-aservice (BPaas)

Business Process
outsourcing to
cloud

reSource ManageMenT challengeS

D. Taxonomy of
Cloud Computing
Resource
Management

The field of CC
resource management has its roots in
distributed and paralsoftware-asPerformance and
comPlete
a-service
cost of an
aPPlications,
lel systems resource
(saas)
aPPlication
accessiBle via
management. In this
thin clients
paper, we propose a
Platform-asPerformance of
Platforms to
simple taxonomy of
a-service
develoP and run a Platform given
(Paas)
availaBle resources
aPPlications
CC resource management problems
infrastructure- virtualized
allocation of virtual
as-a-service
infrastructures
resources on
(Fig. 1), based on the
(iaas)
the hardware
more elaborated taxhardware-ason-demand access utilization of the
onomy of Casavant
a-service
to fully
hardware
and Kuhl [5].
(haas)
configuraBle
The main criterihardware
on for the classification of CC resource
management problems is based on the
to the users but requires management
problem's dynamics. In the case of static
by providers. However, in specific
(or offline) problems, the full list of recases, explicit content sharing is still
quests is known a priori, while for dyrequired [2]. Virtualization is commonnamic (or online) problems, resource
ly used to create pools of virtual
requests become apparent (or may
resources on top of physical resources.
change) over time. CC typically offers
Virtualization techniques enable partion-demand services. Therefore CC retion of physical resources and ensure
source management has a tendency to be
isolation of different user spaces. Isolaintrinsically dynamic. Indeed, although
tion is indeed crucial to ensure securifuture users' behaviors and interactions
ty and performance warranties [3]. The
are unknown, the system must remain
multi-tenancy of physical resources
operational. In some less delay-sensitive
results in challenges for each of
use cases, such as batch scheduling or
the above-mentioned aspects of
load balancer training, the optimization
isolation [4].
Quality of service
(Qos) and the
cost of Processes

can be performed offline. A detailed discussion on specific CC resource management problems is presented in
Section III. Most resource management
problems are by nature NP-hard [6], [7].
CI tools seem to be a natural answer to
the resulting complexity, uncertainty,
scale, and dynamics.
E. Review of Applicable
Computational Intelligence Tools

Evolutionary computation is commonly
used to solve combinatorial optimization problems. However, given its computational cost, it is more suited to static
problems. The need for fast, real-time
response to dynamic problems highlights
the efficiency of Artificial Neural Networks (ANNs) and Fuzzy Systems (FSs),
which, after a training phase, can react
quickly. The dynamic CC environment
presents challenges concerning the
incremental and robust training of these
tools. The resulting solutions are often
hybrid systems, in which advanced
behaviors and algorithms are combined
with faster and simpler systems for realtime decision making.
II. Cloud Computing Resource
Management Models
A. Classic Resource
Management Concepts

Processing systems models are commonly
divided into descriptions of the workload
and the system. Scheduling systems are often described using the
three-field notation [7], [8], while
Cloud Computing
Resource
queueing systems can be
Management
expressed in terms of the Kendall's notation [9]. The scheduling
Static
Dynamic
problem consists in allocating
tasks to machines in given time
slots, while queuing theory studLoad
Capacity
Mapping
Scheduling
ies the behavior of the service sysBalancing
Planning
tems that execute arriving tasks.
What to Do
What to Do
Where and
In contrast to these time-deWhere?
in the
Now?
When?
Future?
pendent concepts, in some CC
problems the time dimension
may be absent, e.g. in the case
Server
Cloud
Cloud
VM
Service
Workflow
Load
Capacity
of long-term allocations of VirBrokering
Placement
Placement
Scheduling
Balancing
Planning
tual Machines (VMs) or services
[10], or in the case of applications that are tolerant to delays.
Figure 1 a simple taxonomy of cc resource management problems.

IEEE ComputatIonal IntEllIgEnCE magazInE | may 2015

Table of Contents for the Digital Edition of Computational Intelligence - May 2015