17-06-2014, 02:59 PM
On the Optimal Allocation of Virtual
Resources in Cloud Computing Networks
On the Optimal Allocation.pdf (Size: 1.43 MB / Downloads: 22)
INTRODUCTION
CLOUD computing promises reliable services delivered
through next generation data centers that are built on
compute and storage virtualization technologies. According
to Buyya et al., [1] “a cloud is a type of parallel and distributed
system consisting of a collection of interconnected and virtualized
computers that are dynamically provisioned and presented as one
or more unified computing resources based on service-level
agreements established through negotiation between the service
provider and the consumers” and accessible as a composable
service via web 2.0 technologies.
Therefore, with respect to cloud computing there exist
the “as a service” definitions, which include software as a
service (SaaS), infrastructure as a service (IaaS), and platform
as a service (PaaS). Each of these has a very different
business value proposition. However, despite of the model
adopted and followed, ultimately the goal of cloud computing
is to create a fluid pool of virtual resources across
computers, servers, and data centers that enable users to
access stored data and applications on an as-needed basis.
In distributed computing environments, up to 85 percent
of computing capacity remains idle [2]. Cloud IaaS emerged
as a solution providing immediate and on demand access to
computing resources with significant cost savings for the
users. The same need for cost efficiency however applies
also to the cloud IaaS providers. Cloud providers try to take
advantage of the cloud’s elastic service provisioning model,
by utilizing solely the needed capacity to satisfy targeted
end-user quality of service (QoS) at any given time. In such
an environment, the probability of all embedded requests in
the cloud utilizing their maximal capacity requirements
simultaneously is low [3]. Therefore, the capacity of
physical resources can be multiplexed among requested
resources allowing us to accommodate more requests [4].
The term thin provisioning is used to differentiate from the
established overprovisioning methodology that plans capacity
for peak workloads [3].
Moreover, in future Internet (FI) vision, where Internet
connection of objects and federation of infrastructures
become of high importance, the cloud involves two
different key players: cloud computing and networking.
For many cloud computing applications, network performance
will be the key to cloud computing performance and
its subsequent adoption. QoS delivery in the cloud is
intrinsically integrated with the network, its infrastructure,
and capacity. Therefore, the convergence between cloud
computing and networking is becoming more a requirement
than a desire, motivating and driving the creation of
networked cloud paradigm.
Paper Contributions and Structure
Promoting a unified management and control framework
for delivering efficiently cloud IaaS, we propose a method
for efficient mapping of user requests for virtual resources
(denoted as virtual network requests) onto a shared
substrate interconnecting previously isolated islands of
computing resources. The problem essentially amounts to
solving optimally the real-time problem of mapping virtual
resources to substrate resources with limited assets (e.g., of
virtual nodes and virtual links also known as virtual
network embedding (VNE) problem [7]).
Specifically, in our work toward optimizing the cloud,
we study and formulate the corresponding VNE problem
in the envisioned networked cloud computing environment
(in the following we will refer to this problem as
networked cloud mapping (NCM)). Following the methodology
introduced in [7] and the above considered cloud
service paradigm, our work aims to:
CONCLUDING REMARKS
In this paper, we study the virtual resource allocation
problem for networked cloud environments, incorporating
heterogeneous substrate resources, and provide an appropriate
approximation approach to address the problem.
Specifically, for the node mapping phase, we provide a MIP
problem formulation capable of taking into consideration
QoS requirements. Appropriate relaxation and application of
a randomized rounding technique leads to a polynomialtime
solution. Following, link mapping is determined by
solving the corresponding multicommodity flow problem.
The proposed solution is compared against two well-known
approaches on embedding virtual resource requests to a
physical substrate. Based on extensive modeling and
experimentation, utilizing CV I-Sim—a simulation/emulation
environment that allows for a flexible and structured
evaluation of the performance and efficiency of the proposed
approach, we conclude that the proposed NCM approach
overall outperforms other commonly applied algorithms.
Specifically, NCM provides a tradeoff between G-SP and
G-MCF in terms of acceptance ratio of NCM requests and
number of hops on the substrate, per virtual link. At the same
time, NCM manages to embed requests that generate
more revenue, at a cost similar to G-MCF. An appropriate
reconfiguration strategy has been also adopted to deal with
the highly dynamic networked cloud environment.