Energy-aware Virtual Network Embedding
Through Consolidation
Sen Su
*
, Zhongbao Zhang
*
, Xiang Cheng
*
, Yiwen Wang
*
, Yan Luo
†
, and Jie Wang
†
*
State Key Laboratory of Networking and Switching Technology,
Beijing University of Posts and Telecommunications
{susen, zhongbaozb, chengxiang, wangyiwen}@bupt.edu.cn
†
University of Massachusetts Lowell
yan luo@uml.edu, wang@cs.uml.edu
Abstract—The rising cost of electricity has forced business
organizations to find new ways to cut energy spending. This
paper studies how to reduce energy consumption in virtual
network embedding, which embeds virtual networks (VNs) re-
quested by users to a shared substrate network (SN) run by an
infrastructure provider (InP). Previous research has primarily
focused on finding embedding methods to increase revenues by
accommodating more VN requests in a fixed SN, with little
attention to reducing the energy cost. To fill this gap we formulate
an energy consumption model and devise an efficient energy-
aware VN embedding algorithm using a consolidation technique.
Through preliminary simulations we show that our algorithm
can significantly reduce energy consumption by up to 30% over
the existing energy-oblivious algorithm, while obtaining attractive
revenues for the InPs.
I. INTRODUCTION
Network virtualization, allowing multiple heterogeneous
virtual networks (VNs) to coexist on the same shared substrate
network (SN), has been studied in recent years as an enabler
for the polymorphic future Internet for sharing resources and
providing adequate flexibility for network innovations [1].
In the settings of network virtualization, infrastructure
providers (InPs) and service providers (SPs) play different
roles: InPs manage the physical infrastructure while SPs create
VNs and offer end-to-end services [2]. Mapping/Embedding
1
VN requests with constraints on both nodes (e.g., CPU) and
links (e.g., bandwidth) onto an SN of the InPs, also known as
VN embedding, has been studied extensively in [3–10]. These
early studies have primarily focused on finding embedding
methods to increase revenues generated from accommodating
more VN requests in the same SN. However, little attention
has been paid to reducing the energy cost. The rising cost
of electricity also affects the profits of InPs. Recent studies
estimated that energy-related costs represent already almost
50% of the operating costs and they are growing faster than
compute-related costs (i.e., servers and network equipments)
[11]. Thus, the energy cost has become the most important
factor for InPs to maximize their profits.
In this paper, inspired by virtual machine (VM) consol-
idation, a promising method to conserve energy in cloud
1
The terms “mapping” and “embedding” are used interchangeably through-
out this paper.
datacenter environment [12, 13], we study the novel energy-
aware VN embedding problem by consolidating VNs to the
smallest number of substrate nodes and powering down unused
substrate nodes. However, adopting the technique of consoli-
dation in the VN embedding context is facing the following
two challenges:
1) Energy consumption modeling. Abstracting the energy
consumption model, is challenging since these substrate
nodes may play different roles to satisfy the CPU and
bandwidth requirements of VN requests and thus they
have different energy consumption.
2) Energy-aware embedding algorithm design. VN embed-
ding is known to be NP-hard [3], thus when considering
energy efficiency simultaneously, it is more complex to
devise such an algorithm.
Addressing the first challenge, we identify active sub-
strate nodes and differentiate them into working nodes and
intermediate nodes. Working nodes provide sufficient com-
puting capacity and communication bandwidth for sending
and receiving packets, while intermediate nodes only forward
packets. We abstract an energy consumption model for these
nodes, based on which we present quantitative analysis of the
overall energy consumption, including the energy consumption
of virtual nodes and virtual links for accommodating a VN
request. We then formulate an integer linear programming
(ILP) problem with an objective of minimizing the number
of active substrate nodes.
To address the second challenge, we devise an energy-
aware VN embedding heuristic called EA-VNE, which is a
two-stage VN embedding algorithm. In the node mapping
stage, we use best-fit and worst-fit schemes to rank nodes for
node mapping. The best-fit scheme is used for satisfying the
node requirements of VN requests to decrease the number
of active working nodes. The worst-fit scheme is used for
satisfying the connectivity constraint to benefit the subsequent
link mapping stage and increase the probability of accepting
VN requests. In the link mapping stage, we devise a special
link mapping mechanism to decrease the number of active
intermediate nodes.
We carry out preliminary simulations and show that our
The 1st IEEE INFOCOM Workshop on Communications and Control for Sustainable Energy Systems: Green Networking and Smart
Grids
978-1-4673-1017-8/12/$31.00 ©2012 IEEE 127