Towards Efficient Virtual Network Embedding
Across Multiple Network Domains
Meng Shen
∗
, Ke Xu
∗
, Kun Yang
†
, and Hsiao-Hwa Chen
‡
∗
Tsinghua National Laboratory for Information Science and Technology
Department of Computer Science, Tsinghua University, P. R. China
†
School of Computer Science & Electronic Engineering, University of Essex, UK
‡
Department of Engineering Science, National Cheng Kung University, Taiwan
{shenmeng, ke.xu}@csnet1.cs.tsinghua.edu.cn; kunyang@essex.ac.uk; hshwchen@mail.ncku.edu.tw
Abstract—Network virtualization provides a promising way
to run multiple virtual networks (VNs) simultaneously on a
shared infrastructure. It is critical to efficiently map VNs onto
substrate resources, which is known as the VN embedding
problem. Most existing studies restrict this problem in a single
substrate domain, whereas the VN embedding process across
multiple domains (i.e., inter-domain embedding) is more practical,
because a single domain rarely controls an entire end-to-end path.
Since infrastructure providers (InPs) are usually reluctant to
expose their substrate information, the inter-domain embedding
is more sophisticated than the intra-domain case.
In this paper, we develop an efficient solution to facilitate
the inter-domain embedding problem. We start with extending
the current business roles by employing a broker-like role,
virtual network provider (VNP), to make centralized embedding
decisions. Accordingly, a reasonable information sharing scheme
is proposed to provide VNP with partial substrate information
meanwhile keeping InPs’ confidential information. Then we
formulate the embedding problem as an integer programming
problem. By relaxing integer constraints, we devise an inter-
domain embedding algorithm to handle online VN requests
in polynomial time. Simulation results show that our solution
outperforms other counterparts and achieves 80%-90% of the
benchmarks in an ideal scenario where VNP has complete
knowledge of all substrate information.
I. INTRODUCTION
Network virtualization has emerged as a powerful way to
allow multiple virtual networks coexist in a shared infras-
tructure. This is done by decoupling infrastructure providers
(InPs), who are responsible for managing the connectivity
of the substrate networks, from service providers (SPs), who
are dedicated to providing services to end users [1]. Re-
cent advances in network equipment virtualization [19] and
software-defined networking (SDN) [20] open possibilities for
implementing such an Infrastructure as a Service (IaaS) model,
where service providers deploy their protocols on customized
virtual networks and pay for the resource usage.
A virtual network (VN) is a collection of virtual nodes
(e.g., routers) interconnected via a set of virtual links. Each
VN, with requirement constraints on its virtual nodes and links
(e.g., CPU and link bandwidth [7]), should be mapped onto
specific physical nodes and links in the substrate network,
which is known as the VN embedding problem. Since multiple
VNs share the same substrate resources, an efficient embedding
algorithm is crucial to increase the utilization of the substrate
resources and the revenue of InPs [8]. In this paper we use
embedding and mapping interchangeably.
Many algorithms and mechanisms have been proposed
to make efficient embedding decisions, e.g., [4, 5, 7–10].
These studies restrict the embedding problem within the same
substrate domain, which is referred to as the intra-domain em-
bedding hereafter. However, an individual InP rarely controls
an entire end-to-end path [1], service providers thus desire
virtual networks across multiple substrate domains so as to
offer value-added services to end users. For example, Voice
over IP (VoIP) needs to run a dedicated virtual network that
provides services with guaranteed performance (e.g., latency)
to users in multiple network domains; live-streaming video
providers require high-throughput virtual networks that deliver
real-time video resources to their geographically distributed
private servers. In this paper, we refer to the VN embedding
across multiple substrate domains as the inter-domain embed-
ding.
We argue that the inter-domain embedding is greatly differ-
ent from the intra-domain one, and thus algorithms proposed
for the latter cannot be directly applied to the former case.
Within a single domain, both the detailed substrate network
topology and VN requests are visible to an InP, who, in
turn, makes optimal embedding decisions according to its
operational goals (e.g., minimizing the embedding cost of the
current VN request). In the inter-domain scenario, however,
InPs are typically reluctant to share substrate topology with
each other. Therefore, no individual InP can make a global
embedding decision for a cross-domain VN request.
A straight-forward solution to the inter-domain embedding
can be described as follows: a service provider, who requires
a cross-domain VN, should first coordinate with each of
the potential InPs and then decide to require which InP to
accommodate which component of the entire VN. It makes
the service provider involved in multiple bilateral coordina-
tions and negotiations, and thus sacrifices a major benefit of
decoupling these two roles. In particular, it is even more costly
for SPs who need to quickly set up short-term VNs.
In this paper, we develop an efficient solution to the inter-
domain embedding, which frees service providers from sophis-
ticated negotiations, while protecting infrastructure providers
from revealing their confidential information. The major con-
tribution of this paper, which comes along with addressing978-1-4799-4852-9/14/$31.00
c
2014 IEEE
2014 IEEE 22nd International Symposium of Quality of Service (IWQoS)
978-1-4799-4852-9/14/$31.00 ©2014 IEEE 61