1563
Session-Based Cloud Video Delivery Networks in Mobile Internet
joint content delivery path construction. They formulated
the problem as an integer-program problem and provided
of ine and online heuristics to solve it. In [19], the authors
proposed a resource auto-scaling method to provide the
bandwidth guarantees for video-on-demand applications.
However, they did not consider the replica placement
problem and storage constraint. Papagianni et al. [20] built
a hierarchical framework for the CDN deployment problem
on the multi-provider network cloud environment. Hu et al.
[21] studied the resource allocation and replica placement
problems for cloud-based CDNs by considering diversities
in user demand patterns. However, they all assumed that
contents are xed and of the same size.
Overall, the traditional algorithms are inadaptable
for delivery of videos in mobile Internet because of the
mobility of mobile users. The dynamics of the explosively
increasing video contents aggravate the problem further,
requiring new solutions to deliver videos ef ciently.
3 Network Model
Traditional VDNs generally consist of three kinds of
logical nodes [4][22] (Figure 1). Origin servers import
videos into the networks. Surrogate sites serve the end
users directly. Re ectors forward video contents from the
origin servers to the surrogate sites. In this situation, video
contents should be pulled or pushed to the surrogate sites
from an origin server, which can lead to instantaneous
congestion problems and low scalability. In contrast, we
propose to rent cloud resources to build a cloud “ecosystem”
in the mobile Internet. In this system, the VDN can be
deployed dynamically and the resource of networks can be
updated periodically.
Figure 1 Traditional VDN Structure
3.1 Confi guration of the Cloud-Based VDN
We assume that one cloud VDN can support multiple
providers (or origin servers). In addition, a control center
is provided to administer these providers. The origin
servers can replicate the video content at multiple cloud
sites (or surrogate sites) to serve all mobile user requests.
We assume the existence of M origin servers, N surrogate
sites, and U mobile users. The set of origin servers can be
expressed by M = {m
1
, m
2
, ..., m
M
}. The surrogate sites can
be expressed by N = {n
1
, n
2
, ..., n
N
}. Each node in M and N
has location coordinates (x
m
, y
m
) or (x
n
, y
n
). We denote the
mobile user set as U = {u
1
, u
2
, ..., u
U
}. The cloud VDN can
be modeled as an undirected graph G = (H, E), where H
∈
{M
∩
N}. Each link e(i, j)
∈
E between two sites i and j is
associated with a bandwidth capacity weight value BW(e(i,
j)) (or BW
ij
).
Figure 2 shows an example scenario of the cloud
VDN. The video content of different providers is
maintained on their origin servers (m
1
– m
2
). The origin
servers are far from the mobile users. Thus, providers rent
surrogate servers (n
1
– n
4
) to place replicas and satisfy
user demands (or requests). The bold lines among the n
1
– n
4
and m
1
– m
2
represent the potential replica delivery
paths. In this model, the surrogate sites can also be used
as intermediate nodes for routing videos from the origin
to other surrogate sites. Every surrogate site has two radio
interfaces: one for communication with other surrogate
sites (backhaul connections) and another for providing
access to mobile users. Mobile users can be any devices
using any accessing technologies (e.g., laptops or desktop
computers using Wi-Fi, cell phones using Global System
for Mobile communication/General Packet Radio Service,
smart phones using 3G/4G/NxtG accessing technology,
and so on). These mobile users (u
1
– u
4
) can only access
(n
1
– n
4
) within their coverage (drawn as dashed lines). If
proper surrogate sites exist among the n
1
– n
4
that requested
video contents and also satisfy the QoS requirements, user
requests can be directed to these proper servers. Otherwise,
some requests may be routed to the root m
1
– m
2
through
the multi-hop route forwarded by the n
1
– n
4
.
Figure 2 Cloud-Based VDN Model
JIT 18(7)-09 Hu.indd 1563 2017/12/20 下午 07:42:51