Journal of Communications and Information Networks
30
minal. The last-hop access is an essential part in the
entire end-to-end video delivery pipeline. It often
directly impacts the QoS and QoE of users. The wire-
less and mobile edge access always presents tricky
issues considering the high dynamics in the wireless
networks. Such dynamics could lead to annoying
QoE degradation such as video re-buffering and vid-
eo quality fluctuation. These QoE impairments will
get even more complicated due to the diversity of ter-
minals, network modes, and QoS/QoE requirements
among different users.
In order to enhance the effectiveness of the these
three components and eventually enrich users’ video
experience, cloud services have been introduced in
the preparation, distribution, and access of video con-
tents by reshaping a pool of computing, storageand
networking resources. Cloud service providers build
data centers that house a plethora of regular servers/
workstations as well as video-dedicated CPU/GPU
arrays to empower the efficient video processing.
Besides, sizable storage capacity are also equipped
for video storage. With these video-friendly resourc-
es, it is becoming possible to improve users QoE by
making proper use of rich cloud resources. It is also
important to note that the boundary between cloud
services providers and the above three architectural
components of video delivery system are becoming
vague nowadays. For instance, video services and
cloud services can be deployed jointly by one single
provider, e.g., Amazon.
In the remaining of this article, we adopt this end-
to-end architectural framework to examine emerging
research on how cloud computing can support the
three components in order to enhance users’ QoE in
various video services.
3 Cloud computing in video servers
In this section, we examine how the cloud resource
can provide support in backend video servers. Due to
the mismatch between the computation-intensive vid-
eo services and the resource-limited thin clients, an
increasing number of video services will need to shift
the video preparation tasks from traditional single
device to the cloud in order to achieve an effective
video content provisioning.
We illustrate a general modular design of the
cloud-assisted video servers in Fig.2. A service cli-
ent first sends a service request to the cloud video
servers. The network server in the cloud handles the
request and delivers the request to the load balancer.
The load balancer then schedules dedicated media
GPU/CPU in the cloud, i.e., the video processor,
based on the current resource allocation in the cloud
and the characteristics of the request. Finally, each
individual video processor takes over the processing
tasks of content provision assigned to them and ef-
ciently completes the tasks in a parallel fashion via
virtualized instances of machines. The prepared video
content will then be transmitted to the frontend users via
the network server using standard network protocols.
Figure 2 A general modular design of video servers in the
cloud
3.1 Video encoding in the cloud
Video encoding has been recognized as a key task
that consumes substantial computation resources. Its
task is to compress the raw video data captured by the
cameras or video recorders into encoded video bit-