PROBABILISTIC MODEL-DRIVEN VIDEO SCHEDULING FOR THE SOCIAL MEDIA
APPLICATION IN THE FEDERATED CLOUD
Zhen YANG
School of Computer Science, Beijing University of Posts and Telecommunications, Beijing 100876,
China.
yangzhen@bupt.edu.cn
ABSTRACT
1
Federated cloud is a trend in cloud computing which, by
interconnecting multiple separate clouds at different
geographical locations, can provide a cloud platform with
much larger resource capacities. Such a cloud is ideal for
supporting the social media applications. The videos in the
social media applications are always short clips and stored
in different cloud sites. Usually, when finishing a current
video, a user quickly tracks and loads second video
according to their interest. The behaviors are called as User
Interest Tracking behaviors (UIT behaviors). The frequent
UIT behaviors pose a great challenge on the UIT waiting
delay which is defined as the interval between the request
time and the playback time of a video.
In this work, we have developed a video scheduling
framework to reduce the waiting delay for social media
applications in the Federated cloud. We first propose the
UIT probabilistic model which can estimate the video access
probability based on the UIT statistics. Based on the
probability, we further develop a probabilistic model-driven
video scheduling scheme to minimize the expected UIT
waiting delay. With extensive simulations, we demonstrate
that the proposed scheduling scheme works very well.
Index Terms—video scheduling, federate cloud, UIT
behavior, UIT waiting delay
1. INTRODUCTION
The services provided by one individual cloud provider are
typically limited to one or a few geographic regions,
prohibiting it from serving more users [1]. Federation of
geo-distributed cloud service is a trend in cloud computing
which, by interconnecting multiple separate clouds (data
centers) at different geographical locations, can provide a
cloud platform with much larger resource capacities. The
federate cloud has some advantage, such as to locate
resources closer to users, to reduce bandwidth costs, to
1
This work reported in this paper was supported by the
NSFC under Grant 61173017.
increase availability, etc. The elastic and on-demand nature
of resource provisioning in the federated cloud is perfect for
supporting large-scale social media streaming applications.
Recently there is an upsurge of interest in the research
community in issues arising from running traditional media
streaming applications on clouds (e.g., VoD) [3] [4]. The
researchers have started to investigate the deployment of
elastic video streaming services over the cloud infrastructure.
For example, Li et al. [3] discuss video scheduling of VoD
services onto a cloud platform, by exploring demands and
user patterns in a conventional VoD application. Wu et al. [4]
deploy a VoD application on an IaaS cloud containing a
single data center.
However, compared with traditional video services, social
media applications feature highly dynamic demands. Users
are interconnected in a social network. Understanding user
behavior in social network is a key of provider QoS [2].
User behavior covers various social activities that users can
do online, such as friendship creation, content publishing,
video viewing, messaging, and commenting [6].
Usually, when finishing a first video, a user quickly
tracks and loads second video according to their interest [2].
For example, after a user watching a goal video of a NBA
match, he or she may want to watch another goal video. The
behaviors have also been frequently observed in [2,6]. We
call the behaviors as User Interest Tracking behaviors (UIT
behaviors). Cheng et al. [8] present an in-depth and
systematic measurement study on the characteristics of
YouTube videos and user behavior, they find that the length
of a YouTube video is short, and the UIT behaviors appear
frequently.
The frequent UIT behaviors pose a great challenge on the
UIT waiting delay, which is defined as the interval between
the request time and the playback time of a video. Since
most of the videos in the social media applications are short,
e.g., few minutes, a response time of more than several tens
of seconds would be unbearable to a user [5]. If the second
video is held by the local cloud site (i.e. the nearest cloud
site from the user), the system can quickly locate and
dispatch it to achieve a smooth transition; if the second one