Heterogeneous Embedded Multicore Processors
LiangZhang, PeiYi Shen
XiDian University
Xi’an, China
liangzhang@xidian.edu.cn
pyshen@xidian.edu.cn
JuanSong, Lb Dong, GuoXiWang, YuXin Cai
XiDian University
Xi’an, China
songjuan@mail.xidian.edu.cn
caiyuxin@stu.xidian.edu.cn
Abstract—Video and Image processing applications are
notably time consuming, especially for large number of small
image files and huge video files. Cloud computing gives a new
processing method for such requirement, but almost all of the
cloud computing platforms are based on super computers. In
this paper, we propose a new method for one big image file's
and large number of small image files' processing based on
MapReduce and give a new idea of constructing the cloud
computing platform based on heterogeneous embedded multi-
core processors. From the experiment results, it can be seen
that the proposed method is feasible and after analysis, we find
the problems of embedded cloud computing platform based on
Hadoop with MapReduce, and conclude the future research
direction.
Keywords-component;Cloud;MapReduce;heterogeneous;
embedded multi-core processors
I. INTRODUCTION
As the evolvement of computer and information science,
the volume of video and image data to be processed
increases significantly, either as the expanding amount of
information available
[20]
or the popularization of video
surveillance
[2]
, in which the HD IPCAM and HD DVR have
been widely used
[13]
. According to a recent research report
from the International Data Corporation, the amount of
digital information has exceeded the zettabyte barrier in
2010 and this trend is expected to continue to grow “as more
and more embedded video terminal and systems pump their
bits into the digital cosmos”. Nevertheless, when user wants
to deal with these big video data instantaneously, such as
real time video analysis, video retrieval, video transcending,
video service or semantic analysis, simple computer can't
satisfy. Now the cloud computing platform gives us another
resolution to deal with such requirements
[15]
. In recent years,
MapReduce framework
[7][3]
has emerged as one of the most
widely used parallel computing platform for processing the
extremely large-scale data, while how to deal with the video
and image files have not been researched deeply within such
framework.
MapReduce is a programming model for the data-
parallel
[17]
programs , which simplifies the parallel
programming, hides the synchronization and dynamic task
allocation Now many parallel programming tools based on
MapReduce have been developed, such as Phoenix
[22][18]
,
Metis
[12]
, and so on. In video processing application, Keimel
et.all
[4, 9]
proposes a video Quality Crowd framework based
on Cloud Computing, in which the crowd-based video
quality assessment is realized with a simple web
interface.Zhenghua Li
[11]
proposes and implements a
“Cloud Transcoder”, which utilizes an intermediate cloud
platform to bridge the format/resolution “gap”. But the
exactly parallel transcending method has not been released,
and only the architecture of the cloud is given. Rafael
Pereira
[14]
presents architecture for distributed high
performance video processing in cloud with Hadoop and
MapReduce framework. The Split and Merge architecture
for high performance video processing are proposed.
Now many data centers and those cloud computing
platforms are constructed by personal computers
[16]
, in
which power consumption is a big problem
[19]
. Embedded
processors, such as multicore processor, DSP processor or
FPGA, have been developed and widely used in video and
image processing. Such processors have advantages of low
power consumption and high image processing capability.
YoungHoon Jung proposes a broadband embedded
computing system with MapReduce
[8]
. This system
combines a central cluster of Linux servers with a
broadband network embedded STB devices, word count and
some benchmarks are tested, but it has not involved the
image processing.
Our work is motivated by the idea of real time processing
for image and video file based on the embedded multicore
processor, especially for the processor which contains not
only the arm processor used to control, but also the DSP
processor for image processing. First, we propose the
methods for large numbers of small image files and single
big image file processing within Hadoop framework.
Second, we illustrate that how to construct an embedded
An Image Processing System Based on
2014 Second International Conference on Advanced Cloud and Big Data
978-1-4799-8085-7/14 $31.00 © 2014 IEEE
DOI 10.1109/.26
157
2014 Second International Conference on Advanced Cloud and Big Data
978-1-4799-8085-7/14 $31.00 © 2014 IEEE
DOI 10.1109/.26
157
2014 Second International Conference on Advanced Cloud and Big Data
978-1-4799-8085-7/14 $31.00 © 2014 IEEE
DOI 10.1109/CBD.2014.27
157
2014 Second International Conference on Advanced Cloud and Big Data
978-1-4799-8085-7/14 $31.00 © 2014 IEEE
DOI 10.1109/CBD.2014.27
157