IJRET: International Journal of Research in Engineering and Technology
eISSN: 2319-1163 | pISSN: 2321-7308
USER PREFERENCE PROFILE BASED CACHING FOR VIDEO
STREAMING IN MOBILE NETWORKS
Jimy George1, Shinto Sebastian2
1
PG Scholar, Department of Electronics and Communication Engineering., Amal Jyothi College of Engineering,
Kerala, India
2
Assistant Professor, Department of Electronics and Communication Engineering., Amal Jyothi College of
Engineering, Kerala, India
Abstract
In radio access networks, mobile devices access internet content through base stations, which are connected to the internet via
gateway routers. A number of studies have been carried out to improve the performance of internet access by mobile hosts.
Among these, caching of popular web documents at locations close to the mobile clients is an effective method. The benefits of
caching data in the core network of mobile carrier have been investigated in the literature. There has been some research in
caching of videos for online video sites like YouTube, which aims at improving the performance of the mobile internet access
networks. Caching of Most Popular Videos (MPV) together with LRU (Least Recently Used) - replacing the least recently used
video when cache is full- algorithm is an efficient method for large sized caches which are only possible for internet content
delivery networks. Caching algorithms for ad-hoc networks has also been developed. These are not applicable for radio access
networks. This paper investigates the effectiveness of placing caches at base stations in cellular networks and considers the UPP
(User Preference Profile) of each cell-site for caching of videos.
Keywords: Radio Access Networks, MPV, LRU, UPP
--------------------------------------------------------------------***---------------------------------------------------------------------1. INTRODUCTION
There is a tremendous growth in the use of smart phones and
tablets for accessing internet content in past few years.
Recent studies revealed that the number of mobile users,
streaming videos from online video sites like YouTube is
increasing. When internet video is accessed by a mobile
device, the video must be fetched from the servers of a
Content Delivery Network (CDN) [1]. Mobile phones are
primarily connected to their respective base stations for
communicating with each other. In LTE (Long Term
Evolution) system, the base stations in Radio Access
Networks (RAN), also known as eNodeBs connect with the
User Equipment (UE) directly. Videos delivered from CDN
reach UE via gateway routers which interconnect RAN and
the Internet. Bringing each requested video in this manner
incurs video latency and may lead to congestion in the
wireless carrier core network. In this context, caching of
most probable videos gained relevance.
Caching of popular web documents at proxy servers in
computer networks was studied extensively in the literature.
The relative frequency with which web pages are requested
follows Zipf’s law [2]. The Zipf’s law states that the relative
probability of a request for a web page is inversely
proportional to its statistical rank. That is for ith most
popular page, the probability is proportional to 1/i. A
number of works have been done on the popularity of online
videos. Studies by M. Cha et al. in [5] revealed that video
popularity also follows a Zipf distribution. It indicates that a
mere 10% of the online videos have nearly 80% views,
while the remaining accounts for only 20% of views.
Caching of videos based on UPP calculates the probability
of each video by considering the preference of users in a
cell.
This paper presents a survey on
different caching
techniques for online videos developed so far, taking into
account the location of caches and criterion for selecting
videos to be cached. The effectiveness of placing video
cache at each eNodeB in the RAN and caching policy based
on UPP is also studied.
2. ANALYZING CACHING BENEFITS
To improve the performance of video delivery on radio
access networks various caching techniques have been
proposed. Some important properties of video traffic that
have an impact on a cache are: size of a video, number of
views for a particular video and inter-arrival time of requests
for a video [4]. As referred from the literature, video
popularity follows a Zipf distribution i.e., most popular
videos contribute to only 10% of the total video population.
Hence caches need to consider only those videos which have
most of the views i.e., by caching only 10% of the total
video content, most of the requests can be served. The
number of videos that can be cached depends on the size of
the cache. Most of the caching strategies in the literature
prefer caching of video chunks video whereas some
investigates caching of full videos. If only some parts of a
requested video is in the cache, the rest need to be fetched
from the main server. This may cause stalling/buffering
during playback and may deteriorate the user’s quality of
experience. The technique involving caching of full videos
_______________________________________________________________________________________________
Volume: 05 Issue: 07 | Jul-2016, Available @ http://ijret.esatjournals.org
528
IJRET: International Journal of Research in Engineering and Technology
limits the number of videos that can be cached and may lead
to low cache-hit ratio. Cache replacement policies are also
important for maintaining the cache up-to-date. Zink et al. in
[5] propose least recently used replacement scheme: if the
cache is full and a new video needs to be cached, the video
that has been used least recently is evicted to create space
for the new one. Caches need to be updated with the
changes in the video popularity statistics for achieving a
good cache-hit ratio. The experimental results in [5] show
that cache-hit ratio improves with the cache size.
3. CACHING TECHNIQUES
With the growing interest in web videos, numerous strategies
for video caching have been developed. Caching of
frequently requested videos reduces the user perceived
latency and server/wireless network loads. Proxy server
architecture for content caching in high speed local area
networks has been proposed in [3]. The architecture consists
of a number of proxy servers, where one of these acts as a
broker or central controlling agent and others act as sibling
caching proxies. All the client requests are routed through the
broker which is an enhanced RTSP proxy server.
The performance metric used to evaluate various caching
techniques is cache-hit ratio. Whenever the requested video
can be found in the cache, we say there is a cache-hit
whereas its absence results in a cache-miss. If the number of
videos stored in cache is large enough so that most of the
requests can be served from the cache, high cache-hit ratio
can be achieved. Hence for large sized caches the hit ratio
will be large compared to small micro-caches.
Caching of Most Popular Videos (MPV) is a proactive
caching policy, where most popular videos according to the
available video popularity distribution are cached. In this
case, cache update occurs only when the video popularity
distribution changes. MPV technique is used in CDNs where
caches are of large size.
Cache replacement policies play an important role in
achieving high cache-hit ratio. When cache is full and a new
content which is not in the cache and became popular within
a short span of time needs to be cached, some content in the
cache should be replaced. Cache replacement policies
determine the content to be replaced to create space for new
content. There are two categories of replacement policies:
single-factor policies and multiple factor policies [6]. FirstIn-First-Out (FIFO) policy, random replacement policy,
LRU, and LFU (Least Frequently Used) are some examples
of single-factor replacement policies. Single factor
replacement policies consider only a single factor such as
popularity, age or cost for selecting the content to be
replaced while multiple factor policies take into account a
combination of these factors for selecting the content to be
evicted.
LFU policy maintains a reference count for each object in
the cache. When cache is full, the object to be replaced is
selected as the one with the least reference count. If a tie
occurs, LRU is applied which replaces the least recently
eISSN: 2319-1163 | pISSN: 2321-7308
used object. Another replacement policy discussed in the
literature is Greedy Dual Size (GDS) [7] which combines
temporal locality, size and other cost information. In [6], a
new rank value based replacement algorithm is proposed.
By considering the size, cost factor and age of the video
along with the Zipf-like law for video popularity
distribution, they have obtained better results compared with
other algorithms.
A collaborative content caching algorithm for mobile ad-hoc
networks is proposed in [9]. The algorithm consists of:
initial selection of backbone nodes, cache placement policy
and cache replacement policy. Selection algorithm selects
between different nodes and forms a network of virtual
access points. These nodes are responsible for caching video
segments. They jointly decide which of the nodes should
cache the requested video segments. Replacement policy
used is LRU.
Video content delivery using coded distributed caching for
wireless networks is discussed in [10]. The key idea is to use
helper stations to cache video files and to deliver a requested
video to UE via short-range wireless links. The files are
encoded such that distributed storage is made possible and
hence improving robustness and storage capacity. Helpers
are placed uniformly in a square grid spanning the cellular
region. The range of each helper is 100m.
Content caching at the base station of RAN is a way to
reduce backhaul transmission and to improve the quality of
experience. Reference [11] introduces an edge caching
mechanism where caches are located at the edge base
stations. These caches are fully interconnected and are able
to disseminate contents via direct links. Thus all
interconnected caches are considered as a single entity and
duplication of contents in these caches can be avoided. This
facilitates placement of other unavailable contents in the
cache.
In [12], video caching problem in mobile networks is
addressed where the caching nodes are distributed along
with the mobile gateways. That is, caches are located in the
core network of the wireless carrier. With collaborative
caching, the cached nodes at gateways are interconnected.
When a request for a video arrives at a gateway node, the
node checks whether it has the video in its local cache. If the
video is available in the cache, it is directly delivered to the
client and if it is not there, the video is fetched from any one
of its neighboring nodes. Each serving gateway has a cache
associated with it and they are interconnected via Packet
Data Network (PDN) gateways and possibly by other
network routers.
4. UPP BASED CACHING
The popularity distribution of videos at national level does
not reflect the local video popularity [5]. So in a mobile
network the local popularity distribution of videos must be
calculated on a cell-by-cell basis. The active users in a cell
at any given time determine the probability distribution of
videos in that cell. Hence a User Preference Profile (UPP)
_______________________________________________________________________________________________
Volume: 05 Issue: 07 | Jul-2016, Available @ http://ijret.esatjournals.org
529
IJRET: International Journal of Research in Engineering and Technology
can be assigned with each active user which indicates the
probability that a user request a particular video given all
video categories. Active users hereby mean the mobile users
who have watched a video when present in the cell or those
who are having an active video session [1]. Since in LTE
networks, the connection statistics of each UE is known to
their respective eNodeBs the number of active users in a cell
can be determined. Thus placing caches at each eNodeB in
the RAN is beneficial in the sense that the cache serving
each cell-site contains the most probable videos according to
the UPP of users in that cell.
Overall video popularity distribution can be used to find the
popularity of each video in different video categories. Once
individual user preferences are available, the probability that
a video being requested can be calculated. Based on the
probabilities videos can be classified as Most Likely
Requested(MLR) and Least Likely Requested(LLR) sets.
Cache update is performed based on these probability
values. The caching algorithm ensures that the cache
memory space is always loaded with most probable videos.
5. SIMULATION RESULT
Simulations are performed using MATLAB. A reactive
caching policy which updates cache only upon receiving
requests from clients is simulated. The cache hit ratio
obtained for cache sizes from 10GB to 200GB is evaluated
and the result is compared against two conventional cache
replacement policies – LRU and LFU. The results indicates
that UPP based caching has better performance for all the
cache sizes.
Fig-1: Cache hit ratio v/s cache size
6. CONCLUSION
With the increasing popularity of video streaming on mobile
devices, new methods should be formulated to efficiently
manage the RAN backhaul bandwidth and to provide the
users with better quality of experience. Caching of most
frequently requested videos at locations close to mobile
clients is found to be an effective method to reduce the
video latency. This paper surveyed on different caching
eISSN: 2319-1163 | pISSN: 2321-7308
policies adopted in the literature. Different criteria for the
selection of videos to be cached and various cache
replacement algorithms are discussed. User Preference
Profile based caching will be useful as it considers the
popularity of videos within each cell to select the videos to
be cached and those to be evicted when the cache is full. If
cache locations are at each edge nodes, video latency can be
significantly reduced.
REFERENCES
[1] H. Ahlehagh and S. Dey, “Video Aware Scheduling
and Caching in the Radio Access Networks,”
IEEE/ACM Trans. Network., vol. 22, no. 5, Oct. 2014.
[2] Lee Breslau, Pei Cao, Li Fan, Graham Phillips, Scott
Shenker,“Web Caching and Zipf-like Distributions:
Evidence and Implications”,IEEE INFOCOM, vol. 9,
no. 5, 1999.
[3] N. Laoutaris, “A closed form method for LRU
replacement under generalized power law demand,” in
Proc. MMNC, San Jose, CA, USA, Jan. 2008.
[4] S. S. Krishnan and R. K. Sitaraman, “Video stream
quality impacts viewer behavior: Inferring causality
using quasi- experimental designs,” in Proc. ACM
Conf. Internet Meas., Conf., 2012, pp. 211-224.
[5] M. Cha, H. Kwak, and P. Rodrigues, “Analyzing the
Video Popularity Characteristics of large scale user
generated content systems,” IEEE/ACM Trans.
Network., vol. 17, no. 5, pp. 1357-1370,Oct. 2009.
[6] G. Nair and P. Jayarekha, “A Rank Based Replacement
Policy for Multimedia Server Cache Using Zipf-like
Law,” J. Comput., vol.2, no.3, Mar.2010.
[7] Breslau, L., Cao, P., Fan, L., Phillips, G., and Shenker,
“Web Caching and Zipf-like Distributions: Evidence
and Implications,” in Proc., IEEE/INFOCOM,
USA,vol.1, pp. 126-134, Mar.1999.
[8] N.Wakamiya,M.Murata, and H.Miyahara,, “Video
Streaming System with Co-operative Caching
Mechanisms,”inProc.,SPIE Int. Symp.,2002, pp.305315.
[9] Y. Abdelmalek, A. Abad El Al, and T. Saadawi,
“Collaborative Content Caching Algorithms in Mobile
Ad-Hoc NetworkEnvironment,” IFIP/IEEE Int. Symp.
on Integrated Network Mgmt., 2009.
[10] NeginGolrezaei, KarthikeyanShanmugam, “Wireless
Video Content Delivery through Coded Distributed
Caching,” IEEE Comm. Theory, 2012.
[11] Kai Dong, Jun He, and Wei Song, “QoE-Aware
Adaptive Bit-Rate Video Streaming over Mobile
Networks with Caching Proxy,” Int. Conf. on
Computing, Networking and Communications., 2015.
[12] Yang Li, YuemeiXu, “A Novel Co-ordinated Edge
Caching with Request Filtration in Radio Access
Networks,” The Scientific World Journal., vol.2, no.3,
Nov. 2013.
_______________________________________________________________________________________________
Volume: 05 Issue: 07 | Jul-2016, Available @ http://ijret.esatjournals.org
530
IJRET: International Journal of Research in Engineering and Technology
eISSN: 2319-1163 | pISSN: 2321-7308
BIOGRAPHIES
Jimy George is a PG
Communication Engineering.
scholar
in
Mr. Shinto Sebastian is working as
Assistant Professor in Electronics and
Communication Engg. His area of
specialization
is
Communication
Engineering.
_______________________________________________________________________________________________
Volume: 05 Issue: 07 | Jul-2016, Available @ http://ijret.esatjournals.org
531