Cache in the Air: Exploiting Content Caching and Delivery Techniques for5G Systems

Mr. S. Saravanan ,
Research Scholar, Bharath Institute of Higher Education and Research /Assistant Professor,
Jeppiaar Institute of Technology, [email protected]
Dr. B.Selva Priya,
Assistant Professor, Dept of Computer Science,
Bharath Institute of Higher Education and research, [email protected]

ABSTRACT
The demand for rich multimedia services over mobile in recent years, the capacity of the wireless link , mobile
networks has been soaring at a tremen- dous pace over radio network, mobile backhaul, and mobile core
recent years. However, due to the centralized architecture network cannot practically cope with the explosively
of current cellular net- works, the wireless link capacity growing mobile traffic due to the centralized nature of
as well as the bandwidth of the radio access networks and mobile network architec- tures [1]. Indeed, despite the
the backhaul network cannot practically cope with the continuous efforts of mobile network operators (MNOs)
explosive growth in mobile traffic. Recently, we have and network equipment vendors to enhance the wireless
observed the emergence of promising mobile content link bandwidth by adopting sophisticated techniques at
caching and delivery techniques, by which popular both the physical (PHY) layer and medium access control
contents are cached in the intermediate servers (or (MAC) layers in Long Term Evo- lution (LTE) and LTE-
middleboxes, gateways, or routers) so that demands from Advanced systems, such as massive multiple-input
users for the same content can be accommodated easily multiple-output (MIMO), carrier aggregation, and
with- out duplicate transmissions from remote servers; coordinated multipoint (CoMP) transmission, the
hence, redundant traffic can be significantly elim- inated. utilization efficiency of the radio spectrum is notably
In this article, we first study techniques related to caching reaching its theoret- ical cap.
in current mobile networks, and discuss potential Stemming from the observation regarding the traffic
techniques for caching in 5G mobile networks, including explosion problem as studied in [2–4], an important
evolved packet core network caching and radio access portion of mobile multimedia traffic is due to duplicate
network caching. A novel edge caching scheme based on downloads of a few popular contents (e.g., popular music
the concept of content-centric networking or information- videos) with large sizes. Therefore, researchers and
centric networking is proposed. Using trace-driven engineers have been investigating effective ways to
simulations, we evaluate the performance of the proposed reduce the duplicate content transmissions by
scheme and vali- date the various advantages of the adopting intelligent caching strategies inside the mobile
utilization of caching content in 5G mobile networks. networks, and enabling mobile users to access popular
Further- more, we conclude the article by exploring new content from caches of nearby MNO gateways (e.g.,
relevant opportunities and challenges. using selective IP traffic offload techniques [5, 6]). From
INTRODUCTION the perspective of Inter- net service providers (ISPs),
Along with recent advances in mobile communi- cation this also helps reduce traffic exchanged inter- and intra-
technologies, an ever growing number of mobile users are ISPs [3], not to mention significant reduction in the
continuously attracted to enjoy a wide plethora of response time required to fetch a content file. Thus, the
multimedia services using smart- phones and tablets [1]. impact of Internet traffic dynamics on variation of the
While the demand for rich multimedia content has response latency can be elimi- nated. Furthermore,
increased tremendously reducing traffic load via intelligent caching of popular
content would enhance the energy efficiency of 4G
networks contributing to the evolution of green 5G net-
works effectively.
 .. highest revisit rate, and cacheable con- tents from
CP C Facebook have the highest probabili- ties of revisits
4G Int 3G Conte . 2 P
C
nt
LT ern UM Ev
files C .3 P evolved packet core, EPC, and/or the RAN). The figure
E et TS olv Extern 4 illustrates how traffic is duplicated and can be reduced
P
ed
pac
al 1
. thanks to caching. We assume that 9 out of 12 mobile users
P- G
G G GS
ket
cor
traffic
MN . (u1, u2, ..., and u12) are requesting the same content from CP
W s n N e O Int via several eNBs in the MNO (n1, n2, ..., and nN). Green
5 M Ra
S- SG cac
dio Backhacore ern lines are for cables connecting EPC and RAN, while black
G s M Iu SN
hin
acc
g
ul . et lines represent the deliv- ered copies of the requested content.
ess
Ws 1 s1- .
E traffic
- Iur net
RNC Inter- The rect- angle at the MNO core and eNBs represent the
1 1 MM IPS wor n n
E k n . traffic
cell
. nN
caching storage. Without any in-network caching, as shown in
u cac 1 2 Fig. 2a, traffic is transmitted in a duplicative manner (with a
3
u8 .
b hin u u4 u
x eN g u
high redundancy of nine copies). When caching at EPC is
2 u 5 u3 u 9
ode No 6 u7 . 1 used, as illustrated Fig. 2b, the inter-ISP traffic can be
U B de u 2 u10 1 significantly reduced to just one single copy, but the intra-ISP
E B 1
. u12 traffic from EPC to users via eNBs in the RAN remains the
Intra-
cell . same (i.e., nine copies) with high redundancy. Furthermore, if
traffic
. caching is deployed at the RAN, as shown in Fig. 2c, the
intra-ISP traffic can be reduced to four, and each eNB can
Fig:Network Architecture locally satisfy the mobile users’ requests, while the EPC
caching storage can be reduced. If eNBs can exchange
which is becoming much cheaper However, the scale of content with neighbor eNBs, the caching redundancy can be
content acquired by content pro- viders (CPs) is growing further reduced.
significantly and it is thus all but impossible to cache all In this article, we investigate emerging
caching techniques for mobile cellular networks, and explore
content. It is hence important to decide what content to potential research challenges and opportunities. This article is
cache taking into account content popularity. As practi- organized as fol- lows. We discuss EPC and RAN caching
cally captured in [3, 4], only a small amount of popular tech- niques, respectively. CCN-based caching is
content is accessed by a large portion of mobile user further discussed. Related performance evalua- tion is
requests, while a long tail of con- tents remains unpopular shown. We detail new challenges and open issues. The
[4]. ). It shall be noted that different file types also have article concludes.
different “cacheability,” as studied in [4]. For instance,
among all the content types, images and videos have the

Where to cache?: As illustrated in Fig. 1, due to the

Inter Inter
networks, two places can be envisioned for net net
•The evolved packet core (EPC), which con- sists of EPC
(among many other nodes) the serv- ing gateway (S-GW), cac
cach hing
packet data network gateway (P-GW), and mobility
management entity (MME) in LTE networks apparent n n . n n .
that the tech- nical key issues fall into the following . nN . nN
u
u
u u .. u9
u
u
u u .. u9
three questions: nature of current cellular u u
.. u11 u u
.. u11
Furthermore, from relat- ed studies, it has become
•The radio access network (RAN), which con- sists of . ( .
b
NodeBs in 3G networks and evolved NodeBs (eNBs) in Inter Inter
4G LTE networks, as well as non-Third Generation net net
Partnership Project (3GPP) accesses, such as Wi-Fi EPC+
access points and worldwide interoperabili- ty for RAN
the inter-ISP traffic bandwidth, and the storage cost,
Caching in 3G mobile networks [2] and caching in n n . n n .
. n . n
4GLTE [4] networks have both been proven to be able u .
u.
u u .
u .
u
u u u
N u u u
N
to reduce mobile traffic by one third to two thirds. u .. u10 u .. u10
u u
. ( .
d
CACHING WITHIN EPC

currently contents are transmitted via eNBs period of time. As shown in Fig. 1, 3G and 4G cellular
By caching content at the mobile core net at the end of reduce user-perceived latency as well as the transmission of
at the EPC than at the RAN. redundant traffic over the network. Furthermore, caching at
higher hit rate. However, down- stream of the EPC (i.e., Both EPC caching and RAN caching can signifi- cantly
characteristics so that caching at the EPC will enjoy a EPCs will suffer from huge traffic loads.
improve the visibility of content dynamics and user Current widely deployed caching functions mostly take
work of 3G [2] and 4G [4] systems, one to two thirds place within the EPC, that is,
of mobile traffic can be reduced. Current- ly, there are at the P-GW forming the so-called mobile content
generally two kinds of caching tech- niques: delivery network (mobile CDN) [3, 4]. Locating cache
Web caching, which is object-oriented caching servers in a centralized fashion with EPC nodes at the
premises
with content awareness,and consists of web caching
based on uniform resource locator (URL) and
WEB CACHING
prefix-based web caching
URL-Based Web Caching — A large portion of mobile
Redundancy elimination (RE), which is pro- tocol-
traffic (i.e., 82 percent) uses HTTP [2]. It is therefore
independent and flow-oriented or packet-oriented,
beneficial to deploy web caching middle-boxes either
and consists of chunk-level RE, TCP-RE, and
between EPC and the gate- way to the Internet (or
packet-level RE. These are described in details
any packet data network
hereunder.
— PDN) or between RAN and EPC [3]. Every web
EPCs will suffer from huge traffic loads.
Current widely deployed caching functions mostly take content has a URL address, and each URL can
place within the EPC, that is, identify one content in most cases. If a user requests
at the P-GW forming the so-called mobile content content by a specific URL, while there is a cached
delivery network (mobile CDN) [3, 4]. Locating cache content with the same URL address in the middle-
servers in a centralized fashion with EPC nodes at the box, the content is directly returned without
premises (e.g., data center) of an MNO admittedly eases fetching from a remote server. The caching server will
the management of both EPC and mobile CDN. This can maintain the table of every content’s URL and count
also easily scale horizontally with increased demand for the access fre- quency by users’ URL requests.
content as well as the diversity of content, and will While object caching is the most effective caching
networks, as well as the expected 5G networks, generally
method from a response time and band- width
have a centralized architec- ture where all packets are
concentrated within a small number of EPCs in a country. utilization perspective, it has three fairly significant
With the improvement of RAN technologies and wider limitations:
deployments of cell towers, the backhaul links to those Aliasing of content (same content with dif- ferent URLs)
Uncacheability (i.e., temporary content or one-time-use content)
the edge of the network has the effect of smoothing traffic Content updates (same URL with different content)
spikes and balancing the backhaul traffic over a long
servers in a centralized fashion evolved version of URL-based web caching. It
with EPC nodes at the premises identifies the duplicated content cache not only by
•The radio access network (RAN), which con- sists URL or a part of the prefix of the URL, but also by
of NodeBs in 3G networks and evolved NodeBs further confirmation of the hash (prefix key) of the
(eNBs) in 4G LTE networks, as well as non-Third first N bytes of the objects and the content-length
Generation Partnership Project (3GPP) accesses, fields. It regards Prefix two content objects as of the
such as Wi-Fi access points and worldwide same content if their prefix keys match .and their
the inter-ISP traffic bandwidth, and the storage cost, sizes are equal. If N is small, there may be false
Caching in 3G mobile networks [2] and caching positives, but if N is large, the computation overhead
in 4GLTE [4] networks have both been proven to be is signifi- cant. Hence, the proper setting of N is of
able to reduce mobile traffic by one third to two vital importance
thirds
-Based Web Caching — Prefix-basedweb caching is an
REDUNDANCY ELIMINATION
The RE technique is protocol-independent as it does Small chunks can easily explode the index sizes in a high-
not depend on the delivered service or application, but speed network.
rather on the monitored chunks or packets. RE can A small chunk size would incur a higher hashing and
remove duplicate byte strings from arbitrary network content reconstruction cost, stressing the memory system,
flows by deploying two middle-boxes on the link between which typical- ly becomes the performance bottleneck in
PDN (e.g., Internet) and P-GW, and on the link between memory-based RE.
S- GW and RAN. The first middle-box divides MNOs may want to apply RE to only large file
incoming content into small chunks or packets, and downloads or specific content types that exhibit high
verifies that there are existing cached chunks with the redundancy with a small manage- ment overhead (e.g.,
same size and hash; it then sends a stream of chunk multimedia content like videos, images, and content from
hashes to the second middle- box, which reconstructs the famous social network services such as Facebook). Based
original content with those hashes and subsequently on [3], bypassing any HTTP responses smaller than 32
delivers it to mobile users. RE needs to be fast, adaptive, kbytes, the cost of managing 90.6 percent of the flows
and parsimonious in memory usage in order to could be eliminated while still ensuring
opportunistically leverage the caching resource. There are 31.6 percent bandwidth savings.
three popular types of RE techniques: CACHING AT RAN
Chunk-level RE Moving application processing resources toward the
TCP-level RE network edge closer to mobile users will make it
Packet-level RE, depending on the scale of the possible to simultaneously reduce net- work traffic and
penetration of the traffic flows improve quality of experience.
Chunk-Level RE — In chunk-based caching, all files are
split into chunks, while each chunk is hashed by the Every web content has a URL address, and each URL
caching servers. If the hash and the size of two chunks can identify one content in most cases. If a user requests
are the same, they are considered as duplicates, and future content by a specific URL, while there is a cached
requests will go to the same cached chunk. Chunk-level content with the same URL address in the middle-box,
RE solves issues relevant to aliasing, uncacheable the content is directly returned without fetching from a
content, and content update. However, it lacks awareness remote server. The caching server will maintain the table
of content and does not facilitate cooperative caching of every content’s URL and count the access fre- quency
strategies for global opti- mization by users’ URL requests.
TCP-Level RE — Due to the wide dominance of HTTP- While object caching is the most effective caching
based traffic (above TCP/IP), we hereby also discuss method from a response time and band- width utilization
TCP-level RE, which is similar in spirit to chunk-based perspective, it has three fairly significant limitations:
caching. TCP-RE works over the content of each TCP Aliasing of content (same content with dif- ferent
flow. The caching middle-boxes reassemble the segments URLs)
in the same TCP flow and divide them into smaller Uncacheability (i.e., temporary content or one-time-use
chunks (either fixed-sized or variable-sized) that serve as content)
the unit of caching. While TCP-level RE has Content updates (same URL with different content)
additional overhead to manage TCP flows, it allows
scalable cache management as well as flow-based RE can remove duplicate byte strings from arbitrary
caching policies network flows by deploying two middle-boxes on the link
Packet-Level RE — One popular caching approach is between PDN (e.g., Internet) and P-GW, and on the link
packet-level RE, which detects can- didate fragments for between S- GW and RAN. The first middle-box
RE within each IP packet on smaller chunk sizes (e.g., divides incoming content into small chunks or packets,
32–64 bytes), typi- cally using content-based and verifies that there are existing cached chunks with the
fingerprinting func- tions. By packet caching middle- same size and hash; it then sends a stream of chunk
hashes to the second middle- box, which reconstructs the
boxes, the upstream middle-box removes redundant
original content with those hashes and subsequently
bytes, and the downstream middle-box reconstructs full delivers it to mobile users. RE needs to be fast, adaptive,
packets. However, there are a few drawbacks with and parsimonious in memory usage in order to
packet-level RE in high-speed cellular back- haul opportunistically leverage the caching resource
networks:
This can further optimize the most expensive part of the referred to [10, 11].
network operating cost of the various fiber leased lines In CCN, a user requests a particular content by issuing
that connect eNBs to EPC. The research work conducted interest packets to neighbors. If the requested content can
in [7] focuses on caching popular video clips in the be retrieved in the caching store (CS) of any device, the
RAN. The “FemtoCaching” concept in [8] also discusses content is directly delivered from the device. Otherwise,
and evaluates the video content delivery through routers propagate interest towards appropriate content
distributed “caching helpers” in femtocell net- works. sources and store information for each forward- ed
However, most RAN caching solutions face the same interest in the pending interest table (PIT). Routers and
implementation issues: eNBs establish “tunnels” other network nodes will push the cached or fetched
between users and the EPC, while content files are content toward their requester(s) based on the
packetized and then encapsulated by GTP tunneling, information stored in their respective PITs and also on a
making it diffi- cult to carry out object-oriented or strategic caching policy. In mobile networks, there are
content- aware caching. many routers within the EPC and RAN whereby CCN-
To overcome this shortcoming, the natural evolution of based caching can be enabled.
caching is to cache repetitive por- tions of an object, 5G mobile networks are expected to include CCN-
known as byte caching [9]. Byte caching is a protocol-, capable gateways, routers, and eNBs. For example, the
port-, and IP- address-independent bidirectional caching adaptive mobile video streaming with offloading and
tech- nology that functions at the network layer by sharing in wireless named data networking (AMVS-
looking for common sequences of data in the bytes of NDN) framework [11] has realized a CCN architecture
packet flows. Byte caching represents an enhanced fine- in a commer- cialized WiMAX BS implementing CCN-
grained approach similar to the packet-level RE scheme. based caching and achieving significant traffic reduc-
It does not need to split flows into segments, but tion. CCN-based caching will be universal and pervasive
continuously pene- trates into the byte strings to cache in future mobile networks. Due to the wide distribution
the often used bytes and eliminate any redundant ones. of caching resources, coopera- tive caching policies
For example, if an enterprise logo is used on all should carefully consider content popularity, freshness,
documents, byte caching will identify this com- mon part diversity, and replica locations over the network
of bytes on the different files and pre- vent that data from topology in order to achieve the following objectives:
being transmitted redundantly. Minimization of inter-ISP traffic (outbound traffic): This
Unlike centralized EPC caching, one impor- tant issue of can be guaranteed when the cached content in all storage
RAN caching consists of the fact that the caching space has the highest diversity so that content can be fetched
at each eNB is practically small, and the number of users within the same ISP (supporting the same EPC and
served by indi- vidual eNBs is usually small (i.e., unless RAN) as much as possible. That is, for any content, only
in highly dense areas such as Times Square in New York one copy is stored, and content is cached according to
City), resulting in low-to-moderate hit ratios. Therefore, popularity until the whole storage of caches within the
intelligent caching resource allocation strategies and EPC and RAN is full. Minimization of intra-ISP traffic
cooperative caching policies among (neighboring) eNBs (traffic within the EPC and RAN): This can be guaran-
is mandatory for effi- cient RAN caching. teed when most popular contents are cached at each
CCN-BASED CACHING eNB, so most requests can be locally satis- fied without
An important requirement behind 4G has been the all-IP many exchanges among eNBs. This objective may
feature of its architecture. Incorporat- ing CCN induce situations when many eNBs cache the same
techniques could become an important feature of 5G popular content, so it is somehow contradictory to the
mobile networks. As per the need for decentralizing requirement of content diversity in the aforementioned
mobile CDN services, CDNs are getting distributed objec- tive.
further, incorporating information-centric and content- Minimization of content access delay of all users: Users
aware caching techniques, and forming the so-called may fetch content from local eNBs, routers in the RAN
informa- tion-centric (ICN) or content-centric and EPC, and even from remote CP servers with
networking (CCN) architecture [10] for the future different delays. To opti- mize the quality of service
Internet. The primary goal of CCN is to facilitate in-net- (QoS) of all users, it is important to consider effective
work data storage for universal caching in every network caching policy (i.e., placing an object oi in a caching
node. Major CCN designs have the fol- lowing common eNB node, RAN node, and/or EPC node) to minimize
attributes: the total delay (Eq. 1), where oi denotes a content, pi is
Receiver-oriented and chunk-based trans- port the popularity (request frequency) of oi, and teNB,
In-network per-chunk caching tRAN, tEPC, and tCP represent the transmis- sion delays
Name-based forwarding of the corresponding part of the net- work. There should
Uniquely identifiable content naming be a balance among caching choices at the local eNBs,
For further details on CCN, the interested read- er is RAN, and EPC with respect to content popularity. Good
two-day measurements of a 3G network in 2010 [2]. work OPEX is shown in Table 1. From the eval- uation of
While the fraction of cacheable objects is small (53.9 traffic reduction and the cost estima- tion, it is proved that
percent) in the measure- ment conducted in [3], the caching in the EPC and RAN can significantly enhance
cache hit ratio is similar (38.3 percent) to that obtained the capacity of current 3G and 4G networks, ensuring
in [2]. The week-long measurements in [3] with over 370 high energy efficiency, and thus should be included in
Tbytes of 3G mobile traffic reveal that standard web the solutions toward the evolution to green 5G mobile
caching can reduce download bandwidth consumption networks.
by up to 27.1 percent, while simple TCP-level RE can OPEN ISSUES AND CHALLENGES
save bandwidth consumption by 42.0 percent with a DISTRIBUTED CACHE RESOURCE MANAGEMENT
cache of 512 Gbytes of RAM. AND COOPERATIVE CACHING POLICY
Furthermore, we also carry out preliminary trace-driven An important limitation of RAN caching con- lies in
simulation to compare the perfor- mance of the the fact that the caching space at indi- vidual eNBs is
aforementioned caching techniques with respect to the usually small and the user base is quite limited, which
user access delay and traffic load if we vary the portion will potentially result in low hit ratios. Generally
of total caching size to the total content size speaking, a RAN con- sists of thousands of eNBs
(excluding duplicated interconnected via high-capacity links. A well tuned
ones). The traffic load is calculated by ForAllDe- inter-eNB cooperative caching policy is therefore
livery(ContentSize DeliveryHopCount). We use the needed that considers the dynamics of local user
IRCache HTTP content caching and for- warding traces demands, caching status of neighboring eNBs, and
[12], collected for 30 days from June to July in 2013, to global optimization of caching resource utilization as
represent the user requests of popular Internet well as the maximization of QoS of all users.
content. Then, referring to the illustrated infrastructure There is practically a trade-off between the diversity of
in Fig. 1, we apply the trace-driven traffic to a randomly content stored inside EPC and RAN and the redundancy
generalized MNO topology with 300 mobile users of the replicas of popular content in eNBs. Figure 4
attached to 30 eNBs that are interconnect- ed with 5 shows a typical sce- nario with a cooperative caching
routers and 1 EPC node. Link delay is randomly assigned policy in the EPC, RAN, and intermediate routers,
(ranging from 10 to 20 ms) for per transmission hop. wherein some popular content are stored at local eNBs,
Note that the com- monly used LFU caching policy is and some are cached in the EPC, while in a small area
utilized. As shown intuitively in Fig. 3, legacy cellular with several neighboring eNBs, one content does not
net- works without caching techniques do not necessarily need to be cached with multiple copies. In
improve users’ experienced delays much; nor do they order to achieve effi- cient cooperation among nodes
reduce overall traffic load. Depending on the (devices) in the EPC and RAN, the caching status
deployment architecture of EPC nodes, the improvement should be effectively shared among nodes (devices),
of EPC caching in users’ delay is also limited, as a certain most importantly incurring only a minimal signaling
portion of the delay is due to the link between EPC overhead. Additionally, as there are always
nodes and RAN. On the other hand, RAN caching numerous new content coming, the underlying caching
largely con- tributes to reduction in both delay and traffic policy should also consider the dynamic changes of the
load as popular content are mostly cached in the last hop. content popularity and user demands so that adequate
The proposed CCN-based universal caching in the EPC caching decisions are made online and in real time.
and RAN exhibits the best performance. This is mainly Another impor- tant issues of RAN caching pertains to
due to the fact that it caches content at every network service continuity (i.e., upon a handoff operation, termi-
device in both the EPC and RAN, employing cooperative nation of content transmission from source eNB, and
opti- mization on content distribution. prefetching of content in the target attach- ing eNB).
In addition to their advantages in mobile traf- Mobility-aware caching and prefetch- ing in the RAN are
fic reduction, EPC and RAN caching can also reduce the highly needed. In this vein, solutions similar to the
operation expense (OPEX) of MNOs by shrinking inter- Follow Me Cloud con- cept can be envisioned [14].
ISP bandwidth payment and lowering the cost due to REALISTIC RAN CACHING IMPLEMENTATION
deployment and main- tenance of cables and transport In spite of the above mentioned challenges, there are
devices in the EPC and RAN. Based on a recent analysis already a few RAN products with caching support.
of three main mobile network markets, Chicago (9.5 Notable examples are Blue
million users with about 5600 BSs), Munich (2.6 million
users with about 1300 BSs), and Hangzhou (1.8 million
users with about 1100 BSs) [13], a consistent pattern of
savings in net-
 1 1 aware at all. They thus may suffer from high
0 0 N
0 0 o implementation complexity, as well as a
9 9 c
scalability problem. There is already one study
0 0 a
c
for CCN-based content-aware RAN caching in
8 8 h the AMVS-NDN framework [11]. In this study, the
0 0 i
n CCN architecture is imported into a WiMAX
7N 7 g
femtocell station, while CCN- based caching for
o
0 0 E
c P LTE eNBs is still under real- ization
16a 4 5 6 16 2 3 4 C5 6
00c Cache size
0 0(percentage
0 00 0 0oftotal
0 0 0
h content size) c Total cache size (percentage of total content size)
5i 5 a
0n 0 (a) c (b)
g h
Figure 3. 4Performance evaluation
4 ofi EPC caching, RAN
0E CCN caching: a)
caching, and n
0 user content access delay; b)
P g
number of 3content
C transmissions.
3
0c 0 R
A
a N
2 2
c
0h 0
Application i
Cloud Server (RACS) c
a
for liquid
1n 1 Cache size
application0g
service for0
BSs. These
c exist- ing RAN (percentage
h
caching techniques
0
are 0
mainlyi based on packet-
R n
level RE or A byte caching, which g are not content-
N EPC
+RA effective caching solution for small cells
c
a
N
cach
deploying LIPA and SIPTO [8] may be required,
c
h
ing
CCN
especially for coor- dinating among neighboring
Coat ProxySGi
n
appliances, CacheMARA’s
cach small cells, and proactively and predictively
ing
caching proxy
g techniques, Altobridge’s “Data- at- caching content for home users regarding the
EPC
the-Edge”+RA solution, Saguna Networks’ Con- tent continuity and opti- mality of the content delivery
N
Optimizationcach Delivery System with caching in may become a promising research topic, which
ing
small cells,CCNand Nokia Siemens Networks Radio may bring fur- ther technological advances toward
Application cac Cloud Server (RACS) for liquid
hing
the realiza- tion of next generation content-
application service
2 for BSs. These exist- ing RAN aware 5G mobile network
caching techniques are mainly based on packet-
0 We have also presented a number of promising
level RE or byte
3 caching, which are not content- research opportunities and relevant challenges,
aware at all. They
0 thus may suffer from high particularly related to distributed cache resource
implementation complexity, as well as a management, cooperative caching policy, and content-
scalability problem. There is already one study aware RAN caching. Conclu- sively, we have
for CCN-based content-aware RAN caching in highlighted the roles that NFV, eMBMS, and D2D can
the AMVS-NDN framework [11]. In this study, the play in further improving the gains that can be
CCN architecture is imported into a WiMAX acquired from caching in mobile networks.
femtocell station, while CCN- based caching for
LTE eNBs is still under real- ization.The 3GPP Another impor- tant issues of RAN caching pertains
specifications relevant to local IP access (LIPA) to service continuity (i.e., upon a handoff operation,
and selective IP traffic offload (SIPTO) via a home termi- nation of content transmission from source
eNB [15] enable 3G/4G small cells (femtocells) eNB, and prefetching of content in the target
to directly connect to the Internet. Hence, attach- ing eNB). Mobility-aware caching and
caching at eNBs in the RAN will become like prefetch- ing in the RAN are highly needed. In this
caching at the distribut- ed and collocated EPC vein, solutions similar to the Follow Me Cloud con-
(since local PDN GWs will be a part of EPC). An cept can be envisioned [14].
 How- ever, it is quite challenging to design efficient
Inter
net
caching and sharing strategies in the RAN (eNBs)
and user devices. A potential research direction for
offloading 5G traffic may be how to effectively exploit
Route user social relationships to facilitate caching and sharing
rA
activities by D2D communication [21]. sharing among
{O1,O3,
n n n mobile users via D2D can significantly reduce the
u u1 traffice load over the cellular eNBs to users
u
u
u

INTEGRATION WITH VIRTUALIZATION OF CONCLUSIONS

CELLULAR NETWORKING DEVICES It is generally agreed that current RAN deploy- ments
The potential use of software-defined network- ing and mobile backhaul networks cannot cope with the ever
(SDN) [16] in 4G mobile networks as well as future growing demand of mobile users for rich multimedia
5G networks has been gaining lots of momentum services. Caching can be a poten- tial solution. In this
recently. In 4G mobile networks, the control and data article, we have provided an overview of some emerging
planes are separated, but with use of the emerging caching techniques for current 3G, 4G, and future 5G
network function vir- tualization (NFV) technique [17], networks, comparing between EPC caching, RAN
networking nodes become controllable, caching, and CCN-based caching. Based on trace-driven
programmable, and, most important, scalable and evaluation and related analysis, we have demon- strated
elastic for resource utilization adapting to user and con- that the deployment of in-network caching into mobile
tent dynamics. This may handle the capacity needs of networks can potentially help reduce mobile traffic. We
each station and ramp up or scale down node have also presented a number of promising research
capability depending on the mobile traffic dynamics. opportunities and relevant challenges, particularly related
SDN is mostly focusing on the adaptability and to distributed cache resource management, cooperative
controllability of network functions using virtualized caching policy, and content-aware RAN caching. Conclu-
resource management systems. SDN provides a fine- sively, we have highlighted the roles that NFV, eMBMS,
grained packet classifier and flex- ible routing, which can and D2D can play in further improving the gains that can
easily direct a chosen subset of traffic through a set of be acquired from caching in mobile networks.
middle-boxes. Content-aware forwarding and caching ACKNOWLEDGMENT
may then become easier, particularly with the support of Prof. Min Chen’s work was supported by the Youth
deep-packet inspection (DPI) technologies. New caching 1000 Talent Program, China National Natural Science
and delivery mechanisms for an SDN- based 5G mobile Foundation (No. 61300224), China Ministry of
network environment become more critical and Science and Technology (MOST), the International
important, but are still pend- ing. Science and Technol- ogy Collaboration Program
CACHING WITH MULTICAST AND (Project No.: 2014DFT10070), and the Hubei
DEVICE-TO-DEVICE Provincial Key Project (No. 2013CFA051).
Although caching techniques can reduce traffic within a REFERENCES
mobile network, this does not impact the load of X. Wang et al., “A Survey of Green Mobile Networks:
conveying the content via the wire- less link. For Opportunities and Challenges,” ACM/Springer Mobile
effective delivery of the same con- tent to a group of Networks and Applications (MONET), vol. 17, no. 1, Feb.
mobile users, the evolved multimedia broadcast 2012, pp. 4–20.
multicast service (eMBMS) [18] of LTE may be J. Erman et al., “To Cache or Not to Cache — The 3G
envisioned, inte- grated along with necessary caching Case,” IEEE Internet Computing, vol. 15, no. 2, Mar.
techniques in the EPC and RAN. Furthermore, the 2011, pp. 27–34.
wireless link load can be further reduced with the use of S. Woo et al., “Comparison of Caching Strategies in
local caching and sharing techniques such as device-to- Modern Cellular Backhaul Networks,” ACM MobiSys,
device (D2D), a hot topic currently under investigation June, 2013.
within 3GPP 4G LTE- Advanced [19]. Using D2D, B. A. Ramanan et al., “Cacheability Analysis of HTTP
mobile users can operator authorized spectrum for direct traffic in an Operational LTE Network,” Wireless
commu- nication without the support of infrastructure, so Telecommun. Symp., Apr.
users can cache content locally in their own devices.
2013.
N. Golrezaei et al., “FemtoCaching: Wireless Video Students in the IT Field by NIPA from 2008 to 2011,
Con- tent Delivery through Distributed Caching and he also received the Global Outstanding Chinese
Helpers,” IEEE INFOCOM, Mar. 2012. Ph.D. Stu- dent Award in 2012. His current research
F. Le, M. Srivatsa, and A. K. Iyengar, “Byte Caching in interests are social-aware multimedia service in cloud
Wireless Networks,” IEEE ICDCS, June 2012. computing, coop- erative backhaul caching, and traffic
B. Ahlgren et al., “A Survey of Information-Centric offloading in mobile content-centric networks.
Networking,” IEEE Commun. Mag., vol. 50, no. 7, July
2012, pp. 26–36. MIN CHEN [SM’09] ([email protected]) is a
B. Han et al., “AMVS-NDN: Adaptive Mobile Video professor in the School of Computer Science and
Streaming with Offloading and Sharing in Wireless Technology at HUST. He was an assistant professor
Named Data Networking,” IEEE INFOCOM, NOMEN in the School of Computer Sci- ence and Engineering
Wksp., Apr. 2013. at SNU from September 2009 to February 2012. He
Nat’l. Lab. for Applied Network Research, Weekly worked as a post-doctoral fellow in the Department of
Squid HTTP Access Logs, http://www.ircache.net/. Electrical and Computer Engineering at UBC for three
H. Sarkissian, “The Business Case for Caching in 4G years. Before joining UBC, he was a postdoctoral fellow
LTE Networks,” LSI tech. rep., 2012. at SNU for one and a half years. His research focus- es
T. Taleb and A. Ksentini, “Follow Me Cloud: on Internet of Things, machine-to machine communica-
Interwork- ing Federated Clouds and Distributed tions, body area networks, body sensor networks, e-
Mobile Net- works,” IEEE Network, vol. 27, no. 5, healthcare, mobile cloud computing, cloud-assisted
Sept./Oct. 2013, pp. 12–19. mobile computing, ubiquitous networks and services,
Local IP Access and Selected IP Traffic Offload (LIPA- mobile agents, multimedia transmission over wireless
SIPTO), http://www.3gpp.org/ftp/Specs/html- net- works, and so on.
info/23829.htm.
X. Jin et al., “SoftCell: Scalable and Flexible Cellular TARIK TALEB ([email protected]) received his B.E.
Core Network Architecture,” ACM CoNEXT, Dec. 2013. degree (with distinction) in information engineering,
Global Network Operators, “Network Functions and M.Sc. and Ph.D. degrees in information sciences
Virtu- alization: An Introduction, Benefits, Enablers, from GSIS, Tohoku University, Sendai, Japan, in 2001,
Chal- lenges, & Call for Action,” white paper, Oct. 2003, and 2005, respectively. He is currently a senior
2012. researcher and 3GPP standards expert with NEC Europe
S. Ramesh, I. Rhee, and K. Guo, “Multicast with Cache Ltd., Hei- delberg, Germany. Prior to his current
(Mcache): An Adaptive Zero-Delay Video-on-Demand position, until March 2009, he was an assistant
Service,” IEEE Trans. Circuits and Systems for Video professor with the Graduate School of Information
Technology, vol. 11, no. 3, Mar. 2001, pp. 440–56. Sciences, Tohoku Uni- versity. From October 2005 to
K. Doppler et al., “Device-to-Device Communication as March 2006, he was a research fellow with the
an Underlay to LTE-Advanced Networks,” IEEE Com- Intelligent Cosmos Research Institute, Sendai. His
mun. Mag., vol. 47, no. 12, Dec. 2009, pp. 42–49. research interests include architec- tural enhancements
X. Wang et al., “Content Dissemination by Pushing to 3GPP networks, mobile multime- dia streaming,
and Sharing in Mobile Cellular Networks: An wireless networking, intervehicular
Analytical Study,” IEEE MASS, Las Vegas, NV, Oct. communications, satellite and space communications,
2012. congestion control protocols, handoff and mobility man-
X. Wang et al., “TOSS: Traffic Offloading by Social agement, and network security.
Network Service-Based Opportunistic Sharing in Mobile
Social Networks,” IEEE INFOCOM, Toronto, Canada, ADLEN KSENTINI ([email protected]) is an
Apr. 2014. associate pro- fessor at the University of Rennes 1,
BIOGRAPHIES France. He is a mem- ber of the INRIA Rennes team
XIAOFEI WANG ([email protected]) is currently a Dionysos. He received an M.Sc. in telecommunication
post- doctoral research fellow in the Department of and multimedia networking from the University of
Electrical and Computer Engineering at the University of Versailles. He obtained his Ph.D. degree in computer
British Columbia (UBC). He received M.S. and Ph.D science from the University of Cergy- Pontoise in 2005,
degrees from the School of Computer Science and with a dissertation on QoS provisioning in IEEE
Engineering, Seoul National University (SNU) in 2008 802.11-based networks. His other interests include future
and 2013, respectively. He received his B.S. degree Internet networks, cellular networks, green net- works,
from the Department of Com- puter Science and QoS, QoE, and multimedia transmission. He is involved
Technology, Huazhong University of Sci- ence and in several national and European projects on QoS and
Technology (HUST) in 2005. He is a recipient of the QoE support in future wireless and mobile networks.
Korean Government Scholarship for Excellent Foreign He is a co-author of over 60 technical journal and
interna- tional conference papers. He received Best
Paper Awards from IEEE ICC 2012 and ACM
MSWiM 2005. He has been on the Technical Program
Committees of major IEEE Com- Soc conferences,
including ICC/GLOBECOM, ICME, WCNC, and
PIMRC.

VICTOR C. M. LEUNG [F] ([email protected])

received B.A.Sc. (Hons.) and Ph.D. degrees, both in
electrical engineering, from UBC, where he holds the
positions of professor and TELUS Mobility Research
Chair in the Department of Electri- cal and Computer
Engineering. He is a Fellow of the Royal Society of
Canada, Engineering Institute of Canada and the
Canadian Academy of Engineering. He has contributed
more than 700 technical papers, 26 book chapters, and 5
book titles in the areas of wireless networks and mobile
systems. He was a Distinguished Lecturer of the IEEE
Com- munications Society. He has served or is serving
on the edi- torial boards of IEEE Transactions on
Computers, Wireless Communications, Vehicular Technology,
IEEE Wireless Com- munications Letters, and several other
journals, and has contributed to the Organizing and
Technical Program Com- mittees of numerous
conferences. He was a winner of the 2012 UBC Killam
Research Prize and the IEEE Vancouver Section
Centennial Award.