Appl. Math. Inf. Sci.9, No. 2L, 671-680 (2015) 671

Applied Mathematics & Information SciencesAn International Journal

http://dx.doi.org/10.12785/amis/092L43

An Optimization-based Routing Forwarding Algorithm inICNSong Guo∗, Muqing Wu and Qian Hu

Laboratory of Network System Architecture and Convergence, Beijing University of Posts and Telecommunications, Beijing, 100876,China

Received: 10 Aug. 2014, Revised: 11 Nov. 2014, Accepted: 12 Nov. 2014Published online: 1 Apr. 2015

Abstract: In dynamic routing construction of ICN, traditional CCN routing mechanism can improve the diversity and reliability ofdata forwarding. But it also brings problems in retrieval redundancy, taking some load to the network. SoCCeR strategy uses antcolony algorithm to finish single-path routing of CCN in distribution, while it has defects in ant agency control and convergencefeature. Therefore, based on rapid routing this paper proposes a service nodes selecting algorithm with optimized ACO algorithm. Thealgorithm integrates path time-delay and load of service node into optimized selecting algorithm of routers, which makes full used of thevisiting feature of users to improve visiting quality and reduce request loss efficiency. In our method, the probabilistic state forwardingrules are modified and adaptive pheromone updating formulasare introduced, to prevent ant agency to fall into search stagnation, dueto abnormal accumulation of pheromone density in current optimal path. So the routing algorithm is more responsive to the dynamicchanges of network topology.

Keywords: CCN, Pheromone, Content activity, CAACO

1 Introduction

Along with rapid development of resource scale and userquantity on Internet, the network communicationaggravates local flow of network and displays weaknesssuch as load imbalance, low efficiency, etc. In order tochange this situation fundamentally, researchers proposea novel network system framework, Information-centricNetwork (ICN) [1]. It increases copy and storage functionof routing node on resources, so that routing node canalso be a resource provider to change the mode to obtainresources in IP network according to terminal position.Thus, users only concern the resource itself so as to adaptto large-scaled data sharing. The purpose of ICN is todevelop a more effective network framework for contentdistribution, access and sharing. It satisfies clients requestfor resource by content replica and caching. At present,there has appeared many research engineer projects whichmainly include Data-OrientedNetworkArchitecture(DONA), Publish-Subscribe InternetRouting Paradigm(PSIRP)Network ofInformation(NetInf), Content-Centric Networking(CCN),etc [2,3,4,5] The realization structure of these projects isdifferent but their key design ideas are content-centered

network. They all no longer use IP address on forwardingrealization of system framework and promote routingnode to own function of resources storage. In these years,the mode of directly deploying cache in router, that is,internal network cache, has been paid attention by manyresearchers [6]. The initial purpose of design on CCN isthat the functions of network have changed fromterminal-to-terminal of host communication todistribution and acquisition of content.

In CCN, common strategies include full forwardingstrategy(FF) , random forwarding strategy(RF) and theshortest path forwarding(SPF) [7]. In FF, the nodeprovides forwarding interest packet to all correspondinginterfaces of most content sources. However, mostinterest packets will result in generating redundant flow innetwork. RF strategy randomly selects one forwardinginterest packet in corresponding interface of contentsource, but RF cannot acquire effective performance.NCE adopts SPF strategy and it transmits the interestpacket to the content source which has the least hop tocurrent node. NCE method cannot guarantee selecting theoptimal content source since it does not concern linkstatus and node load. In addition, caching content replica

∗ Corresponding author e-mail:guosong@bupt.edu.cn

c© 2015 NSPNatural Sciences Publishing Cor.

672 S. Guo et. al.: An Optimization-based Routing Forwarding

on nodes is frequently changing so the selected contentsource nodes cannot effectively guarantee requestcontent. Reference [8] combines network system ofservice center with network system of content center. Itproposes a service route and service selecting methodSoCCeR (Services over Content-Centric Routing).However, SoCCeR does not consider dynamiccharacteristics and volatility of the content on nodes, so itcannot be effectively applied in processing of CCNreplica nodes. In network route processing, the flowdistribution is constantly changing and network link ornodes will be randomly invalid or rejoined. Self-catalysisand positive feedback mechanism of ACO effectivelymatch the solution feature of these problems, so antcolony intelligence is effectively applied to networkrouting area [9]. Related scholars propose that ant colonyalgorithm is applied in content center network to solve itsproblems in retrieval redundancy in related references in2011, and obtain better effects.

Thus, contraposing to current CCN routingoptimization algorithm, that is, the defect of SoCCeRalgorithm in ant agent control and algorithm convergence,we offer improvement in two aspects: At first, contentactivity is introduced in routing optimization algorithm.Itconsiders the behavior characteristics of users access andanalysis on interest preference, and merges them intooptimized selection algorithm of routing. So it can fullyutilize features of users to improve access quality of usersand reduce failure rate of users access, which alsooptimizes the route selection algorithm and QoS ofnetwork. Second, the probabilistic state transition rule ofant colony algorithm is modified and updating formulasof adaptive pheromone are introduced to avoid thatpheromone density of current better path is constantlyincreasing. It avoids ant agent falls into search stagnancy,so that they have more chances to explore more optimalpaths which has not been excavated. Finally, based onabove improvement, we design an ant colonyoptimization-based service node selecting algorithm, thatis, Content- Activity-Based Ant Colony Optimization(CAACO) algorithm. We specifically discuss theoptimization process and realization module of thisalgorithm. Then the performance of algorithm is verifiedby some simulations. Compared to existing methods, ouralgorithm has fast convergence speed by distributedrealization. It can be effectively applied in theenvironment of dynamic change of CCN content replica.

2 Analysis of Content-Center NetworkRouting Mechanism

CCN proposes a novel network system structure, takinginformation name as routing identity and IP as lower levelnetwork, without conception of ICP layer definition,adding the strategy layer and security layer [10,11]. Asthe next layer of network layer, strategy layer provides

decisions for routing. As the last layer of network layer,security layer provides security for network. In CCN,URL is used to express information name to provide twotypes of packets: Interest and Data. Interest packet isrequest packet and Data packet is returned informationpacket. The workflow of CCN is shown as figure 1. Atfirst, the server with information is flooding to the wholenet. After router receives flooding data, informationrouters are calculated and established. When Client1sends Interest packet, routers will search in cache first. Ifthere is no corresponding information, the informationname according to requirement will be routed to thenearest mytimes server [12]. Then it records Interestgrouping track, provides path for information returningand cached mytimes information at the same time.Information packet returns Client1 according to originalpath. When Client2 also requires interest packet and it isrouted to B. If there is mytimesinformation in B, theinformation will be returned fromB.

Routing mechanism of content center network isbased on Name-Based Routing. It does not need to searchcorresponding object storage position of content markname. It just requires content to be routed tocorresponding storage position to obtain information, andit directly routes request data packet to one or morecontents to offer nodes based on name marker of contentobject [13]. When multi-duplication of one content isuseful, FIB in router may connect between multiplenext-hop port and the same one content name prefix.Then, one interest data packet will be replicated tomultiple C, to trigger repeated returning content datapacket. Although this feature offers effectiveness anddiversity of data forwarding, the repeated interest datapacket may cause huge energy consumption of servicenode. Thus, we believe that CCN node needs to extend itsmechanism and make route optimization to avoidrepeated calling of one request for multiple contents, soas to obtain a distributed, scalable and fault-toleratesystem. Because CCN only routes request packet and datapacket return to original node according to request packetpath. So routing problem of CCN is how to choose anoptimal content providing node for request packet. Basedon above CCN node forwarding model, selecting anoptimal content node equals to that node seeks the bestnext hop port for corresponding content item. That is, FIBof node only connects one next-hop port with this contentitem. Therefore, request packet will be routed to theoptimal content provider along the optimal path, avoidingsearch redundancy effectively.

SoCCeR is an ant colony intelligence-baseddistributed path selection strategy which is applied incontent center network. The main job of this strategy is toexplore different next-hop port path information byforwarding exploration message in the network .Then itfurther discover the best port connected with prefix nameof this content to complete path selection. SoCCeR designis based on ant colony optimization. It is a distributed andprobability-based optimization method. So it is applicable

c© 2015 NSPNatural Sciences Publishing Cor.

Appl. Math. Inf. Sci.9, No. 2L, 671-680 (2015) /www.naturalspublishing.com/Journals.asp 673

Fig. 1: Work flow of CCN

for route selection in CCN. It provides informationretrieval on the some one content object in network andobtains perfect research achievements. The optimizationframework of SoCCeR strategy is described as followingsin detail: each network node periodically sends interestant” at the beginning of each local time window. It takesrandomly selected content item as the destination. Thisinterest ant has one time stack to record forwarding timeof its traversing various nodes. The nodes to generateinterest ants push current forwarding time in stack andsend it to each port which is related to this content item inits FIB. The nodes receiving interest ants will push itslocal time into the stack at first and transfer it to the portof the highest frequency value in pheromone table. In thisway, it continues until this interest ant reachescorresponding nodes of content provider, as described infigure 2. To avoid stagnation, for instance, sustainablestrengthening of current optimal path may possibly resultin congestion and constant decrease of probability valueof other ports. It will lead that all interest ants will betransmitted towards one direction. So nodes will transmitthe interest ants to a random port in a small probability.This probability is the exploration probability and thisensures that interest ants can traverse a path which may

be currently optimal but was unknown before. Repeatingabove process till returned ants traverse all the nodes inpath and their corresponding pheromone of ports areupdated, to finally reach original nodes. Interest ants anddata ants periodically traverse various nodes in thenetwork, but normal CCN data packet is only transferredon the optimal path.

The defect of SoCCeR is that its ant agent controlstrategy is completely random. That is to say, thegeneration of explorative agent in network is completelyrandom. While FIB forwarding table of SoCCeR isupdated for optimization, it randomly takes one reachablecontent item of this node as destination to sendexplorative message. The explorative message willcomplete path optimization with this content item in thisperiod. Such random selecting strategy will lead thatcontent item with higher request rate cannot obtain morefast and effective path optimization. In comparison, thepath optimization of content item with lower request rateis too frequent.

c© 2015 NSPNatural Sciences Publishing Cor.

674 S. Guo et. al.: An Optimization-based Routing Forwarding

Fig. 2: Working principle of SoCCeR ant agents

3 Distributed Service Node Selection Basedon Optimized ACO

3.1 Principal Idea

Basic ant colony optimization algorithm aims to searchthe shortest path. However, in terms of service nodesselection, ants not only need to detect the time-delay inpath but also detect service load and content activity ofthe nodes. Thus, the basic ant colony optimizationalgorithm needs to be modified. Since ant colonyalgorithm needs to explore optimal service nodes, theimproved ant colony algorithm proposed in this paper isapproximate to AntNet [14], which needs two kinds offorward and backward artificial ants. Research shows thatmost dataflow of current internet service is fromforwarding flow of streaming media data. However, 80%of users’ access on streaming media concentrates in 20%of researching content. Obviously, it is essential for thisaccess preference research to improve the performance ofour optimization algorithm. Specifically, CCN nodes willgenerate a forward ant at the beginning of each timewindow. The forward ants explore path information withthe purpose of one reachable content item on this node. Itrecords the number of path hop whose initial value is zeroand the load of content providers.

Destination selection of the forward ants is related tothe activity distribution of different content items. It isproposed that there areN content providing nodes andeach node can provide one content item. Reachablecontent item quantity of current nodes isn. Among allNcontent items, contents has the highest activity and itsprobability of users’ access isfs. Correspondingly,forward ant selects content itemS generated by currentnodes as destination with probabilityfs. The higheractivity of content item is, the more explorative message

from nodes sending to offering nodes are, and morefrequent update will occur on the paths providing nodes.Different paths of nodes owing higher activity will obtainmore frequent update and they will promote the optimalpath of practical selection on real data packet in networkto be more approximate to the absolute optimal path. Thereason is the pheromone density of links with linkdeficiency, congestion or other situation will volatile in afast probability. In contrast, pheromone density on anoptimal path with optimal network situation will beincreased more quickly and efficiently. Therefore, bycorresponding control with content activity distributionasguidance, the performance of routing optimizationalgorithm can be effectively improved.

Besides original forwarding information table FIB,CCN node increases status information table. Consideringrandom CCN nodev and recording tv as statusinformation table ofv. For any contenti in tabletv, statusinformation table contains corresponding forwardinginterface ofFv

i of this content and pheromone value aswell as forwarding probability of each interface, as shownas figure 3.

τvi, j (d), τv

i, j (l)andvi, j(α) respectively denote the path

time delay of transmitting interfacej, load of servicenode and the normalized pheromone corresponding tocontent activity. ∀ j ∈ Fv

i , ∑j∈Fv

i, j

τvi, j(x) = 1, ∀i ∈ F ,

x ∈ {l ,d,a}. Pvi, j is forwarding probability which shows

the quality factor of j relative to other interfaces. It iscalculated by weighted calculation of above threepheromone. α, β , γ are weight parameters andα +β + γ = 1.

Pvi, j =

ατvi, j (d)+β τv

i, j(l)+ γvi, j(α)

∑j∈Fv

i, j

(ατvi, j (d)+β τv

i, j(l)+ γvi, j(α))

(1)

c© 2015 NSPNatural Sciences Publishing Cor.

Appl. Math. Inf. Sci.9, No. 2L, 671-680 (2015) /www.naturalspublishing.com/Journals.asp 675

Fig. 3: Design of CCN nodes

Fig. 4: Accelerating ant colony diagram

CCN nodes will periodically generate forward antscontaining some content name. The forward ants arerandomly transmitted to each interface to the next hopnode, according to each interface probability in statusinformation table. After repeated hop forwarding theyfinally reach the service node of content. The servicenode generates backward ants. Backward ants updatepheromone value of each link on path during the processof generating nodes, from service nodes to forward ants.Routing node will update the forwarding probability ofeach interface with the pheromone.

3.2 Adaptive Pheromone Updating Strategy

Specifically, when mediate node in network receives aforward ant, it will select a next hop node based oncurrent pheromone density in different paths and make itroute towards destination node. Meanwhile, the returnedbackward ant will update various nodes pheromonedensity on path, based on path quality, that is, length ofpath time-delay and node load situation. Further itupdates selection probability value on different paths by

updating pheromone density. From this we know that theupdating formulas of pheromone density are essential forprobability status transformation rules of ACO. Wedesign an adaptive pheromone updating formula, which isshown as following:

τi(x) = τi(x)+∆x,∆x =−n(x−1)exp(x)(1− τi(x))/gen(2)

In this formula, ∆x denotes the pheromone. Whenreturned backward ants update corresponding pheromonedensity of nodes, they will calculate increment∆x basedon adaptive pheromone updating formula. Based onabove formula, on one hand, the pheromone incrementwhich backward ants carry appears inverse relation of veilindex. That is, the larger node load of path hop counts andcontent are, the smaller pheromone increment is; On theother hand, it is related to iteration times in currentexperiment. The larger iteration times are, the smallerpheromone increment is. This avoids falling into searchstagnation of exploration agent and it has moreopportunities to explore better paths which are notexcavated. Particularly in unexpected situations, for

c© 2015 NSPNatural Sciences Publishing Cor.

676 S. Guo et. al.: An Optimization-based Routing Forwarding

instance, when current optimal next hop interfacesuddenly fails, the exploration agent will react morequickly to find out another optimal interface. It will notlet the pheromone value of failed interface be higher topromote CCN node to still select failed Fast-Path totransmit packets.

When backward ants update pheromone density ofnodes, they will further update the selected probabilityvalue of corresponding path according to probabilitystatus transferring formula. Specific formula is shown as:

P=i

[(1−α)τ(i d)+ατ(i l)]N∑

i=1[(1−α)τ(i d)+ατ(i l)

(3)

At the same time, nodei performs pheromoneevaporation for the interfaces exceptj in Fv

i , that is,pheromone weakening:

τvi, j (x) = (1−ρ)τv

i, j(x), j ′ ∈ Fvi , j ′ 6= j,x∈ {dv

i, j , lvi, j ,a

vi, j}(4)

If the node does not receive backward ant in timewindow δt , it performs pheromone evaporation to all theinterfaces.

When explorative agent carrying path information isupdating the pheromone of interface, the pheromoneincrement of interface is not only related to carryingpheromone of backward ants, according to adaptivepheromone updating formula, but it also relates topheromone value of current interface as well as iterationnumber of current experiment. So it can avoid falling insearch stagnation with increasing experimental iterationnumber. In addition, while network status is changing, itcan find optimal path in time.

3.3 Backoff Mechanism and ACO Acceleration

Since content in cache of CCN node are dynamic andvaporable, the route of these caching nodes cannotguarantee accuracy of 100%. Therefore, under conditionsthat node caches are missing, backoff mechanism issignificant. Probability forwarding can retransmit otherservice nodes to obtain request content, when a servicenode is missed. Reference [15] notifies the cachingmissing by sending feedback packet of content to originalnode of request packet. The source node of interest packetwill reselect service node to forwarding. This paperinherits this kind of mechanism to solve caching missingof service nodes, and make an improvement of backoffmechanism to accelerate ACO.

Figure 4 refers to the schematic diagram if feedbackpacket to accelerate ACO. Content request packets aresent to service nodeD from node A. Node D checkscontent caching table: If cache is missing, feedbackpacket will be sent along the inverse path of requestpacket. Feedback packet is approximate to reverse proxy

in ACO which contains content activity and its value isset as 0. When it passes one node, it updates pheromonevalue of corresponding content activity in this node statusinformation table, so as to reduce forwarding probabilityof corresponding interface. The content feedback packetis transmitted from serviceD to A. On one hand, backoffmechanism is established to weaken pheromone of cachemissing path; on the other hand, it reduces the probabilityof forwarding request packet to service nodeD andaccelerates the process of ACO.

Based on above improvement, the flow of CAACOrouting optimization is shown as figure 5. The processesare classified into three modules: exploring managementmodule, exploring agent path-finding module andexploring agent path updating module. The nodes innetwork periodically send exploring agent while usersrequest packet and returned the packet which is onlytransmitted in current optimal path. When solving theproblems in search redundancy, QoS of network is alsoimproved.

4 Simulations

We will study the performance of CAACO in CCN infollowing simulations. The experiment adopts GT-ITM togenerate CCN network structure and MATLAB forsimulation. The parameters in experiments are: There are30 nodes in the network and the connection probabilitybetween any two nodes is 0.3; Link bandwidth is100Mbps; In the edge nodes, we randomly select onenode as content source server to publish original content;5 nodes are selected as content agent, that is, servicenodes, to be in charge of publishing and service ofreplica; the other nodes are general CCN routing nodesdirectly connected with users. The cache replacingstrategy is LRU. The number of content in network is2000 and the content data block obey uniform distributionbetween 128KB and 512KB; User request obeys Poisondistribution with meansλ = 8 and the duration timeobeys negative index distribution with means 50s.

4.1 Content Request Efficiency and Time Delay

In the simulation process of figure 6, it describesstatistical results of request failure rate of CAACO andSoCCeR algorithms. In this figure, abscissa denotesiteration times and ordinate denotes failure quantity ofinterest data packet of No.41 node in each iterationprocess, that is, the interest data packet that does notsuccessfully reach content to offer acquired content. Westudy item s of that request from NO.41 to providecontent by NO.65. The blue line refers to SoCCeRalgorithm while red line refers to CAACO algorithm.From this figure we can see, in most cases, requestefficiency of CAACO algorithm is lower than SoCCeR

c© 2015 NSPNatural Sciences Publishing Cor.

Appl. Math. Inf. Sci.9, No. 2L, 671-680 (2015) /www.naturalspublishing.com/Journals.asp 677

Fig. 5: Routing optimization flow of CAACO

algorithm. By observation of failure cased, the reason offailure is path looping. However, in CAACO algorithm,exploring agent distribution will explore path informationwhich goes towards node of content provider with highactivity in a larger probability, according to contentactivity distribution. So it can find loops of path leading tothese content providing nodes in time. On the pathsproducing loops, exploring agent can not reachdestination node within the largest hop count. It willreduce pheromone value of various nodes on this path byevaporation operation. Thus, CAACO algorithm can findpath loop of content providing nodes in higher activity in

time, and reduce practical efficiency of most users, whichbetter meet the demand for content.

Figure 7 compares average request time delay ofalgorithms. During simulation, the completedaccumulating request is used to calculate average timedelay. We select three common methods in CCN: FF, RFand SPF in this experiment. It demonstrates that theaverage time delay of request gradually increases withincreasing reaching request quantity. When requestnumber is 410, time delay tends to be stable. The requestarrival results in reducing system performance duringinitiative stage of simulation. When request times reach

c© 2015 NSPNatural Sciences Publishing Cor.

678 S. Guo et. al.: An Optimization-based Routing Forwarding

Fig. 6: Simulation of request failure efficiency

410, the existing request service completes and theresource is released, so as to provide service for newrequest and the system tends to be stable in the wholeperformance.

Fig. 7: Average time delay of content request

Since FF forwarding strategy will transmit interestpacket to all content service node, it results in largeincrease of load for various nodes and its performance isworse. The average time delay of SPF is small at initialstage of simulation. This is because the hop count mainlyaffects the performance at initial stage. With increasingload of service node, the processing time delay starts toaffect the performance. In comparison with other threestrategies, the load of CAACO algorithm among variousservice nodes is balanced so it has the smallest averagetime delay. When simulation tends to be stable, there isnearly 4% reduction of CAACO algorithm in comparison

with SPF strategy, in average time delay of contentrequest.

4.2 Convergence of Forward Ants ProducingModules

In figure 8 and 9, when algorithm starts, the node selectsnode No.33 as the next hop node to destination node withprobability about 0.8. When algorithm iterates 25 times,node No. 33 fails. We can see that the selectionprobability of No.33 rapidly decreases. However, for thecurrent optimal port, that is, the selection probability ofNo.42 rapidly increases to become the next hop port ofinterest data packet forwarding. However, in convergencefigure of SoCCeR algorithm, the selection probability ofNo.33 does not rapidly decrease after its effectiveness.Similarly, the selection probability of No.42 is not stableto become the optimal port. During iteration, there alsoappears the case that the selection probability of failedport 33 and port 42 is approximately equal. This easilycauses misjudgment of interest data packet to result inrequest failure. Based on above discussion, theconvergence of CAACO is also superior to SoCCeR.

Fig. 8: Simulation of convergence results for SoCCeR

4.3 Load Simulation of Content Source Server

If less source server receives interest packet, it indicatesmore requests obtain content from agent service node toreduce the load of source server. The receiving contentrequest of source server, that is, the number of interestpackets is taken as load indicator for simulation. Thereare totally 5 times and their average value is taken as thefinal result. Figure 10 shows the comparison of four

c© 2015 NSPNatural Sciences Publishing Cor.

Appl. Math. Inf. Sci.9, No. 2L, 671-680 (2015) /www.naturalspublishing.com/Journals.asp 679

Fig. 9: Simulation of convergence results for CAACO

algorithms on load performance of content source servers.We can see that the source server is the heaviest in FF,because if transmits the interest packets to all servicenodes. RF is approximate to SPF on performance: theinterest packets are sent to single service node to reduceload of source server. However, received interest packetsof CAACO source server is the least because thealgorithm accumulates content request to make morerequests service for agent node, which further reduces theload of source server. Compared to SPF, the source serverload reduces about 41%.

Fig. 10: Load of content source servers

5 Conclusion and Future Work

In the solution of CCN routing optimization, differentcontent activities in network is introduced to analyze andcontrol ants agent with distribution of content activity. In

specific, for content items with higher activity, thereexists higher probability for nodes to choose them asdestination of searching path of ants agent. To getdifferent paths of node with higher activity, it willpromote the optimal path of practical selection on realdata packet in network, which is more approximate toabsolute optimal path. Therefore, by correspondingcontrol with the content activity distribution taken by antagent as instruct, the routing optimization algorithm canbe effectively improved. Secondly, by the modification ofcalculating formulas of ant current colony algorithms andintroduction of adaptive pheromone updating formula, theconvergence of existing ACO is improved. Simulationexperiment shows that CAACO has 4% reduction inaverage time delay, for the shortest strategies taking theleast number of hops as server node selection. The loadon source server is also reduced about 41%.

The distribution law of different content activity incontent center network remains accurate assessment. Asis discussed in the paper, current academia has onlyqualitative analysis on content activity in network andthere are few analyses in quantitative assessment.However, CCN is still developing and there is not anylarge-scale implementation environment which alsobrings difficulty to provide quantitative assessment oncontent activity in real network environment. In addition,since there are many defects of ACO itself, in furtherresearch, we will focus on the content activity distributionin real network environment and combine ACO withother intelligence optimization algorithms, such asgenetic algorithm, to find better routing solutions.

Acknowledgements

This work was supported by the National Science andTechnology Major Projects of China under Grant No.2011ZX03001-007-03.

References

[1] Nelson T Literary, machines: the report on, and of, projectXanadu concerning word processing, electronic publishing,hypertext, thinker toys, tomorrows intellectual revolution,and certain other topics including knowledge, education andfreedom, Sausalito, California: Mindful Press, 1981.

[2] Koponen T, Chawla M, Chun B-G, et al, A data-orientednetwork architecture, Proceedings of the Conference onApplications, Technologies, Architectures, and Protocols forComputer Communications, New York, NY, USA: ACM,2007: 181-192.

[3] Rhea S, Godfrey B, Karp B, et al, ”OpenDHT: a publicDHT service and its uses”, ACM SIGCOMM ComputerCommunication Review, 2005,35, 73-84.

[4] Eugster P T, Felber P A, Guerraoui R, et al, ”The many facesof publish/subscribe”, ACM Computing Surveys, 2003,35,114-131.

c© 2015 NSPNatural Sciences Publishing Cor.

680 S. Guo et. al.: An Optimization-based Routing Forwarding

[5] XIA Chunmei, XU Mingwei, ”Survey of information-centricnetworking, Journal of Frontiers of Computer Science andTechnology, 2013,7, 481-493.

[6] Tolga Bekt, Jean-Franois Cordeau, Erhan Erkut, ”Exactalgorithms for the joint object placement and request routingproblem in content distribution networks”, Computers &Operations Research, 2008,35, 3860-3884.

[7] Severance Charles,”Van Jacobson: Content-centricnetworking” , Computer, 2013,46, 11-13.

[8] Shanbhag S, Schwan N, Rimac I, et al, ”SoCCeR: servicesover content-centric routing, ACM SIGCOMM Workshop onInformation-Centric Networking, Toronto, Canada, 2011:62-67.

[9] Li Chengming, Okamura Kouji, Liu Wenjing, ”Ant colonybased forwarding method for content-centric networking”,Proceedings of 27th International Conference on AdvancedInformation Networking and Applications Workshops,WAINA, 2013, pp.306-311.

[10] Almeida J, Eager D, Vernon M, et al, ”Minimizing deliverycost in scalable streaming content distribution systems”,IEEE Transactions on Multimedia, 2004,6, 356-365.

[11] Li Jun, Liu Bin, Wu Hao, ”Energy-efficient in-network caching for content-centric networking”, IEEECommunications Letters, 2013,17, 797-800.

[12] Wang Kai, Chen Jia, Zhou Huachun, ”Content-centricnetworking: Effect of content caching on mitigating DoSattack”, International Journal of Computer Science Issues,2012,9, 43-52.

[13] Wang Y, He K, Dai Het al, ”Scalable name lookup in NDNusing effective name component encoding, Proceedings ofIEEE 32nd International Conference on IEEE, Tokyo, 2012:688-697.

[14] Gao Qiang, Wang Weiming, Zhou Jingjing, ”Researchprogress for the interest control protocol of content-centricnetworking”, Sensors and Transducers, 2013,154, 87-93.

[15] Lee Jihoon, Kim Daeyoub, ”Proxy-assisted content sharingusing CCN for resource-limited mobile consumer devices”,IEEE Transactions on Consumer Electronics, 2011,57, 477-483.

Guo Song, was bornin Jilin, China, in 1987.9.He received his Bachelor’sdegree in telecommunicationengineering from BeijingUniversity of Posts andTelecommunications,P. R. China, in 2010.Now he is doing research asa Ph.D. candidate in Beijing

University of Posts and Telecommunications, P. R. China.His current research interest includes content delivery androuting technology of information-centric network.

Wu Muqing, was born inBeijing, China, in 1693.7.Hereceived his Ph.D. degreein Beijing University of Postsand Telecommunications.He is a professor at BeijingUniversity of Posts andTelecommunications, isa senior member of the Chinainstitute of communications.

His current research interest includes future networks, AdHoc wireless networks, high-speed network traffic controland performance analysis, GPS locating and services, andteaching basic.

Hu Qian, was bornin Hebei, China, in 1988.1.He received his Bachelor’sdegree in telecommunicationengineering from XiDianUniversity, P.R. China,in 2010. Now he is doingresearch as a Ph.D. candidatein Beijing University of Postsand Telecommunications,P.R. China. His current

research interest includes caching and routing technologyof information-centric network. theories of telecomnetworks.

c© 2015 NSPNatural Sciences Publishing Cor.