HCOR: a high-throughput coding-aware opportunistic routing for inter-flow network coding in wireless mesh networks

Transcription

1 Ha et al. EURASIP Journal on Wreless Communcatonsand Networkng 2014, 2014:148 RESEARCH Open Access HCOR: a hgh-throughput codng-aware opportunstc routng for nter-flow network codng n wreless mesh networks Long Ha, Hongyu Wang *, Je Wang and Zhenzhou Tang Abstract Codng-aware routng s an effectve approach to create more codng opportuntes n nter-flow network codng. To the best of our knowledge, most of the codng-aware routng schemes focus on maxmzng the codng opportuntes. However, for opportunstc transmsson, the throughput s not always ncreased wth the ncrease of codng opportuntes. In ths paper, we explore why ths case wll happen and how to measure the benefts of network codng n the opportunstc routng. Accordng to the above conclusons, we propose a novel hgh-throughput codng-aware opportunstc routng (HCOR) to acheve the maxmal throughput gan n wreless mesh networks. HCOR s based on anypath routng and takes advantage of the network codng gan to fnd out the route wth mnmal anypath cost reasonably. Meanwhle, t s also a multhop network codng and changes the route wth dynamcal data loads adaptvely. Smulaton results demonstrate that HCOR has better performance than codng opportunty-aware routng and also obtans a sgnfcant throughput gan n wreless mesh networks. Keywords: Inter-flow network codng; Opportunstc routng; Codng-aware routng; Multhop network codng 1 Introducton Network codng (NC) s ganng popularty as a new transmsson method that can mprove network performance n terms of ncreasng throughput at the codng level. NC was frst proposed by Ahlswede et al. [1] n 2000, and s bascally classfed nto two types. One s called ntra-flow network codng (IANC) whch encodes multple packets belongng to the same flow. The other s called nter-flow network codng (IRNC) whch encodes multple packets from dfferent flows. IANC s a knd of relable transmsson method. For n natve packets sent from one flow, the sources and ntermedate nodes are allowed to encode these packets together before sendng them to the destnatons. The generated encoded packets are redundant aganst lossy lnks untl the destnatonrecevesand decodesn ndependent encoded packets for recoverng all natve packets. By dong ths, packet losses are masked and data transmssons are robust [2]. *Correspondence: whyu@dlut.edu.cn Faculty of Electronc Informaton and Electrcal Engneerng, DUT, Lnggong Road, Dalan , Chna Dfferent from IANC, IRNC s an effcent transmsson method. Intermedate nodes encode the natve packets from dfferent flows and broadcast the encoded packets to dfferent destnatons. By explotng the broadcast nature of wreless channel, destnatons can overhear some of sde nformaton of the encoded packets before decodng them. So the codng nodes can broadcast the encoded packets to dfferent destnatons for dfferent flows at thesametme.bymaknguseofthsfree-rde transmsson [3], IRNC saves many tme slots to send dfferent flows at ntermedate nodes smultaneously. In ths paper, weonlyfocusonanddscussirnc.sotheterm network codng whch appears n the followng only means IRNC. Much work ponts out that the benefts of network codng are dfferent for the same flow by usng dfferent routes. So the queston of how to fnd out the best route wth network codng s stll open untl now. Most of the codng-aware routng schemes are based on determnstc routng protocols [3,4]. Due to the characterstc of network codng, these schemes cannot work well when the data flows change frequently. The reason s that route 2014 Ha et al.; lcensee Sprnger. Ths s an Open Access artcle dstrbuted under the terms of the Creatve Commons Attrbuton Lcense ( whch permts unrestrcted use, dstrbuton, and reproducton n any medum, provded the orgnal work s properly credted.

2 Ha et al. EURASIP Journal on Wreless CommuncatonsandNetworkng 2014, 2014:148 Page 2 of 13 must be preprogrammed n determnstc routng protocols. However, network codng depends on the flows and routes smultaneously. So dong codng-aware routng n determnstc routng protocols must bnd flows and routes together. Ths causes low stablty of routes when the state of flows (such as on or off, or the rate of flow) s changed frequently. For example, assume that the best NC-aware route s R a for flow a to be encoded wth flow b. Whenflowb s end, R a may not be the best route for a. The follow-on problems are that hgh delay wll be caused f a does the reroutng at ths tme or thethroughputwllbedegradedfa stll uses R a as ts route. Network codng-aware opportunstc routng (NCOR) has been proposed recently [5-8]. Unlke classcal determnstc routng, opportunstc routng (OR) always fnds potentally feasble paths as the route. In OR, every packet s forwarded by a forwardng lst whch s composed of the neghbors closer to the destnaton. Hence, unstable routng problem can be solved n opportunstc routng. Meanwhle, due to the forwardng lst adopted, whch means many nexthops can be consdered n a codng structure, more codng opportuntes may be found n an opportunstc transmsson. Conventonal NCOR schemes [5,8] only consder the maxmum codng opportuntes as the selecton mechansm for the codng node. For example, the authors n [5] propose that they use the number of the neghbor nodes that can decode a codng pattern as the codng gans of that codng pattern. In IRNC, the maxmum number of the neghbor nodes that can be the decodng nodes suggests the maxmum codng opportuntes. The authors n [8] clam that f a node wth the most codng opportuntes can be chosen as the forwarder n each packet transmsson, the throughput of network wll be greatly mproved as a result. From our study, however, ths mechansm s not very sutable for ncorporatng NC nto OR. When multple packets are encoded together, the recevers of ths encoded packet should be determned. After that, the set of new recevers may be possbly shrunken from the set of orgnal recevers. As we know, the shrnkage of a forwardng lst wll reduce the transmsson gan n an opportunstc transmsson. So n ths case, s the network codng stll benefcal for ths transmsson? How should we decde whch s the best choce, codng or not? In addton, f we consder the gan of opportunstc transmsson n dong network codng, how to determne the forwarder lst s complcated. In ths paper, amng to address the above problems, we study the network codng n anypath routng [9] whch provdes a calculaton method to compute the gan of opportunstc transmsson. Our man contrbutons are summarzed as follows: Accordng to the defnton and codng condton of IRNC, we fgure out two followng solutons to help desgn a codng-aware opportunstc routng. The frst s whch nodes should be n the forwardng lst for sendng an encoded packet. The second s how to measure the prce of network codng n an opportunstc transmsson. We propose a novel codng-aware opportunstc routng scheme called HCOR, whch s based on anypath routng and fnds out the mnmal anypath cost path as the route wth network codng. Compared wth conventonal NCOR schemes, HCOR s more feasble on choosng the codng opportuntes. Meanwhle, HCOR s desgned on a multhop network codng structure, n whch much more codng opportuntes can be collected than wth a local codng structure [3]. We mplement the HCOR scheme n ns-2 and carry out extensve evaluaton to show the performance of HCOR. The remander of ths paper s organzed as follows. Secton 2 revews some related work. Secton 3 s the overvew of HCOR. In Secton 4, we show some basc knowledge and key technologes used n ths paper. In Secton 5, we dscuss the characterstc of ncorporatng NC nto opportunstc routng and gve a computng method to calculate the cost of network codng. We present the detals of HCOR mplementaton n Secton 6. The smulatons performed to evaluate the performance of HCOR are dscussed n Secton 7. Fnally, we conclude ths paper n Secton 8. 2 Related work Ahlswede n [1] presented that the network capacty can be ncreased sgnfcantly by employng network codng. Snce then, many varous studes have proved the benefts of network codng [10,11] and gven many codng methods to acheve the gan of network codng for dfferent networks [12-17]. For nter-flow network codng, COPE [18] was proposed n 2008 as the frst practcal XOR (exclusve or)-based network codng scheme. Authors n [11] showed the codng opportuntes of COPE n a multhop wreless network. For practcal wreless network codng, authors of [19] gave the bound of encodng number and the gan of throughput n a general class system of random access mechansms. In lossy networks, the performance of nter-flow network codng a has been wdely studed. The work on XOR-based IRNC wasproposedn[20].wustudedthecaseofrandom lnear network codng-based IRNC wth the number of sessons M 3 n [21]. Wang extended Wu s study to the case wth M > 3 n [22]. Amermehr and Ashtan

3 Ha et al. EURASIP Journal on Wreless Communcatonsand Networkng 2014, 2014:148 Page 3 of 13 analyzed the trade-off between delay and throughput for opportunstc network codng n a two-way relay network [23]. Yang et al. gave a study of bnary multuser NC to mprove the symbol error probablty n a N-way relay network [24]. Codng-aware routng has been regarded as an effectve approach to actvely create more codng opportuntes [3-5,25-27]. In [25] and [27], authors proposed codngaware routng methods based on COPE. Ther methods detect codng opportuntes from a local codng structure. In [3], authors proposed a codng-aware routng method called dstrbuted codng-aware routng (DCAR) based on (dynamc source routng) DSR [28] routng protocol. DCAR works n more general cases than the COPEbased method. The codng opportuntes are detected n the routng process. So more codng opportuntes can be detected from multple codng structure. Free-rdeorented routng metrc (FORM) [4] s a varaton of DCAR and also works on multple codng structure. Most of the above codng-aware routng schemes are focused on the determnstc routng protocols. Recently, network codng-aware opportunstc routng schemes were proposed to acheve more throughput n wreless networks [5,8,29-31]. Authors of CORE [5] proposed an NCOR method by combnng hop-by-hop opportunstc forwardng and localzed nter-flow network codng. Authors of [8] proposed a practcal NCOR scheme for wreless mesh networks. Authors of [31] gave a mult-rate approach to realze NCOR n wreless mesh networks. Authors of [29] proposed a relable multcast protocol based on NCOR to acheve hgh throughput and farness n lossy wrelessnetworks.in[32],authorsgaveageneralsurveyof codng-aware routng n wreless networks. Compared wth these NCOR schemes, HCOR consders the network codng cost n an opportunstc transmsson and does codng-aware opportunstc routng based on anypath cost. 3 OvervewofHCOR HCOR s a codng-aware opportunstc routng method. It manly works between Internet Protocol (IP) and meda access control (MAC) layers and nserts a codng layer to do the calculaton of codng gan and serve the sendng and recevng of encoded packets. HCOR adopts the XORbased codng method. The foundaton of HCOR s a dstrbuted system based on anypath routng. As llustrated n Fgure 1, HCOR ams to solve whether the network codng should be done and whch nodes should be used as the codng nodes to acheve the maxmal throughput gan. To realze HCOR, every node wll calculate a temporary anypath cost ndependently for each encoded flow. Ths temporary anypath cost s survvng on ths flow and marks ths node wth ths cost. All flows always choose the lowest anypath cost path as ther transmsson route. Each Fgure 1 Conceptual vew of HCOR. Aflowf SD s generated from S to D. Assumng a multhop codng structure, node C s a codng node and encodes f SD nto other flows. The nodes wth red color are other potental codng nodes for f SD. The encoded flow s denoted by f. The blue crcle presents the forwardng lst of C to D. The red crcle presents the decodng set of f to acheve f SD. The decodng set may be smaller than the forwardng lst. So the anypath cost of forwardng f SD by the decodng set may be more expensve than that by the forwardng lst. The red dashed arrow presents dong network codng. The blue dashed arrow presents forwardng f SD wthout network codng. potental encoded packet uses opportunstc codng, n whch a node ams to acheve the maxmal throughput but not maxmze the number of network codng n a sngle transmsson. HCOR s also a multhop network codng scheme. That means HCOR has the ablty of detectng codng opportuntes from a multhop network topology. The multhop codng opportunty detecton s an extenson of local overheard nformaton detecton. HCOR uses learnng neghbor state whch s a guessng method used n COPE for local detecton and transmts ths neghbor state on the path. So we need a neghbor state mantenance mechansm whch s very common n ad hoc routng protocols. If a node makes an ncorrect guess occasonally, whch possbly causes the encoded packet to be undecodable at decodng nodes, the relevant natve packet should be retransmtted to help decode ths encoded packet. To make HCOR easy, we leverage routng pre-computaton to compute the delvery probablty between every par of nodes and use t to dentfy lnk status. Routng precomputaton works before anypath routng. That means the status of all lnks s known when we begn to do the anypath routng. Before delvng nto detals, we defne the terms used n the rest of the paper and lst them n Table 1. 4 Network codng and anypath cost 4.1 Opportunstc transmsson and anypath cost In opportunstc transmsson, a packet sent from one nodemayberecevedpossblybyanyoftsneghbors.

4 Ha et al. EURASIP Journal on Wreless CommuncatonsandNetworkng 2014, 2014:148 Page 4 of 13 Table 1 Defntons of terms used n ths paper Term Natve packet (flow) Encoded packet (flow) Vrtual natve packet (flow) Overheard nformaton Codng node Decodng node Forwardng lsts of an encoded packet Packet ID Output queue Packet pool Defnton A non-encoded packet (flow) A packet (flow) that s the XOR of multple natve packets (flows) The natve packet (flow) that s XORed n an encoded packet (flow) The set of nodes whch have overheard the natve packet (flow) The node whch encodes natve packets together The node whch decodes the encoded packets The set of forwardng lsts for the vrtual natve packets A 32-bt hash of the packet s IP source address and IP sequence number A FIFO queue at each node to buffer the packets that need to be sent A buffer where a node stores all packets sent and overheard n the past T seconds Theforwardersarethenodesntheseneghborswhch have receved ths packet and are closer to the destnaton. A forwarder has hgher prorty to do the forwardng f t s closer to the destnaton. A forwardng lst (or say a forwardng set) s the set of forwarders. Apparently, more forwarders whch a forwardng lst has cause a hgher probablty of forwardng a packet successfully. In anypath routng [9], authors desgn a calculaton method of anypath cost to present ths relatonshp. They defne an anypath cost C p for forwardng packet p at node by C p = c p J + Cp J, (1) where c p J s the broadcast cost from to a forwardng lst J, andc p J s the anypath cost of J. Accordng to the ETX metrc [33] and assumng that the loss for each lnk s ndependent, cost c p J can be calculated by J = 1 1 ( ), (2) j J 1 dj c p where d j s the packet delvery rato (PDR) from to j.the remanng anypath cost C p J s defned as a weghted average of the costs of the nodes n forwardng set J by C p J = j J w j C p j, (3) where the weght w j s the probablty of node j beng the relayng node of p from node and j J w j = 1. Let J = {1, 2,..., n} wth anypath cost C p 1 Cp 2... Cp n,the weght w j s smplfed to w j = d j 1 j k=1 (1 d k) 1 ( ). (4) j J 1 dj Wth the above settngs, some conclusons can be acheved n the anypath cost. Lemma 4.1. For a fxed transmsson rate, let C J be the cost b of a node va forwardng set J and set C,J be the cost va forwardng set J = J {k}. WehaveC J C J f and only f C C k. Lemma 4.1 s the key concluson n anypath routng. Wth ths lemma, f we want to reduce the cost of a forwardng set, we need to add a new forwarder whch has lower cost nto the forwardng set. In other words, the cost of a forwardng set s optmal f and only f there s no node n the neghbors wth lower cost than that of ths sender. Lemma 4.2. For a fxed transmsson rate, assume a forwardng set {1, 2,..., k} wth cost C 1 C 2... C k.if C ( j) s the cost from node va forwardng set { 1, 2,..., j }, for 1 j k, then we always have C (1) C (2)... C (k) = δ,whereδ s the optmal cost of node. Lemma 4.2 shows the varaton tendency of anypath cost wth varyng number of forwarders. It s the extenson of Lemma 4.1 to demonstrate that the reducton of the forwarders n a forwardng lst wll cause the ncrease of the anypath cost of ths opportunstc transmsson. 4.2 Inter-flow network codng Before dscussng the ssues of IRNC n opportunstc routng, we would lke to gve a defnton of IRNC formally. For a network codng structure, w.l.o.g., we assume that there s only one codng node and multple decodng nodes. All potental encoded flows should be encoded at the codng node and decoded at the decodng nodes.

5 Ha et al. EURASIP Journal on Wreless Communcatonsand Networkng 2014, 2014:148 Page 5 of 13 Consderng a multhop wreless network, a flow runs possbly on a multhop path whch starts at a source and ends at a destnaton. We denote a flow from source to destnaton j by f j.ifanatveflowf j s encoded nto an encoded flow, we mark t as a vrtual natve flow by f j.a path P j means a set of lnks connectng node to node j. WesayaflowonapathP j f and only f ths flow passes the nodes and j. Accordng to the denotaton of a flow, we dstngush a flow on a path by the start and end of ths path. For example, gven two flows f ab and f cd,ff ab s on the path P j,wesayf cd s dstngushable from f ab on P j f and only f f cd s not on P j.basedon ths settng of path dstncton, we gve the defnton of IRNC by Defnton 4.1. For path dentfable flows, we say a network codng s nter-flow network codng f and only f all vrtual natve flows of one encoded flow are dstngushable on the paths from codng node to the decodng nodes. Defnton 4.1 can be used as the rule to dstngush IANC and IRNC. Based on the defnton of IRNC, we can acheve two followng lemmas. Lemma 4.3. In nter-flow network codng, for each encoded flow, the decodng nodes for achevng dfferent natve flows must be dfferent. Proof. Ths lemma s easy to be proved by Defnton 4.1. If two vrtual natve flows have the same decodng node, they cannot be dstngushed on the path from the codng node to decodng node. Lemma 4.4. In nter-flow network codng, for each encoded flow, no node s both the decodng node of one vrtual natve flow and also the forwardng node of another vrtual natve flow. Proof. If node m s the decodng node of a vrtual natve flow f from the encoded flow f f j, ths suggests m has overheard f j.ifm s also the forwardng node of the encoded flow for the other decodng node to obtan f j, t mples m s the potental decodng node of f j.ths contradcts Lemma General codng condton Consderng an ntermedate node k on the path P j of flow ( f j ), we denote the upstream part of P j at k by ( U k ) fj and the correspondng downstream part by D k fj. Before dong network codng, the codng node must guarantee that the encoded flow s decodable. That means there s at least one node n the downstream whch can decode ths encoded flow. For dong ths, the codng node must know the decodng knowledge before makng any codng decson. In nter-flow network codng, decodng knowledge s the flow s overheard nformaton whch s defned by a set of possble nodes whch had or overheard ths flow. Ths nformaton can be acheved by learnng neghbor state method. We denote the overheard nformaton ( of flow f before node by O ( f ) = j U ( f ) Nj {j} ),wheren j s the neghbor of node j wth a hgh PDR. The nodes n O ( f ) can be consdered as the potental decodng nodes for the flows whch are encoded wth f.hence,thegeneralcodngcondtonsgvenas follows. Codng condton. For f k and f j ntersectng at node, f we denote that f k can be encoded nto f j by f k f j = ( ) ( ) ( ) ( ) O fj D fk,andsmlarly fj f = O k fk D fj, then f j and f k can be encoded together and the followng should hold: f j f k fk f j = 0. (5) For encodng m > 2 flows together, every two of them should hold the codng condton. Intutvely, the more flows are encoded together, the more NC gan can be acheved. How to encode the maxmum flows together, however,canbesummarzedasfndngthemaxmum clque n an undrected graph [3]. The maxmum clque problem s NP-complete. In [19], authors pont out that the maxmum number of flows that can be encoded wth a gven flow s bounded by a small number. Meanwhle, the probablty of encodng m > 2 flows s very low wth random flows n multhop networks. For example, f we assume that every flow s generated randomly and wth probablty p to encode wth another flow, the probablty of encodng m > 2 flows together s only p m(m 1)/2. Moreover, from our work, encodng more flows together wll cost more OR prce n the transmsson (we wll dscuss ths ssue n followng sectons). So t seems to be not worth to desgn an encodng multple flow scheme for OR. Hence, n ths paper, we only consder a two-flow encodng case, and we beleve that the multpleflowencodngcasecanbeachevedbyextendng our work. 5 Network codng n anypath routng After gvng the defnton of nter-flow network codng, we reman to use the word of network codng nstead of nter-flow network codng for clarty. The key queston of ncorporatng NC nto anypath routng s how to calculate the anypath cost.

6 Ha et al. EURASIP Journal on Wreless CommuncatonsandNetworkng 2014, 2014:148 Page 6 of 13 Accordng to characterstcs of NC and anypath routng, HCOR manly overcomes the two followng challenges. What nodes should be contaned n the forwardng set of a codng node n HCOR? What nodes should be the codng nodes n HCOR? 5.1 Forwardng set of a codng node In anypath routng, when an ntermedate node receves a packet from a flow, only the destnaton of ths packet and the nexthop whch mples the neghbors of the codng node can be known as the downstream of ths flow. Accordng to the codng condton, the potental decodng nodes of ths packet are only from ts nexthops and destnaton. Hence, we have Theorem 5.1. In opportunstc transmsson (or say anypath transmsson), f we denote the forwardng set of a codng node for a vrtual natve flow f by J,andtheforwardng set of node for the natve flow f by J, we must have J J. Proof. If the decodng node s the destnaton, we easly have J = J snce all the forwardng nodes just need to forward ths packet wthout dong the decodng. If the decodng node s not the destnaton, we denote the anypath cost for f at node by C and that for f at node by C. Accordng to the anypath transmsson, we should have max j J { } Cj C C. (6) In Lemma 4.1, we know that each neghbor j of node s n the forwardng set J f and only f C j C.Sowehave J J. 5.2 Network codng prce n anypath transmsson In ths paper, we focus the ssues of network codng only on the routng layer. So we study the cost of network codng on the transmsson level and gnore the cost on the codng level, such as addtonal codng header or addtonal computaton for network codng. From our research, the man network codng prce n anypath routng s from the shrnkage of the forwardng set. As llustrated n Fgure 2, we consder a case wth twoflow network codng. Accordng to the codng condton, becausesomeofthenodesntheforwardngsetsj and K may be not the decodng nodes of ths encoded flow, the forwardng sets J and K for encoded flow are not equal to J and K possbly. Hence, the forwardng set of each natve flow wll be changed after dong the network codng. Accordng to Theorem 5.1, the new forwardng sets J and K are the subsets of J and K, respectvely.that means, accordng to the Lemma 4.2, the costs of C J and C K wll be greater than or equal to those of C J and C K, respectvely. So addtonal prce should be pad by dong network codng n ths anypath transmsson. Let C f j,f k be the anypath cost of sendng f j and f k at relay wth uncast. Let C f j,f k be the anypath cost of sendng encoded flow f j f k wth broadcast, where f j and f j are the vrtual natve flows of f j and f k,respectvely.basedon anypath routng, the anypath cost C f j,f k can be calculated by C f j,f k = C j + Ck = c J + c K + C J + C K. (7) Smlarly, we defne the anypath cost C f j,f k by C f j,f k = c {J,K } + C {J,K }. (8) Now,thequestonswhchonesmoreexpensve,C f j,f k or C f j,f k. To ths queston, we dscuss from two cases. The frst case s that the destnatons of vrtual natve flows are the decodng nodes. In ths case, the encoded flow wll always be decoded fnally whatever node s the forwarder. So the forwardng sets J and K are equal to J and K.Due to the free-rde transmsson of network codng, we know that c {J,K} s always lower than c J + c K.Hence,C f j,f k s always more expensve n ths case. The second case s that at least one of the destnatons s not the decodng node. Accordng to codng condton, decodng nodes are only from the forwardng set or destnatons because the downstream s only composed of them n anypath routng. So there may be a forwardng set only composed of the decodng nodes, and we call ths decodng forwardng set. The nodes n the decodng forwardng set are called decodng forwardng nodes. Apparently, f destnatons are not the decodng nodes, there s no forwardng set whch contans both forwardng nodes and decodng forwardng nodes. So n the second case, at least one of the vrtual natve flows f j and f k has the decodng forwardng set. Accordng to Lemma 4.4, f at least one of J and K s the decodng forwardng set, we have J K =.Fromths, Equaton 8 s transformed to C f j,f k = c {J,K } + C J + C K. (9) Accordng to Equaton 2 of [34], we have c {J,K } = c J + c K c J c K c J + c K 1. (10) Based on the calculaton of anypath cost and the above dscusson, we know c {J,K } < c J + c K and C J + C K C J + C K.Hence,whchoneofC f j,f k and C f j,f k s more expensve s unknown n the second case. In summary, due

7 Ha et al. EURASIP Journal on Wreless Communcatonsand Networkng 2014, 2014:148 Page 7 of 13 Fgure 2 The change of forwardng set after dong network codng. Node s a relay node. The sold lne means uncast, and J and K are the forwardng sets of flow f j and f k wth uncast, respectvely. The dashed lne means broadcast, and J and K are the forwardng sets of f j and f k after dong the network codng, respectvely. to the exstence of network codng prce n opportunstc routng, we should check whether t s worth to do the network codng even f a network codng opportunty s there. 6 HCORdesgn In ths secton, we present the mplementaton detals of HCOR n a practcal dstrbuted network system. Some defntons of terms used n the followng can be found n Table Multhop overheard nformaton In the prevous sectons, we have mentoned that HCOR has a multhop codng structure. The key of realzng multhop network codng s to do multhop detecton of overheard nformaton (OI). Tradtonal network codng schemes use local OI detecton [18], n whch every node gets the OI only from ts neghborhood, such as opportunstc lstenng or learnng neghbor states n COPE. We call ths local detecton for smplcty. As llustrated n Fgure 3, local detecton msses many codng opportuntes n the multhop networks. In [3], authors proposed a multhop OI detecton method, say multhop detecton for smplcty. Ther multhop detecton s based on local detecton and works n the routng procedure. The OI of a flow s collected by local detecton and transmtted wth routng control packets to all the nodes on ths route. The OI keeps beng updated by addng some new OI from each hop. Ths process can be consdered as a combnaton of two procedures, detectng OI locally and forwardng t to the route. However, ths method has two defects. Frst, t s dependent on the routng procedure. For opportunstc routng methods whch have no routng procedure, ths method cannot work wthout routng control packets. Second, ths method s not accurate. If the route s not broken but the neghbor nodes of ths route have changed, then the OI detected by ths method wll be not rght snce the routng update does not begn. Consderng the above problems, HCOR uses each packet tself to pggyback ts own OI. Before source node sendng a packet, an OI lst wll be added nto the control header of ths packet. When a forwarder receves ths packet, the OI lst wll be updated by addng ts own OI before forwardng. The cost of nsertng an OI lst nto the control header may be senstve f we consder a large-scale ntensve network. However, compared wth broadcastng the recevng reports perodcally, our method s stll benefcal for reducng the packet collsons and control packet schedulng. In HCOR, every Fgure 3 Lmtaton of local detecton n multhop topology. There are two flows f AD and f EG n ths multhop topology. Node C s a codng node because ({A, G}, C, {D, E}) can be made as a codng structure of X topology. However, nodes B and F cannot overhear each other. Ths codng structure cannot be found f local detecton s used. Hence, many codng opportuntes are mssed n ths case.

8 Ha et al. EURASIP Journal on Wreless CommuncatonsandNetworkng 2014, 2014:148 Page 8 of 13 node mantans a neghbor table. Before sendng a packet, the node wll guess whch nodes wll overhear ths packet from ts neghbors. Only the neghbor wth hgh PDR wll be chosen (we set 0.7 as the default). If a node makes an ncorrect guess occasonally, the relevant natve packet should be retransmtted to help decode ths encoded packet. 6.2 Vrtual flow lst In HCOR, each node mantans a vrtual flow lst. The vrtual flow lst records the flows whch are passng ths node. The format of a vrtual flow lst s shown n Fgure 4. Source and Destnaton dentfy a flow unquely. Codng Flag marks that the vrtual flow s a potental encoded flow, or say beng an encoded status. Codng Expre s the expred tme of ths flow beng the encoded status. Forwardng Lst s the temporary forwardng lst of ths encoded flow. It contans the anypath cost of every potental nexthop and s set to null f no network codng happens. Expre s the expred tme of ths vrtual flow. When a node receved the frst packet from a new flow, t wll create a vrtual flow n ts vrtual flow lst. Every followng packet from ths flow wll trgger an update of ths vrtual flow. There are two dfferent updates for a vrtual flow n HCOR. One s normal update whch s trggered by every packet from the natve flow. In normal update, only the flag of Expre wll be reset. The other s codng update whch begns only after recevng an ACK packet from a codng node (we wll dscuss ths ACK process n the followng secton). In the codng update, the Codng Flag wll be set to 1, Codng Expre wll be reset, and the Forwardng Lst wll be modfed. In every node, there s a tmer to purge the vrtual flow lst. The vrtual flow wll be removed f t s expred. The Codng Flag wll be set to 0 f ths vrtual flow s not beng an encoded flow. 6.3 Codng procedure and codng feedback Whenever a node receves a new packet p k from flow f k, t executes the codng procedure llustrated n Algorthm 1. Fgure 4 Format of vrtual flow lst. Algorthm 1 Codng procedure 1: Get the forwardng lst F k, overheard nformaton O k and destnaton dst k of packet p k ; 2: f F k = then 3: return NO_ROUTE 4: end f 5: Pck a natve packet p j n the head of output queue; 6: f p j has been operated then 7: p = p k and jump to lne 36; 8: end f 9: Get the forwardng lst F j, overheard nformaton O j and destnaton dst j of p j ; 10: f dst j s n O k. then 11: codng j = 1; 12: next j = F j, decodng j = { } dst j ; 13: else 14: codng j = 0; 15: next j = F j O k, decodng j = next j ; 16: end f 17: f dst k s n O. then 18: codng k = 1; 19: next k = F k, decodng k = {dst k }; 20: else 21: codng k = 0; 22: next k = F k O j, decodng k = next k ; 23: end f 24: f next k.sze > 0 && next j.sze > 0 && decodng j decodng k ==. then 25: C k,j (F j, F k, dst j, dst ); 26: C k,j (next j, next k, decodng j, decodng k ); 27: f C k,j < C k,j or codng j &&codng k then 28: p = p j p k ; 29: nexthop = { } next j, next k ; 30: decodng = { } decodng j, decodng k ; 31: Jump to lne 36; 32: end f 33: end f 34: Insert p j nto the tral of the output queue; 35: Repeat lne 5; 36: Insert p nto the output queue; Lne 24 mples the codng condton and Lemma 4.4. The costs C k,j and C k,j obey Equatons 7 and 8. In lne 27, we use anypath cost as the judgng crtera of network codng. Meanwhle, as we analyzed n Secton 5.2, we always do network codng f the decodng node s the destnaton for each vrtual natve flow. As mentoned n Secton 4.3, ths codng procedure only realzes two-flow XORng. For m > 2 flow XORng, f the calculaton of anypath cost can be expanded to the multple flow case, t can be realzed wth greedy strategy by only removng the word natve n lne 5.

9 Ha et al. EURASIP Journal on Wreless Communcatonsand Networkng 2014, 2014:148 Page 9 of 13 Fgure 5 Illustratve scenaros. (a) Chan topology. (b) X topology. (c) Multhop topology. When node does the network codng, after sendng m encoded packets n a perod of tme (we set 0.1 s as default n our smulaton), an ACK packet s fed back from to the prevous hop of p k.thsackcanbeconsderedasaknd of routng feedback, and we call t codng feedback.codng feedback contans three attrbutes n ts control header: src, dst,andcost.thesrc and dst record the source and destnaton of p k,respectvely.theparofsrc and dst dentfes the flow f k.thecost records the anypath cost of f k after dong network codng. In Equaton 8, anypath cost C k,j s related to both f k and f j.meanwhle, the gan of network codng s only from one broadcast of the encoded packet p. Hence, we defne the temporary anypath cost of f k after dong network codng by C k = d j Ck,j, (11) d k + d j where d k s the PDR of encoded packet p from to nexthop k. After the neghbor of receves ths codng feedback, ts vrtual flow lst wll be updated. The vrtual flow (src, dst) updates ts forwardng lst by modfyng the anypath cost of nexthop to cost. After dong ths, the anypath cost of node for flow f k s updated to C k = αc k + (1 α) C k,whereα = 1f0< Ck < C k, and α = 0 otherwse. Based on the update of vrtual flow lst, the nexthops recorded n the vrtual flow wll be more compettve to acheve a hgh prorty to forward ths packet. If network codng dsappears, no codng feedback causes no vrtual flow update. After reachng the expraton tme, vrtual flow wll be purged, and then the new packet from ths flow wll be forwarded followng ts orgnal anypath cost. 6.4 Decodng In HCOR, every node works n the promscuous mode and stores all packets sent and overheard by tself n the past T seconds n ts Packet Pool. The decodng process happens when an encoded packet arrves at a node and theaddressofthsnodesnthedecodng. Becausewe use the XOR codng method, the encoded packet can be decoded only by XORng any of the vrtual natve packets, such as p 1 = (p 1 p 2 ) p 2. 7 Experment results Ths secton uses a ns-2 smulator to present the performance evaluaton. We study the performance of HCOR by comparng wth one non-nc scheme and one network codng scheme. The non-nc scheme s just the anypath routng scheme wthout dong any network codng.thencschemescalledcoor,whchsakndof Fgure 6 End-to-end throughput performance n chan topology. (a) Throughput performance wth a fxed PDR of 0.8 and varyng loads from 10 kb/s to 1 mb/s. (b) Throughput performance wth fxed loads of 400 kb/s and varyng PDR from 0.3 to 1.

10 Ha et al. EURASIP Journal on Wreless CommuncatonsandNetworkng 2014, 2014:148 Page 10 of 13 (a) Fgure 7 End-to-end throughput performance n X topology. (a) Throughput performance wth a fxed PDR of 0.8 and varyng loads from 10 kb/s to 1 mb/s. (b) Throughput performance wth a fxed rate of 400 kb/s for each load and varyng PDR from 0.3 to 1. (b) codng opportunty-aware OR method and runs on a multhop codng structure. The dfference between HCOR and COOR s that COOR does the network codng whenever a codng opportunty happens, whle HCOR does network codng only f a gan can be acheved. In our smulatons, all the nodes are set to the promscuous mode wth a modfed IEEE standard as the MAC protocol whch supports the opportunstc transmsson. We use User Datagram Protocol (UDP) traffc sources, and all the flows are constant bt rate (CBR), wth a fxed packet sze of 512 bytes. The transmsson range s set to 250 m, and the nterference range s set to 550 m wth the TwoRayGround propagaton model [35]. 7.1 Results from llustratve scenaros We present the performance of HCOR compared wth thenon-ncschemensomebasctopologes.frstly,we study the chan topology wth bdrectonal flows shown n Fgure 5a. We set all lnks wth a fxed PDR. Two symmetrcal loads have the same rate from A to B and B to A, respectvely. The measurements of average end-to-end throughput are plotted n Fgure 6. The throughput gan s about 23% after symmetrcal load runnng n the saturaton levels shown n Fgure 6a. The reason why HCOR and non-nc have dfferent varaton trends n Fgure 6b s because the relay node n non-nc suffers from bottleneck, and hence, the bandwdth of ts output lnks wll be saturated when the loads keep ncreasng. In HCOR, the relay node mxes the two flows together and sends ths mxture as one flow. So the rate of output s always lower than that of nput. If nput and output lnks have the same bandwdth, there s no bottleneck problem n HCOR because output lnks are always starved. Next, we study X topology shown n Fgure 5b. The measurement results are plotted n Fgure 7. We can see that the throughput gan s stll sgnfcant and about 21% n X topology. Dfferent from the chan scenaro, there are (a) Fgure 8 End-to-end throughput performance n multhop scenaro. (a) Throughput performance wth a fxed PDR of 0.8 and varyng loads from 10 kb/s to 1 mb/s. (b) Throughput performance wth a fxed rate 400 kb/s for each load and varyng PDR from 0.3 to 1. (b)

11 Ha et al. EURASIP Journal on Wreless Communcatons and Networkng 2014, 2014:148 Page 11 of 13 Fgure 9 Cellular scenaro. (a) Hexagon topology. (b) Throughput performance. overhearng lnks n the X scenaro. In our smulaton, we set the threshold of guessng a successful overhearng to 0.7 as default. That means a node guesses one of ts neghbors can overhear the packets t sends f the PDR from t to ths neghbor s more than 0.7. Hence, we can see that the performance of HCOR and non-nc s smlar when PDR s no more than 0.7 n Fgure 7b. After that, a gan can be acheved by HCOR. At last, a smple multhop scenaro s studed to present the multhop structure of HCOR. Ths multhop topology s shown n Fgure 5c. We plot the smulaton results n Fgure 8. The gan of HCOR s not as much as the above two scenaros. The reason may be the gan from bottleneck s dluted by bandwdth consumpton from multhop transmsson. But we can stll have more than 10% gan for total end-to-end throughput over non-nc. In Fgure 8b, HCOR almost always has a gan over non-nc because multhop topology has multple codng structure, such as (A, B, C), (G, F, C), and ({B, E}, C, {F, D}), where (A, B, C) and (G, F, C) are not affected by the overhearng lnks. 7.2 Results from specfc scenaros In ths secton, we want to present how flexble HCOR s. We am to pont out the beneft of HCOR compared wth COOR. To do ths, we desgn two specfc scenaros. One s cellular scenaro wth a hexagon topology. The other s damond scenaro wth a double-damond topology. As shown n Fgure 9a, there are seven nodes and two flows f23 and f65 wth the same rate n the cellular scenaro. The PDR of lnks (2, 1), (2, 4), (1, 3), (6, 7), (6, 4), and (7, 5) s set to 0.8. The PDR of lnks (4, 1) and (4, 7) s set to 0.4. The PDR of lnks (4, 5) and (4, 3) s set to 0.5. The PDR of lnks (2, 5) and (6, 3) s set to 1. The potental forwardng nodes for f23 are nodes 1 and 4. The potental forwardng nodes for f65 are nodes 7 and 4. Apparently, there s a codng opportunty at node 4 because the overhearng lnks (2, 5) and (6, 3) have very good qualty. But the output lnks of node 4 are very poor. So ntutvely, t seems that there s no gan for dong network codng at node 4. The smulaton results plotted n Fgure 9b demonstrate that our ntuton s correct. COOR shows a poor performance n ths scenaro and has about 30% lower throughput gan than HCOR. In the Fgure 10a, we present a damond scenaro wth two symmetrcal loads f67 and f21. The PDR of lnks (5, 7), (2, 5), (6, 3), and (3, 1) s set to 0.9. The PDR of lnks (6, 4), (2, 4), (4, 7), and (4, 1) s set to 0.7. The PDR of lnks (4, 5) and (4, 3) s set to Accordng to anypath routng, Fgure 10 Damond scenaro. (a) Double-damond topology. (b) Throughput performance.

12 Ha et al. EURASIP Journal on Wreless CommuncatonsandNetworkng 2014, 2014:148 Page 12 of 13 Fgure 11 Throughput performance of random scenaro. (a) Throughput vs. offered loads. (b) Throughput vs. PDR. node 5(3) s the forwarder of node 4 for flow f 67 ( f 21 )snce node 5(3) has better PDR to the destnaton than node 4. The structure composed of 6, 3, 4, 2, and 5 s a basc codng structure of X topology. So there s stll a codng opportunty at node 4. The smulaton results show that there s no codng gan n ths scenaro. The performance of COOR s 28% lower than that of HCOR. 7.3 Results from random scenaro At last, we consder the performance of HCOR n a larger random mesh network. We construct ten 50-node random topologes wth sze 1,000 1,000 and 20 random loads for each topology. All these loads have random sources and destnatons wth the same rate and random tme of duraton. The PDR of all lnks n these topologes s set to the lowest bound PDR, whch means the PDR of every lnk s set randomly but hgher than the lowest bound PDR value. We compare HCOR wth COOR and plot the average throughput performance of these ten topologes n Fgure 11. We frst set the lowest bound PDR to 0.6 and show the throughput performance wth ncreasng loads n Fgure 11a. The results show that HCOR has more than 25% gan over COOR n ths case. In another case, we vary the lowest bound PDR from 0.3 to 1 and use a fxed load rate of 400 kb/s. The throughput performance of the second case s shown n Fgure 11b. We can see that the gan of HCOR s 13% hgher than that of COOR, and the promoton s obvous between 0.5 to 0.8 of PDR. 8 Concluson In ths paper, we proposed a novel codng-aware opportunstc routng scheme HCOR. We study the characterstc of ncorporatng NC nto opportunstc routng and fgure out some conclusons to help desgn a codng-aware opportunstc routng. Takng advantage of anypath cost, we gve a computng method to calculate network codng cost n opportunstc transmsson. Based on the above study, we desgn a dstrbuted routng scheme to realze HCOR, whch consders the multhop codng structure n the desgn. We mplement the HCOR scheme n ns-2 and carry out extensve smulatons to show the good performanceofhcor.thepropertyofourworkmaybeuseful for future codng-aware opportunstc routng desgn; n partcular, our work leaves wde open on the desgn of codng-aware opportunstc schemes wth encodng of more than two natve flows together. Endnotes a Some also call t nter-sesson network codng. b In the followng of ths paper, we smplfy C p to C f we do not emphasze the packet or flow n the calculaton of anypath cost. Competng nterests The authors declare that they have no competng nterests. Acknowledgements Ths work s fnancally supported by Natonal Natural Scence Foundaton of Chna under grant numbers , , , and ; the Fundamental Research Funds for the Central Unverstes under grant numbers DUT13JS09 and DUT14QY04; and the Specalzed Research Fund for the Doctoral Program of Hgher Educaton of Chna under grant number Receved: 22 Aprl 2014 Accepted: 25 August 2014 Publshed: 11 September 2014 References 1. R Ahlswede, N Ca, SYR L, RW Yeung, Network nformaton flow. IEEE Trans. Inf. Theory. 46(4), (2000) 2. S Chachulsk, M Jennngs, S Katt, D Katab, Tradng structure for randomness n wreless opportunstc routng, n Proc. of ACM SIGCOMM 07, (Kyoto, Japan, August 2007) 3. JL Le, JCS Lu, DM Chu, DCAR: dstrbuted codng-aware routng n wreless networks. IEEE Trans. Moble Comput. 9(4), (2010)

13 Ha et al. EURASIP Journal on Wreless Communcatonsand Networkng 2014, 2014:148 Page 13 of B Guo, HK L, C Zhou, Y Cheng, Analyss of general network codng condtons and desgn of a free-rde-orented routng metrc. IEEE Trans. Vehcular Technol. 60(4), (2011) 5. Y Yan, BX Zhang, J Zheng, M Jan, CORE: a codng-aware opportunstc routng mechansm for wreless mesh networks [accepted from open call]. IEEE Wreless Commun. 17(3), (2010) 6. Y Benfattoum, S Martn, K Al Agha, IROCX: nterference-aware routng wth opportunstcally coded exchanges n wreless mesh networks, n Wreless Communcatons and Networkng Conference (WCNC) 2011 IEEE. (Qunyana-roo, Mexco, March 2011), pp do: /wcnc YJ Ln, CC Huang, JL Huang, PpelneOR: a ppelned opportunstc routng protocol wth network codng n wreless mesh networks, n IEEE 71st Vehcular Technology Conference, VTC 2010-Sprng (Tape, Tawan, May 2010), pp do: /vetecs Y Yan, BX Zhang, HT Mouftah, J Ma, Practcal codng-aware mechansm for opportunstc routng n wreless mesh networks, n IEEE Internatonal Conference on Communcatons, ICC 08 (Bejng, Chna, May 2008), pp do: /icc R Laufer, H Dubos-Ferrère, L Klenrock, Polynomal-tme algorthms for multrate anypath routng n wreless multhop networks. IEEE/ACM Trans. Netw. 20(3), (2012) 10. A Agarwal, M Charkar, On the advantage of network codng for mprovng network throughput, n IEEE Informaton Theory Workshop, ITW 04 (San Antono, TX, USA, Oct 2004), pp do: /itw KK Ch, XH Jang, S Horguch, Network codng opportunty analyss of cope n multhop wreless networks, n IEEE Wreless Communcatons and Networkng Conference. WCNC 2008, (Las Vegas, NV, USA, 31 March 3 Aprl 2008), pp do: /wcnc R Koetter, M Medard, An algebrac approach to network codng. IEEE/ACM Trans. Netw. 11(5), (2003) 13. S Jagg, P Sanders, PA Chou, M Effros, S Egner, K Jan, L MGM, Polynomal tme algorthms for multcast network code constructon. IEEE Trans. Inf. Theory. 51(6), (2005) 14. M Effros, H Tracey, K Sukwon, A tlng approach to network code desgn for wreless networks, n IEEE Informaton Theory Workshop, 2006 (Punta del Este, Uruguay, March 2006), pp do: /itw M Badas, A MacKenze, Many-to-many space-tme network codng for amplfy-and-forward cooperatve networks: node selecton and performance analyss. EURASIP J. Wreless Commun. Netw. 2014(1). 48 (2014). do: / M Nabaee, F Labeau, Quantzed network codng for correlated sources. EURASIP J. Wreless Commun. Netw. 2014(1). 40 (2014). do: / T Lv, S L, W Geng, Combnng cooperatve dversty and network codng n uplnk mult-source mult-relay networks. EURASIP J. Wreless Commun. Netw. 2013(1). 241 (2013). do: / S Katt, H Rahul, H Wenjun, D Katab, M Medard, J Crowcroft, XORs n the ar: practcal wreless network codng. IEEE/ACM Trans. Netw. 16(3), (2008) 19. JL Le, JCS Lu, DM Chu, On the performance bounds of practcal wreless network codng. IEEE Trans. MobleComput. 9(8), (2010) 20. A Khreshah, I Khall, P Ostovar, J Wu, Flow-based XOR network codng for lossy wreless networks. IEEE Trans. Wreless Commun. 11(6), (2012). do: /twc Y Wu, Broadcastng when recevers know some messages a pror, n IEEE Internatonal Symposum on Informaton Theory, ISIT 2007 (Nce, France, June 2007), pp do: /isit CC Wang, On the capacty of wreless 1-hop ntersesson network codng: a broadcast packet erasure channel approach. IEEE Trans. Inf. Theory. 58(2), (2012) 23. MH Amermehr, F Ashtan, Delay and throughput analyss of a two-way opportunstc network codng-based relay network. IEEE Trans. Wreless Commun. 13(5), (2014) 24. A Yang, ZS Fe, CW Xng, M Xao, JH Yuan, JM Kuang, Desgn of bnary network codes for multuser multway relay networks. IEEE Trans. Vehcular Technol. 62(8), (2013) 25. MF Jhang, SW Ln, WJ Lao, C2AR: codng and capacty aware routng for wreless ad hoc networks, n IEEE Internatonal Conference on Communcatons (ICC 10) (Cape Town, South Afrca, May 2010), pp do: /icc LF Xe, PHJ Chong, SC Lew, YL Guan, CEO: consstency of encodng and overhearng n network codng-aware routng. IEEE Wreless Commun. Lett. 2(2013), do: /wcl S Sengupta, S Rayanchu, S Banerjee, An analyss of wreless network codng for uncast sessons: the case for codng-aware routng, n 26th IEEE Internatonal Conference on Computer Communcatons, INFOCOM 2007 (Anchorage, AK, 6 12 May 2007), pp do: /infcom DB Johnson, DA Maltz, Dynamc source routng n ad hoc wreless networks. The Kluwer Int. Seres n Eng. Comput. Sc. Moble Comput. 353, (1996) 29. PL,SGuo,SYu,AVaslakos,Relable multcast wth ppelned network codng usng opportunstc feedng and routng. IEEE Trans. Parallel Dstrbuted Syst. PP(99), 1 1 (2014). do: /tpds T Chen, S Zhong, An enforceable scheme for packet forwardng cooperaton n network codng wreless networks wth opportunstc routng. IEEE Trans. Vehcular Technol. PP(2014), 1 1. do: /tvt M Aajam, H-R Park, J-B Suk, Combnng opportunstc routng and network codng: a mult rate approach, n IEEE Wreless Communcatons and Networkng Conference, WCNC 2013 (Shangha, Chna, 7 10 Aprl 2013), pp do: /wcnc MA Iqbal, B Da, B Huang, A Hassan, S Yu, Survey of network codng-aware routng protocols n wreless networks. J. Netw. Comput. Appl. 34(6), (2011) 33. D Couto, D Aguayo, J Bcket, pr Morrs, A hgh-throughput path metrc for mult-hop wreless routng, n Proc. of the 9th Annual Internatonal Conference on Moble Computng and Networkng, MobCom 03 (San Dego, USA, Sept 2003), pp B N, N Santhapur, ZF Zhong, S Nelakudt, Routng wth opportunstcally coded exchanges n wreless mesh networks, n 2nd IEEE Workshop on Wreless Mesh Networks, WMesh 2006 (Reston, VA, USA, Sept 2006), pp do: /wimesh T Yang, G Mno, L Baroll, A Durres, F Xhafa, Comparson evaluaton for moble and statc sensor nodes n wreless sensor networks consderng tworayground and shadowng propagaton models, n 14th Internatonal Conference on Network-Based Informaton Systems, NBS 2011 (Trana, Albana, 7 9 Sept 2011), pp do: /nbs do: / Cte ths artcle as: Ha et al.: HCOR: a hgh-throughput codng-aware opportunstc routng for nter-flow network codng n wreless mesh networks. EURASIP Journal on Wreless Communcatons and Networkng :148. Submt your manuscrpt to a journal and beneft from: 7 Convenent onlne submsson 7 Rgorous peer revew 7 Immedate publcaton on acceptance 7 Open access: artcles freely avalable onlne 7 Hgh vsblty wthn the feld 7 Retanng the copyrght to your artcle Submt your next manuscrpt at 7 sprngeropen.com