BSMR: Byzantine-Resilient Secure Multicast Routing in Multi-hop Wireless Networks

Transcription

1 Purdue University Purdue e-pubs Department f Cmputer Science Technical Reprts Department f Cmputer Science 07 BSMR: Byzantine-Resilient Secure Multicast Ruting in Multi-hp Wireless Netwrks Reza Curtmla Cristina Nita-Rtaru Purdue University, crisn@cs.purdue.edu Reprt Number: Curtmla, Reza and Nita-Rtaru, Cristina, "BSMR: Byzantine-Resilient Secure Multicast Ruting in Multi-hp Wireless Netwrks" (07). Department f Cmputer Science Technical Reprts. Paper This dcument has been made available thrugh Purdue e-pubs, a service f the Purdue University Libraries. Please cntact epubs@purdue.edu fr additinal infrmatin.

2 BSMR: BYZANTINE-RESILIENT SECURE MULTICAST ROUTING IN MULTI-HOP WIRELESS NETWORKS Reza Curtmla Cristina Nita-Rtaru Nita-Rtam CSD TR # March 07

3 BSMR: Byzantine-Resilient Secure Multicast Ruting in Multi-hp Wireless Netwrks Reza Curtmla Department f Cmputer Science The Jhns Hpkins University crix@cs.jhu.edu Cristina Nita-Rtaru Department f Cmputer Science and CERIAS Purdue University crisn@cs.purdue.edu@ cs.purdue.edu Abstract Multi-hp wireless netwrks rely n nde cperatin t prvide unicast and multicast services. The multi-hp cmmunicatin ffers increased cverage fr such services, but als makes them mre vulnerable t insider (r Byzantine) attacks cming frm cmprmised ndes that behave arbitrarily t disrupt the netwrk. In this wrk we identify vulnerabilities f n-demand multicast ruting prtcls fr multi-hp wireless netwrks and discuss the challenges encuntered in designing mechanisms t defend against them. We prpse BSMR, a nvel secure multicast ruting prtcl that withstands insider attacks frm clluding adversaries. Our prtcl is a sftware-based slutin and des nt require additinal r specialized hardware. We present simulatin results which demnstrate that BSMR effectively mitigates the identified attacks. 1. INTRODUCTION Multicast ruting prtcls deliver data frm a surce t multiple destinatins rganized in a multicast grup. Several prtcls were prpsed t prvide multicast services fr multi-hp wireless netwrks. These prtcls rely n nde cperatin and use flding [1], [I], gssip [2], gegraphical psitin [3], r disseminatin structures such as meshes [4], [5], r trees [6], [7]. A majr challenge in designing prtcls fr wireless netwrks is ensuring rbustness t failures and resilience t attacks. Wireless netwrks prvide a less rbust cmmunicatin than wired netwrks due t frequent brken links and a higher errr rate. Security is als mre challenging in multi-hp wireless netwrks because the pen medium is mre susceptible t utside attacks and the multi-hp cmmunicatin makes services mre vulnerable t insider attacks cming frm cmprmised ndes. Althugh an effective mechanism against utside attacks, authenticatin is nt sufficient t prtect against insider attacks because an adversary that cmprmised a nde als gained access t the cryptgraphic keys stred n it. Insider attacks are als knwn as Byzantine [8] attacks and prtcls able t prvide service in their presence are referred t as Byzantine resilient prtcls. Previus wrk fcused mainly n the security f unicast services. Several ruting prtcls [9]-[12] were prpsed t cpe with utsider attacks. Methds prpsed t address insider threats in unicast ruting include mnitring [13], multi-path ruting [14] and acknwledgment-based feedback [15], [16]. The prblem f secure multicast in wireless netwrks was less studied and nly utside attacks were addressed'[17] - [17].. Security prblems related t multicast ruting can be classified in ruting specific security, such as the management f the ruting structure and data frwarding, and applicatin specific security such as data cnfidentiality and authenticity. Slutins t the latter prblem als referred t as secure grup cmmunicatin fcus mainly n grup key management [18], [19]. In this wrk we are cncerned with multicast ruting specific security. Several differences make the multicast cmmunicatin mdel mre challenging than its unicast cunterpart. First, designing secure multicast prtcls fr wireless netwrks requires a mre cmplex trust mdel, cunter- as ndes which are members f the multicast grup cannt simply rganize themselves in a disseminatin structure withut the help f ther nn-member ndes acting as ruters.

4 Secnd, unlike unicast prtcls which establish and maintain rutes between tw ndes, multicast prtcls establish and maintain mre cmplex structures, such as trees r meshes. Fr example, prtcls relying n trees require additinal peratins such as rute activatin, tree pruning and tree merging. These actins d nt have a cunterpart in the unicast case and may expse the ruting prtcl t new vulnerabilities. Third, multicast prtcls deliver data frm ne sender t multiple receivers making scalability a majr prblem when designing attack-resilient prtcls. In particular, slutins that ffer resiliency against Byzantine attacks fr unicast are nt scalable in a multicast setting. Fr example, multi-path ruting affects significantly the data disseminatin efficiency, while strategies based n end-t-end acknwledgments have high verhead. In this paper we study vulnerabilities f multicast ruting prtcls in multi-hp wireless netwrks and prpse a new prtcl that prvides resilience against Byzantine attacks. Our main cntributins are: We identify several aspects that make the design f attack-resilient multicast ruting prtcls mre challenging than their unicast cunterpart, such as a mre cmplex trust mdel and underlying ruting structure, and scalability. We als discuss ptential attacks against such prtcls. We prpse BSMR, an n-demand multicast prtcl fr multi-hp wireless netwrks which relies n several mechanisms t mitigate Byzantine attacks. BSMR uses a selective data frwarding detectin mechanism that relies n a reliability metric capturing adversarial behavir. Ndes determine the reliability f links by cmparing the perceived data rate with the ne advertised by the surce. Adversarial links are avided during the rute discvery phase. BSMR als prevents attacks that try t prevent r arbitrarily influence rute establishment. We shw thrugh simulatins that the impact f several Byzantine attacks (fld rushing, black hle and wrmhle) n a previusly prpsed secure multicast ruting prtcl is cnsiderable and cannt be ignred. We als demnstrate thrugh simulatins that ur prtcl BSMR mitigates the attacks, while incurring a small verhead. The remainder f the paper is rganized as fllws. Sectin II I1 verviews related wrk. Sectin III I11 presents ur netwrk and system mdel. We discuss the attacks against multicast in IV-B and present BSMR in Sectin V. We present experimental results in Sectin VI and cnclude in Sectin VII. II. RELATED WORK Significant wrk addresses vulnerabilities f unicast ruting prtcls in wireless netwrks. Several secure ruting prtcls resilient t utside attacks were prpsed in the last few years such as Ariadne [11], [l 11, SEAD [], [l], ARAN [12], and the wrk in [9]. Wireless specific attacks such as fld rushing and wrmhle were recently identified and studied. RAP [] prevents the rushing attack by waiting fr several fld requests and then randmly selecting ne t frward, rather than always frwarding nly the first ne. Techniques t defend against wrmhle attacks include Packet Leashes [21] which restricts the maximum transmissin distance by using either a tight time synchrnizatin r lcatin infrmatin, Truelink [22] which uses MAC level acknwledgments t infer if a link exists r nt between tw ndes, and the technique in [23], which relies n directinal antennas. The prblem f insider threats in unicast ruting was studied in [13]-[16]. Watchdg [13] relies n a nde mnitring its neighbrs if they frward packets t ther destinatins. If a nde des nt verhear a neighbr frwarding mre than a threshld number f packets, it cncludes that the neighbr is adversarial. SDT [14] uses multi-path ruting t prevent a malicius nde frm selectively drpping data. ODSBR [15], [16] prvides resilience t Byzantine attacks caused by individual r clluding ndes by detecting malicius links based n an acknwledgement-based feedback techniques. Mst f the wrk addressing applicatin security issues related t multicast in wireless netwrks fcused n the prblem f grup key management in rder t ensure data cnfidentiality and authenticity [24]-

5 [28]. Wrk studying multicast ruting specific security prblems in wireless netwrks is scarce with the ntable exceptin f the authenticatin framewrk by Ry et al. [17]. The framewrk allws MAODV t withstand several external attacks targeted against the creatin and maintenance f the multicast tree. Hwever, it des nt prvide resilience against Byzantine attacks. Multicast ruting specific security was als studied in verlay netwrks [29]-[31]. Slutins prpsed explit verlay specific prperties such as: existence f netwrk cnnectivity between each pair f ndes which allws ndes t directly prbe nn-neighbring ndes, and highly redundant cnnectivity, guaranteeing that many disjint paths exist. Nne f these prperties hld in multi-hp wireless netwrks NETWORK AND SYSTEM MODEL III. NETWORK AND SYSTEM MODEL A. Netwrk Mdel We cnsider a multi-hp wireless netwrk where ndes participate in the data frwarding prcess fr ther ndes. We assume that the wireless channel is symmetric. All ndes have the same transmitting pwer and cnsequently the same transmissin range. The receiving range f a nde is identical t its transmissin range. Ndes are nt required t be equipped with additinal hardware such as GPS receivers r tightly synchrnized clcks. Als, ndes are nt required t be tamper resistant: If an attacker cmprmises a nde, it can extract all key material, data r cde stred n that nde. B. Multicast Prtcl We assume a tree-based n-demand multicast prtcl such as [6]. The prtcl maintains bi-directinal shared multicast trees cnnecting multicastsurces surces and receivers. Each multicast grup has a crrespnding multicast tree. The multicast surce is a special nde, the grup leader, whse rle is t eliminate stale rutes and crdinate grup merges. Rute freshness is indicated by a grup sequence number updated by the grup leader and bradcast peridically in the entire netwrk. Higher grup sequence numbers dente fresher rutes. The main peratins f the prtcl are rute discvery, rute activatin and tree maintenance. During rute rute discvery a nde discvers a path t a nde that is part f the multicast tree. A requester first bradcasts a rute request message that includes the latest knwn grup sequence number. The rute request message is flded in the netwrk using a basic fld suppressin mechanism and establishes reverse rutes t the surce f the request. Upn receiving the rute request, a nde that is part f the multicast tree and has a grup sequence number at least as large as the ne in the rute request, generates a rute reply message and unicasts it n the reverse rute. The rute reply message includes the last knwn grup sequence number and the number f hps t the nde that riginated the rute reply. During rute activatin, the requester selects the freshest and shrtest rute (i.e., with the smallest number f hps t the multicast tree) frm the rutes returned by the rute discvery peratin. The requester activates that rute by unicasting a multicast activatin message. Three main peratins ensure the tree maintenance: tree pruning, brken link repair and tree merging. Tree pruning cclirs ccurs when a grup member that is a leaf in the multicast tree decides t leave the grup. T prune itself frm the tree, the nde sends a message t indicate this t its parent. The pruning message travels up the tree causing leaf ndes that are nt members f the multicast grup t prune themselves frm the tree, until it reaches either a nn-leaf nde r a grup member. A nn-leaf grup member must cntinue t act as a ruter and cannt prune itself frm the multicast tree. A nde initiates a link repair prcedure when the upstream link in the multicast tree breaks. If the nde cannt recnnect t the tree, it means the tree is partitined. In this case the nde runs a special prcedure t prune nn-member leaf ndes and elect a grup leader fr the partitin. When tw partitins f the same tree recnnect, the leader f ne f the partitins crdinates the merge f the partitins, suppressing the ther leader.

6 IV. ATTACKS AGAINST MULTICAST ROUTING A. Adversarial Mdel We assume that ndes may exhibit Byzantine behavir, either alne r clluding with ther ndes. Examples f such behavir include: nt frwarding packets, injecting, mdifying r replaying packets. We refer t any arbitrary actin by authenticated ndes resulting in disruptin f the ruting service as Byzantine behavir, and t such an adversary as a Byzantine adversary. We cnsider a three-level trust mdel that captures the interactins between ndes in a wireless multicast setting and defines a nde's privileges: a first level includes the surce which must be cntinually available and assumed nt t be cmprmised; a secnd level cnsists f the multicast grup member ndes, which are allwed t initiate requests fr jining multicast grups; and a third level cnsists f nn-member ndes which participate in the ruting but are nt entitled t initiate grup jin requests. In rder t cpe with Byzantine attacks, even grup members cannt be fully trusted. An attacker can disrupt the physical layer by jamming, and MAC prtcls such as can be disrupted by attacks using the special RTS/CTS RTSICTS packets. This wrk nly cnsiders attacks targeted against the netwrk level. Als, preventing traffic analysis is nt the gal f this wrk, which fcuses instead n survivable ruting. B. Attacks in Multicast in Multi-Hp Wireless Netwrks An adversary can attack cntrl messages crrespnding t the rute discvery, rute activatin and tree management cmpnents f the ruting prtcl, r can attack data messages. The rute discvery phase can be disrupted by utside attackers creating undesired results by injecting, replaying, r mdifying cntrl packets. Ndes that are nt in the tree can mislead ther ndes int believing that they fund and are cnnected t the tree. Ndes can fld the netwrk with bgus requests fr jining multicast grups. A Byzantine adversary can prevent a rute frm being established by drpping the request and/r respnse, r can influence the rute selectin by using wireless specific attacks such as wrmhle and fld rushing. A Byzantine adversary can als mdify the packets carrying the rute selectin metric such as hp cunt r nde identifiers. An utside attacker can inject bgus rute activatin messages, r prevent crrect rute activatin messages t reach all ndes. Ndes can maliciusly reprt that ther links are brken r generate incrrect pruning messages resulting in crrect ndes being discnnected frm the netwrk r tree partitining. In the absence f authenticatin, any nde can pretend t be the grup leader. Althugh many ruting prtcls d nt describe hw t select a new grup leader when needed, we nte that the leader electin prtcl can als be influenced by attackers. Attacks against data messages cnsist f eavesdrpping, mdifying, replaying, injecting data, r selectively frwarding data after being selected n a rute. A special frm f packet delivery disruptin is a denial f service attack, in which the attacker verwhelms the cmputatinal, sending r receiving capabilities f a nde. In general, data surce authenticatin, integrity and encryptin can slve the first attacks and are usually cnsidered applicatin specific security. Defending against selective data frwarding and denial f service cannt be dne exclusively by using cryptgraphic mechanisms. V. SECURE MULTICAST ROUTING PROTOCOL A. BSMR Overview Our prtcl ensures that multicast data is delivered frm the surce t the members f the multicast grup, even in the presence f Byzantine attackers, as lng as the grup members are reachable thrugh nn-adversarial paths and a nn-adversarial path exists between a new member and a nde in the multicast tree. se-

7 T eliminate a large class f utside attacks we use an authenticatin framewrk that ensures nly authrized ndes can perfrm certain peratins (e.g., nly tree ndes can perfrm tree peratins and nly ndes that pssess valid grup certificates can cnnect t the crrespnding multicast tree). BSMR mitigates inside attacks that try t prevent a nde frm establishing a rute t the multicast tree by flding bth rute request and rute reply such that if an adversarial-free rute exists, then a rute is established. BSNIR BSMR ensures resilience t selective data frwarding attacks by using a reliability metric that captures adversarial behavir. The metric cnsists f a list f link weights in which high weights crrespnd t lw reliability. Each nde in the netwrk maintains its wn weight list and includes it in each rute request t ensure that a new rute t the tree avids adversarial links. A link's reliability is determined based n the number f packets successfully delivered n that link ver time. Tree ndes mnitr the rate f receiving data packets and cmpare it with the transmissin rate indicated by the surce in the frm f an NIRATE MRATE message. If the perceived transmissin rate falls belw the rate indicated in the MRATE message by mre than a threshld, an hnest nde that is a direct descendant f an adversarial nde updates its weight list by penalizing the link t its parent and then tries t discver a new rute t the tree. We nte that a strategy based n end-t-end acknwledgments, althugh shwn effective in unicast [14], [16], is nt scalable: As the size f the multicast grup increases, ACK implsin ccurs at the surce, which may cause a drastic decrease in data delivery [32]. Mrever, slutins that address the prblem f feedback implsin in multicast prtcls (e.g., feedback aggregatin r a cmbinatin f ACWNACK ACKINACK messages [33]) were designed t perate under nn-adversarial cnditins; It is questinable if they will wrk in adversarial netwrks. Withut lss f generality, we limit ur descriptin t ne multicast grup. Belw we describe the previusly mentined authenticatin framewrk, the rute discvery, the rute activatin, multicast tree maintenance and the selective data frwarding detectin mechanisms. B. Authenticatin Framewrk In rder t prtect frm external attacks against the creatin and maintenance f the multicast tree BSMR uses a framewrk similar with the ne in [17]. The framewrk prevents unauthrized ndes t be part f the netwrk, f a multicast grup, r f a multicast tree. These frms f authenticatin crrespnd t the trust mdel described in Sectin IV-A. Each nde authrized t jin the netwrk has a pair f publiclprivate public/private keys and a nde certificate certijicate that binds its public key t its IP address. Each nde authrized t jin a multicast grup has an additinal grup certificate certijicate that binds its public key and IP address t the IP address f the multicast grup. Ndes in the multicast tree are authenticated using a tree tken, which is peridically refreshed and disseminated by the grup leader in the multicast tree with the help f pairwise shared keys established between every direct tree neighbrs. Thus, nly ndes that are currently n the tree will have a valid tree tken. T allw any nde in the netwrk t check that a tree nde pssesses a valid tree tken, the grup leader peridically bradcasts in the entire netwrk a tree tken authenticatr f f(tree tken), where f is a cllisin resistant ne-way functin. Ndes can check the validity f a given tree tken by applying the functin f t it and cmparing the result with the latest received tree tken authenticatr. T prevent tree ndes frm claiming t be at a smaller hp distance frm the grup leader than they actually are, we use a technique based n a ne-way hash chain. The last element f this hash chain, referred t as hp cunt anchr, is bradcast peridically by the grup leader. We assume that ndes have a methd t determine the surce authenticity f the received data (e.g., TESLA [34]). This allws a nde t crrectly determine the rate at which it receives multicast data.

8 C. Rute Discvery BSMR's rute discvery allws a nde that wants t jin a multicast grup t find a rute t the multicast tree. The prtcl fllws the typical rute request/rute requestlrute reply prcedure used by n-demand ruting prtcls with several differences. T prevent utsiders frm interfering, all rute discvery messages are authenticated using the public key crrespnding t the netwrk certificate. Only grup authenticated ndes can initiate rute requests and the grup certificate is required in each request. Tree ndes use the tree tken t prve their current tree status. Several mechanisms are used t address internal attackers: (a) bth rute request and rute reply are flded in rder t ensure that, if an adversarial-free path exists, it will be fund; (b) the path selectin relies n the weights list carried in the respnse fld and allws the requester t select a nn-adversarial path; (c) the prpagatin f weights and path accumulatin is perfrmed using an nin-like signing t prevent frwarding ndes frm mdifying the path carried in the respnse. The requesting nde bradcasts a rute request (RREQ) message that includes the nde identifier and its weight list, the multicast grup identifier, the last knwn grup sequence number, and a request sequence number. The RREQ message is flded in the netwrk until it reaches a tree nde that has a grup sequence number at least as great as that in the RREQ. Only new requests are prcessed by intermediate ndes. When a tree nde receives fr the first time a RREQ frm a requester and the nde's grup sequence number is at least as great as that cntained in the RREQ, it initiates a respnse. The nde bradcasts a rute reply (RREP) message that includes that nde identifier, its recrded grup sequence number, the requester's identifier, a respnse sequence number, the grup identifier and the weight list frm the request message. T prve its current tree nde status, the nde als includes in the respnse the current tree tken, encrypted with the requester's public key. The RREP message is flded in the netwrk until it reaches the requester, using the fllwing weighted Jld fld suppressin mechanism. Tree ndes with a grup sequence number at least as great as that in the RREP RREiP ignre RREP messages. Otherwise, a nde cmputes the ttal path weight by summing the weight f all the links n the specified path frm the multicast tree t itself. If the ttal weight is less than any previusly frwarded matching respnse (same requester, multicast grup and respnse sequence number), and all the signatures accumulated n the reply are valid, the nde appends its identifier t the end f the message, signs the entire message and rebradcasts it. As the RREP message prpagates acrss the netwrk, ndes establish the frward rute by setting pinters t the nde frm which the RREP was received. Althugh several tree ndes may initiate the respnse fld, the weighted fld suppressin mechanism insures the cmmunicatin verhead is equivalent t nly ne fld. When the requester receives a respnse, it perfrms the same cmputatin as an intermediate nde during the respnse prpagatin. The requester updates its infrmatin upn receipt f a valid respnse that cntains a better path accrding t ur reliability metric. D. Multicast Rute Activatin The requester signs and unicasts n the selected rute a multicast activatin (MACT) message that includes its identifier, the grup identifier and the sequence number used in the rute request phase. An intermediate nde n the rute checks if the signature n MACT is valid and if MACT cntains the same sequence number as the ne in the riginal RREQ message. The nde then adds t its list f tree neighbrs the previus nde and the next nde n the rute as dwnstream and upstream neighbrs, respectively, and sends the MACT message alng the frward rute. The requester and the ndes that received the MACT message culd be prevented frm being grafted t the tree by an adversarial nde, selected n the frward rute, which drps the MACT message. T mitigate the attack, these ndes will start a waiting t cnnect timer (WTC-Timer) (WTC_Timer) upn whse expiratin ndes islate a faulty link and initiate Rute Discvery (Event 3 f Sec. V-F). The timers are set t expire

9 after a value prprtinal t a nde's hp distance t the tree, in the hpe that the ndes clser t the tree will succeed in aviding the adversarial nde and will manage t cnnect t the tree. After a nde becmes aware f its expected receiving data rate, it cancels its WTC_Timer WTC-Timer and behaves as described in Sec. V-F. E. Multicast Tree Maintenance The tree maintenance phase ensures the crrect peratin f the prtcl when cnfrnted with events such as pruning, link breakage and nde partitining. Ruting messages exchanged by tree neighbrs, such as pruning messages (described in Sec. III-B) 111-B) are authenticated using the pairwise keys shared between tree neighbrs. If a malicius nde prunes itself even if it has a subtree belw it, the hnest ndes in this subtree will recnnect t the tree fllwing the prcedure described in Sec. V-F. The link repair prcedure is initiated by ndes that detect a brken link and is similar with Rute Discvery. The grup leader peridically bradcast in the entire netwrk a signed Grup Hell message that cntains the current grup sequence number, the tree tken authenticatr and the hp cunt anchr (described in Sec. V-B). F: F. Selective Data Frwarding Detectin The surce peridically signs and sends in the tree a multicast rate (MRATE) message that cntains its data transmissin rate P. p. As this message prpagates in the multicast tree, ndes may add their perceived transmissin rate t it. The infrmatin in the MRATE message allws ndes t detect if tree ancestrs perfrm selective data frwarding attacks. Depending n whether their perceived rate is within acceptable limits f the rate in the MRATE message, ndes alternate between tw states. The initial state f a nde is Discnnected; After it jins the multicast grup and becmes aware f its expected receiving data rate, the nde switches t the Cnnected state. Upn detecting a selective data frwarding attack, the nde switches back t the Discnnected state. A netwrk perating nrmally exhibits sme amunt f natural "lss", which may cause the rate perceived by a nde t be smaller than the rate perceived by its tree parent. This natural rate decrease is cumulative as data travels further away frm the surce. We define a threshld 6b as the upper bund fr the tlerable lss rate n a single link. If a nde's perceived rate is smaller than the last recrded rate in MRATE by mre than 6, b, the nde cncatenates its identifier and its rate t MRATE and signs the entire message befre frwarding it. These added rates serve as prfs that ndes which previusly frwarded the MRATE message did nt perceive lsses much larger than natural lsses. In rder t prevent a malicius nde frm intrducing a rate decrease significantly larger than 6, b, we use anther threshld,6, A > 6. b. Upn receiving an MRATE NlRATE message, each nde first checks if the difference between the last rate in MRATE and the nde's perceived rate is greater than,6,. A. If s, this indicates that there exists at least an adversarial nde in between this nde and the nde that added the last rate t MRATE. The first hnest nde that ntices a difference larger than,6, A incriminates the link t its tree parent as faulty (by using an multiplicative weight increase scheme) and assumes respnsibility fr finding a new rute t the tree. The ndes in the subtree belw this nde will ntice there is a "gap" greater than,6, A between the rates included in MRATE; They will defer taking any actin t islate the faulty link fr an amunt f time prprtinal t the distance frm the nde that already started the repair prcedure, in the hpe that the ndes clser t the faulty link will succeed in islating it. Upn detecting that the expected data packet rate has been restred, ndes cancel the repair prcedure. Figures 1, 2 and 3 describe hw a Cnnected nde reacts t the fllwing events, respectively: (1) receipt f an MRATE message, (2) timeut f the MRATE-Timer, MRATE_Timer, and (3) timeut f the WTC_Timer. WTC-Timer. Pnde pnde dentes the rate at which the nde receives packets frm its tree parent. Tree ndes expect t peridically receive MRATE messages, therwise the MRATE-Timer MRATE_Timer will expire. Nte that each tree nde stres the latest received MRATE message and uses it t re-initiate the prpagatin

10 Fig. 1: I: receipt f MRATE = (p, (P, (idl,pl), 1, PI),""...,(idk,pk)) Pk)) 1. I. if this is the first MRATE message received then 2. switch t Cnnected state 3. cancel WTC-Timer WTC_Timer 4. stre MRATE message and cancel MRATE-Timer MRATE_Timer 5. if state = Cnnected and WTC-Timer WTC_Timer # =I PENDING then 6. if MRATE cntains a "gap" larger than A D then 7. start WTC-Timer WTC_Timer timer 8. frward MRATE 9. return. else if WTC-Timer WTC_Timer = =PENDING then 11. II. if MRATE cntains a "gap" larger than A D then 12. MRATE = cat-and-sign(mrate, cat_and_sign(mrate, (idnde, ' pnde)) Pnde» 13. frward MRATE 14. return 15. else 16. cancel WTC-Timer WTC_Timer 17. switch t Cnnected state 18. if pk Pk - pnde Pnde > A D then 19. MRATE = cat-and-sign(mrate, cacand_sign(mrate, (idnde, pnde)) Pnde». if WTC_Timer WTC-Timer = =PENDING then 21. cancel WTC_timer WTC-timer 22. switch t Discnnected state 23. increase weight f the link t the parent 24. initiate Rute Discvery 25. else if pk Pk - pnde Pnde > b 0 then 26. MRATE = cat-and-sign(mrate, cacand_sign(mrate, (idnde, pnde)) Pnde» 27. frward MRATE message 28. start MRATE-Timer Timer Fig. 2: timeut f MRATE-Timer Timer I. 1. if state = Cnnected and WTC_Timer WTC-Timer =I # PENDING then 2. retrieve stred MRATE = (p, (idl, l, pl), pi),....., (idk, pk)) Pk)) 3. if pk Pk - pnde Pnde > D A then 4. MRATE = cacand_sign(mrate, cat-and-sign(mrate, (idnde, ' pnde)) Pnde» 5. switch t Discnnected state 6. increase weight f the link t the parent 7. initiate Rute Discvery 8. else if pk Pk - pnde Pnde > 0b then 9. MRATE =cacand_sign(mrate, = cat-and-sign(mrate, (idnde, ' pnde)) Pnde» IO.. frward MRATE message Fig. 3: timeut f WTC-Timer Timer 1. switch t Discnnected state 2. increase weight f the link t the parent 3. initiate Rute Discvery f MRATE if MRATE_Timer MRATE-Timer expires. Als, when MRATE_Timer MRATE-Timer expires a nde cmpares its perceived rate with the expected rate frm the stred MRATE message. VI. EXPERIMENTAL RESULTS T the best f ur knwledge, the nly security mechanism prpsed n-demand multicast prtcls is the authenticatin framewrk by Ry et al. [17], t which we refer as A-MAODY. A-MAODV. Althugh A-MAODV withstands several external attacks targeted against the creatin and maintenance f the multicast tree, it des nt prvide additinal resilience against Byzantine attacks. In this sectin, we study the effect f several Byzantine attacks n the perfrmance f A-MAODV and we simulate the same attacks against BSMR in rder t shw its effectiveness in mitigating the attacks. Implementatin. We implemented BSMR using the ns2 simulatr [35], starting frm an MAODV implementatin [36]. We assumed the prtcl uses RSA [37] with l024-bit 24-bit keys fr public key peratins,

11 AES [38] with 128-bit keys fr symmetric encryptins and HMAC [39] with SHAI SHAl as the message authenticatin cde. The values used fr 15 6 and 6A were % and % f the surce's rate, respectively. We develped a prtcl-independent Byzantine attack simulatin mdule fr ns2. A. Experimental Methdlgy T capture a prtcl's effectiveness in delivering data t the multicast grup, we used as a perfrmance metric the packet delivery rati (PDR), defined as: PT PDR= = - P, Ps N. N where P, P r is the number f data packets received by multicast grup members, P, P s is the number f data packets sent by the surce and N is the size f the grup. Because external attacks can be prevented using the authenticatin framewrk described in Sec. V-B, in this paper we fcus n the fllwing three Byzantine attacks: -black hle attack: This is a selective data frwarding attack, in which adversaries nly frward ruting cntrl packets, while drpping all data packets. -wrmhle attack: Tw clluding adversaries cperate by tunneling packets between each ther in rder t create a shrtcut (r wrmhle) in the netwrk. The adversaries use the lw cst appearance f the wrmhle in rder t increase the prbability f being selected n paths; Once selected n a path, they attempt t disrupt data delivery by executing a black hle attack. -fld rushing attack: The attack explits the fld duplicate suppressin technique used by many wireless ruting prtcls. By "rushing" an authenticated fld thrugh the netwrk befre the fld traveling thrugh a legitimate rute, a Byzantine adversary ends up cntrlling many rutes. The attack can be implemented by simply ignring the small randmized delays which are nrmally required t reduce the number f cllisins. Fld rushing can be used t increase the effectiveness f a black hle r wrmhle attack. In rder t quantify the impact f adversarial psitining, we cnsider the fllwing scenaris: -randm placement: adversaries are placed randmly in the simulatin are -strategic placement: adversarial placement is as fllws: black hle attack: adversaries are placed strategically arund the multicast surce, equidistant n a circle with radius f 0 meters. wrmhle attack: Given k adversaries, ne adversary is placed near the surce, at crdinates (650,650). The ther k - 1 adversaries are placed thrughut the simulatin area s that the areas cvered by their transmissin range verlap as little as pssible. Each ne f the k - 1 adversaries is cnnected via a wrmhle tunnel t the adversary placed near the surce, thus creating k - 1 wrmhles. In Fig. 4, we illustrate the strategic adversarial placement fr the black hle attack (in the presence f 6 adversaries) and fr the wrmhle attack (in the presence f 7 adversaries). T study the influence f whether the adversaries explicitly jin the multicast grup and the rder f jining, we cnsider tw scenaris: -N NJOIN: adversarial ndes d nt jin the grup; -JOIN: adversarial ndes explicitly jin the grup befre any f the hnest members jin. The adversaries are cnsidered grup members in the frmula fr PDR. We chse these test case scenaris in rder t study the impact f the attacks under a light set f cnditins (adversaries are placed randmly, r they d nt explicitly jin the multicast grup) and under a mre extreme set f cnditins (adversaries are placed strategically, r they jin the grup befre hnest ndes d).

12 " I....,,,,, 1 +-''''=--::::'OO=--::::'''---="",----='''---C9=-OO-'''=-''---C''=-00--':"=-,,-,-J"OO O L 150 3m ara sm r 9m,050 tzaa 13s w,-----'-, 150 OW 3M _---.J MO ]50 m '.loo 50 OOIO 10 I I IYO lsoo (a) Black hle: 6 adversaries (b) Wrmhle: 7 adversaries Fig. 4: Examples f strategic adversarial placement fr the black hle and wrmhle attacks B. Simulatin Setup We perfrmed simulatins using the ns2 netwrk simulatr [35]. Ndes were set t use radis with 2 Mbps bandwidth and 250 meters nminal range. The simulated time was 600 secnds. We randmly placed 0 ndes within a 1500 by 1500 meter area and the multicast surce in the center f the area at crdinates (750,750). We experimented with different values fr grup size (, (l0, 30 and 50), fr number f adversaries (between 16% and 60% f the grup size) and fr "max" speed (0, 2 and 5 mls). Grup members jin the grup sequentially in the beginning f the simulatin, each ne at an interval f 3 secnds. Then the surce transmits multicast data fr 600 secnds at a rate f 5 packets per secnd, each packet f 256 bytes (resulting in lads between Kbps acrss all receivers). The members stay in the grup until the end f the simulatin. added t the netwrk replace hnest ndes, thus mdeling the capture f hnest ndes. We used a randm way-pint mbility mdel, but we incrprated changes t address cncerns raised in [40] abut the validity f the standard randm way-pint mdel. In particular, ndes select a speed unifrmly between % and 90% f the given maximum speed t achieve a mre steady mbility pattern and ensure that the average speed des nt drp drastically ver the curse f the simulatin. In additin, 300 secnds f mbility are generated befre the start f the simulatin s that ndes are already in mtin. This allws the average speed and nde distributin t stabilize befre the simulatin starts. Each data pint in the figures is averaged ver 30 different randm envirnments and ver all grup members. We evaluate the PDR as a functin f the number f adversaries, fr different grup sizes and levels f mbility. Each graph f sectins VI-C and VI-D illustrates the effect f the attacks with and withut fld rushing. C. Attack Resilience: The black hle attack a) Impact f Adversarial Placement: Figures 5 and 6 shw the results fr randm and strategic adversarial placement, respectively. Fr randm placement we see that, fr the same grup size, the PDR f A-MAODV decreases as the number f adversaries increases. Fr the same number f adversaries, it als decreases as we increase the grup size. Hwever, randm adversarial placement causes the number f grup members in the subtree belw an adversary t be lw; Thus a relatively large number f adversaries is needed t cause a significant disruptin (e.g., 30 adversaries fr a grup f size 50 can cause the PDR t drp belw 50%). In the presence f fld rushing, the PDR decreases further because adversaries actively try t get selected themselves as part f the tree. The impact f fld rushing decreases as the grup size and the ndal speed increase. We ntice that, fr A-MAODV, increasing the ndal speed des nt have a negative effect n the PDR; On the cntrary, at higher speeds we even see a slight increase in PDR. The effect f link breaks due t 1] la,i '" ""'.'

13 1 +A-MAODV ++ A-MAODV-rush 4 BSMR 4% BSMR-rush 1 0 _" c '" 60,. i=' 50 C 40 ti 3D _" '",,+. i='. 50 C 40 _ BO,. ; rj '" 60,. i='. 50 C i i (a) members; 0 mls m/s (b) members; 2 m/s d s (c) members; 5 mls m/s ti 3D ':J i 0,... -" c 70 '" 60 i='. 50 C Q.,. " 'ij '" 70 '" 60 i='. 50 C 40 3D Q. +. _ BO e 70 '" 60 i='. 50 C 40 ]I 30 Q., ; O i O i (d) 30 members; 0 mls m/s (e) 30 members; 2 m/s mls (f) (f) 30 members; 5 mls m/s -" e 70 9O.=::::::::iI. It: ".f 50 C 40 0; g 30 Q. O i ", O i O j (g) members; 0 mls m/s (h) 50 members; 2 m/s d s (i) (i) members; 5 mls m/s Fig. 5: 5: Black hle attack and fld rushing cmbined with black hle: Randm placement (NJOIN) mbility is is cmpensated by the fact that grup members get a chance t t recnnect t t the multicast tree in in a different psitin, pssibly cnnected t t the surce thrugh an an adversarial-free path. Fr the same reasn, the effect f fld rushing is is diminished as as the ndal speed increases. BSMR is is almst unaffected by the black hle attack (see Fig. 5). The PDR drps by by less than % even in in the presence f adversaries. In In additin, the influence f fld rushing is is unnticeable. This shws the effectiveness against fld rushing f BSMR's strategy, which includes the prcessing f f all all respnse fld duplicates and the metric capturing past behavir f f adversarial ndes. Mbility causes a slight PDR decrease, which is is natural because higher speeds will cause mre link breaks.

14 [ft-a-maodv 1 +A-MAODV -B-A-MAODV-rush ++A-MAODV-rush... t BSMR -B-BSMR=ruSilJ +BSMR-rush 1 0,. 0,.. : - " 70 a: Q; Q 40 w tf 30 _e... '0' 70+ '\: C-"..., 60 t...,,""'-:...,. 50 Q; Q 40 + ""'...,,""6 30 _ ==, 70+,,, a: 60,... ""'''. 50 Q; 40j..... Q ti 30 r:. -l-----_---_-----i '0 -l-----_---_------i -l _------i (a) members; 0 m1s m/s (b) members; 2 m1s m/s (c) members; 5 m1s m/s _ 80 '" 70 a: ". 50 Q; Q a. j....'..,=:=::::=:::-4j... _ l0:r.: , ; a: Q; Q 40 ti 30 r:. I ==='P +... _ 80 ""; 70 0, a: Q; Q 40 ]I i i O---_----- Adversanes (d) 30 members; 0 m/s (e) 30 members; 2 m1s m/s (f) 30 members; 5 m1s m/s e " c c a: 60 a: w Q; Q; Q 40 Q 40 Q "" u '0 "" 30 g a. Q. a ' '0 30 (g) 50 members; 0 m1s m/s (h) 50 members; 2 m1s m/s (i) 50 members; 5 m1s m/s Fig. 6: Black hle attack and fld rushing cmbined with black hle: Strategic placement (NJOIN) In the previus experiments, the adversaries were randmly placed. Fig. 6 shws the results when adversaries are strategically psitined as described in Sectin VI-A. Fr A-MAODV we ntice a drastic drp in the PDR. Fr example, at mis, 0 d when the grup size is 30, nly 5 adversaries (representing 16% f the grup size) are able t reduce the PDR t 25% by executing the black hle attack with fld rushing. This is a direct cnsequence f the fact that an adversary is nw selected in the tree clser t the rt and the subtree belw it may ptentially cntain many grup members. Fr the same reasn, the negative effect f the fld rushing attack is nw emphasized when cmpared t the randm placement case. We cnclude that strategic adversarial psitining has a crippling effect n the perfrmance f A-MAODY. A-MAODV.

15 [-A-A-MAODV A A-MAODV -A-MAODV-rush :::::: A MAODV rush -.-BSMR +- -II- BSMR-rush BSMR.ruh <"' ---(.-- ideal] idea p -- L _---.._ _----_._ _ 8D C ; 70 0:: 6D Q 40 3D _ 8D _ 8D, $ '";,g 70 "" ;-,g 70 "" 0:: 1 8D SO 0:: 1 6D ; s 50 Q$ i% 30 3D 5 u 30 L I ; 0 0 f Advenaries 1 (a) hnest members; 0 mls m/s (b) hnest members; 2 mls m/s (c) hnest members; 5 mls m/s D 80 C C 8D :: 6D 0::.0 0:: Q 40 Q 40 Q 40 g 30 ti Advenaries Advenaries (d) 30 hnest members; 0 mls m/s (e) 30 hnest members; 2 rn/s m/s (f) (0 30 hnest members; 5 mls m/s 0Jik..., :: C _ 00 C.e 70 1; 0:: C j. 90 _ '0 c 70 0:: '0. 50 C 40 I 3D, _---_----; -I _-- ; Advenaries (g) 50 hnest members; 0 mls m/s (h) 50 hnest members; 2 mls m/s (i) 50 hnest members; 5 mls m/s Fig. 7: Black hle attack and fld rushing cmbined with black hle: Randm placement (JOIN) On the cntrary, the effect f strategic adversarial psitining n BSMR is minr (Fig. 6). Like fr randm placement, the PDR drps by less than % even in the presence f adversaries, at lw ndal speeds. When mre adversaries are present, we see a slightly larger PDR decrease because there are less available hnest ndes left in the netwrk t serve as intermediaries fr the grup members. The resilience f BSMR t attacks that therwise have a devastating effect n A-MAODV validates the effectiveness f BSMR's apprach. b) Impact fexplicit Jin and Jin Order: T analyze the impact f explicit jin f adversaries t the multicast grup (JOIN), when cmpared t the NJOIN case, we lk again at the cases where adversaries

16 1 + A-MAODV + A-MAODV-rush + BSM R BSMR-rush -:?+- ideal 1 0'-..., "" , 0 70 'g 'X, II: SO II: 60, II: ; 0; 0; Q Q 40 Q J ti 30 ti X (a) hnest members; 0 m/s (b) hnest members; 2 rn/s m/s (c) hnest members; 5 m/s C "" " II: 80 II: 80 II: ; 0; 0; Q 40 Q 40 Q ti ". (d) 30 hnest members; 0 rn/s m/s (e) 30 hnest members; 2 rn/s m/s (f) 30 hnest members; 5 rn/s m/s 90 _ 80 " 70 0"""......, II: SO. 50 0; Q f \:... : _00 "" 70 II: ; Q 40 ti :1 j... 0),0 : 90 _ 00 '0' 70 II: ; Q 40 :: 30 Ii!,..... I"' : : (g) 50 hnest members; 0 m/s (h) 50 hnest members; 2 rn/s m/s (i) 50 hnest members; 5 m/s Fig. 8: Black hle attack and fld rushing cmbined with black hle: Strategic placement (JOIN) are randmly and strategically placed (Figures 7 and 8, respectively). Each figure als shws the ideal PDR (labeled ideal), which wuld be btained if everyne f the hnest grup members wuld receive all the packets sent by the surce. The effectiveness f an attack shuld be interpreted as the difference diflerence between the ideal line and a prtcl's PDR line. An attack is effective if this difference increases as the number f adversarial ndes is increased; On the ther hand, a prtcl resilient t attack appears as a line that remains parallel t the ideal line. Fr randm adversarial placement in Fig. 7, just like in the NJOIN case, we als see a decrease in

17 PDR as the number f adversaries increases. In additin, we see a majr difference frm the NJOIN case: When the adversaries explicitly jin the grup befre the hnest ndes jin, the impact f fld rushing is minimal because the adversaries are already part f the grup and rushing cntrl packets des nt prvide any additinal benefit. On the cntrary, in this case fld rushing may actually imprve the PDR % because, by rushing cntrl packets, adversaries may help legitimate ndes t find rutes faster. Als, in MAODV a jining nde activates a rute nly after waiting an interval f time, during which it may receive several rute reply messages. At the end f this interval, it selects the best rute accrding t a hp cunt metric. Thus, even if a adversarial tree nde managed t rush its rute reply, it may get verwritten by a legitimate rute reply with a better hp metric. A drastic drp in the PDR is bserved fr A-MAODV when adversaries are placed strategically (see Fig. 8). We cnclude that strategic psitining has a crippling effect n the perfrmance f A-MAODV even mre when adversaries explicitly jin the multicast grup. Fr bth randm and strategic adversarial placement, BSMR is barely affected by the attacks: In mst cases the PDR line remains almst parallel t the ideal line, which shws little degradatin ccurs as the number f adversaries increases. The impact f the attacks n BSMR increases slightly when a large number f adversaries have jined the grup, because there are less available hnest ndes left in the netwrk t serve as intermediaries fr hnest grup members. We cnclude that BSMR's strategy is effective in the JOIN case as well. D. Attack Resilience: The wrmhle attack c) Impact f Adversarial Placement: Figures 9 and shw the results fr randm and strategic adversarial placement, respectively. Unlike in the experiments fr the black hle attack, where the maxi- maximum number f adversaries was varied prprtinally with the grup size, fr the wrmhle attack nly up t 12 adversaries (frming 6 wrmhles) and up t 7 adversaries (frming 6 wrmhles) were used fr the randm and strategic placement cases, respectively, regardless f the grup size. When adversaries are placed randmly, the delivery rati f A-MAODV is affected by an increase in the number f adversaries, as expected. In general, we ntice that the wrmhle attack causes mre damage than the black hle attack because less adversarial ndes are needed t decrease the delivery rati by the same amunt. As an example, fr speed f 0 mfs m/s and grup size f 30 members, nly ndes frming wrmhles can reduce the PDR t 60%; On the ther hand, adversaries executing the black hle attack are needed t reduce the PDR by the same amunt. Fld rushing has a nticeable impact especially fr small grup sizes and fr lw mbility levels. Increasing the ndal speed reduces slightly the impact f the wrmhle and fld rushing attacks. The explanatin is similar with that fr the black hle attack: Mbility causes link breaks and hnest ndes reattach themselves in different psitins in the multicast tree, pssibly cnnected t the surce thrugh nn-adversarial paths. The wrmhle attack has a minr effect n the delivery rati f BSMR and the additin f fld rushing has n effect. The PDR drps by less than % fr all simulated scenaris, even at higher levels f mbility and remains abve 90% fr mst simulated scenaris. This cnfirms the effectiveness f BSMR's strategy against the wrmhle attack. Fig. shws the results when adversaries are strategically placed and frm wrmhles as described in Sectin VI-A. Because ne end f each wrmhle is near the surce, the ther end f the wrmhle will present attractive, shrt rutes t the ndes that are in its transmissin range. This has a serius impact n the delivery rati f A-MAODV: When 7 adversaries are present, frming 6 wrmhles, they cver almst cmpletely the entire simulatin are Thus, every hnest nde that is nrmally mre than three hps away frm the surce will be tricked int cnnecting t the multicast tree thrugh ne f the adversarial ndes. When fld rushing is emplyed, even ndes that are nrmally less than three hps away frm the surce are als tricked int cnnecting thrugh adversarial ndes. This. has a serius impact n the PDR f A-MAODV, which drps as lw as 23% (a drp f ver 70%) when fld rushing is present

18 [fra-maodv M A O D V :::-A-MAODV-rush t... BSMR -I)-... BSMR-rush I 1 '00 '00 'GG _80 _80 C C '" 60.i? 50.i?.i? C 40 C 40 C 40 g g 30 ll. ll. ll. bi,,,,,, P (a) lo members; 0 m/s rn/s (b) members; 2 m/s rn/s (c) members; 5 mls 0 ' I ' _ 80 _ 80 C C s C -, e 80 '" i?.i?. C 40 C 40 $ C 40 0 '" 30 '" i;l 30 '" i;l l:. 30 ll. P ll (d) 30 members; 0 rn/s m/s (e) 30 members; 2 rn/s m/s (f) (t) 30 members; 5 m/s rn/s l ' _80 80 C C C '" 60' '" Q C 'G C.. ]1 30 '" 30 '" U i;l i;l 30 ll. ll. ll (g) 50 members; 0 rn/s m/s (h) 50 members; 2 rn/s m/s (i) 50 members; 5 rn/s m/s Fig. 9: Wrmhle attack and fld rushing cmbined with wrmhle: Randm placement (NJOIN) fr 0 rnfs m/s and grup size is 3 We cnclude that a small number f strategically placed adversaries can cause cnsiderable damage when A-MAODV is used. Fr BSMR, strategic placement results in a slim increase in the delivery rati drp, when cmpared t the randm placement case. Fr mst simulated scenaris, the PDR drps by less than %. The slid perfrmance f BSMR serves as prf fr the rbustness f its apprach. d) Impact fexplicit Jin and Jin Order: T analyze the impact f explicit jin f adversaries t the multicast grup (JOIN), when cmpared t the NJOIN case, we lk again at the cases where adversaries are randmly and strategically placed (Figures 11 and 12, respectively). Eacn Eacli figure als shws the ideal

19 (+A-MAODV +A-MAODV-rush -A- BSMR +BSMR-rush I '"(; 70 1 \,,:. '" C 3D.OJ e:. 70+ \,...,... : a: C 40 " 3D O i +---_-_-_-_-----; Adversanes (a) members; 0 mls (b) members; 2 m/s d s (c) members; 5 m/s d s _80... 'Q' "C 40 " 30 : 'Ii a: C.0 0; ". 70 a: C.0 0,..., " -ti 3D J i _-_ ; O _----i (d) 30 members; 0 mls (e) 30 members; 2 m/s d s (f) 30 members; 5 m/s d s... -" 7DI', 1; a: 60 + (\;[...,'.-".. 50 C 40 i "&:::::= ",, ; 0; ti 30 '" :: If r ,------i ' 70' """':..... i ""t,.[ C.0 0; tj ",'''' : a: 60 : 50 C.0 0; _-_-_-_-_-----i (g) 50 members; 0 mls rnls (h) 50 members; 2 m/s d s (i) 50 members; 5 m/s d s Fig. : Wrmhle attack and fld rushing cmbined with wrmhle: Strategic placement (NJOIN) PDR (labeled ideal), which wuld be btained if everyne f the hnest grup members wuld receive all the packets sent by the surce. The effectiveness f an attack shuld be interpreted as the difference diflerence between the ideal line and a prtcl's PDR line. An attack is effective if this difference increases as the number f adversarial ndes is is increased; On the ther hand, a prtcl resilient t attack appears as a line that remains parallel t the ideal line. In general, fr bth randm and strategic adversarial placement, the effect f fld rushing n bth A-MAODV and BSMR is is unnticeable. Hwever, the reasns behind the immunity t fld rushing are different: BSMR prevents fld rushing attacks by using a reliability metric and by prcessing all fld

20 1DO '0 1 +LMAODV + A-MAODV-rush -A- BSMR 4- BSMR-rush.-,-..-.g ±A-MAODV --R-A MAODV-rush 8SM R '--l1li- 8SM R rush id; _.. 0;.;(;, 9D s '; 70 '" C 40 3D 2D 'T" _.. 0)(-, 9D s i 70 '" C.. g l i 0 -I---_ i 0 J..----_ Adversarles (a) members; 0 mls (b) members; 2 mls d s (c) members; 5 mls d s _ '" DO -;,0 +.,_ "'.:::K '" 60 + l'p"'... ". 50 C.. t 'ti 30 :::,.;=. _.. s -; 7D + "''iil ""'''''-..=c± ,'8;".. 50 c 40+ "'..""'''''iii'j g _.. 0k s -; 7D + "":... '" C 40 j 3D 2D J _--- J..---_ _--_--_ Adversarles (d) 30 members; 0 mls d s (e) 30 members; 2 mls d s (f) 30 members; 5 mls _ s '0 '" C 40 g " '"0' 70 '" C 40 i : +---_-_--_--i -1--_-_-_ D '" C 40 i3 30 -l---_-_-_-_-_---i (g) 50 members; 0 mls (h) 50 members; 2 mls d s (i) 50 members; 5 mls d s Fig. II: 11: Wrmhle attack and fld rushing cmbined with wrmhle: Randm placement (JOIN) duplicates, frwarding the nes with a better metric. As already explained in Sec. VI-C, A-MADDV's A-MAODV's immunity t fld rushing is simply an artefact that the adversarial ndes jin the multicast tree befre the hnest ndes d. When adversaries are placed randmly (Fig. 11), the impact f the wrmhle attack n A-MADDV A-MAODV is significant and increases as the number f wrmhles increases. BSMR's delivery rati line remains almst parallel with the ideal line, which means that little degradatin ccurs as mre adversaries jin the multicast grup. The devastating impact f the wrmhle attack n A-MADDV A-MAODV becmes bvius when adversaries are

21 I...,."...A-MAODV 1 +A-MAODV -B-A.MAODV-rush +A-MAODV-rush -+-BSMR -A- -llll-bsmr-rush +."::. ideal id 1 "XC : '" <: Gi c 40 m 30 Q. + """...C""......f " i , ; 70 '" 60.:!: 50 Gi c 40.. ti 3D 1 c-"lil,:..a, " ; '".0 <:-. 50 Gi c 40 m M 30 Q. j "'1'!!!l = " i (a) members; 0 m1s d s (b) members; 2 m1s d s (c) members; 5 m/s mls 70+\ :: "liii '" 60+ \... f <:. 50 Gi c ti 3D "'..., " i ""' A-,,""c :,. f _ '\ ::= ::::m '".0+ I'!l!:. f <:. 50 0; c 401 = """"l'!l\;"' f.. tj 30 " ; _ '" 80 <:-. 50 Gi C e:.="'... f.. tj 30 " ' " (d) 30 members; 0 m1s d s (e) 30 members; 2 m1s d s (f) 30 members; 5 m/s mls -X- -X- -X- -X '".0 50., Gi Gi " c 40 C 40 ti Q. " " 0 ::, f +, 1,,,,,, 1,,,,,, 0 0 J i (g) 50 members; 0 m1s d s (h) 50 members; 2 m1s d s (i) 50 members; 5 m1s d s Fig. 12: Wrmhle attack and fld rushing cmbined with wrmhle: Strategic placement (JOIN) strategically placed (Fig. 12). When 7 adversaries are present, the difference between the ideal line and A-MAODV's line increases by mre than 60% at 0 mis, mls, fr bth grup sizes f 30 and 50 (causing the PDR t drp t 15-%). BSMR is affected slightly mre than when adversaries are randmly placed, but in, all simulated scenaris the difference between the ideal line and BSMR's line des nt grw mre than 8%, even in the presence f 7 adversaries. This validates the effectiveness f BSMR's strategy in the JOIN case a well '" ; C 40.Ṁ 30 Q. "

22 A-MAODV- -S-A-MAODV BSMR-... BSMR-30 A-MAODV-NJOIN -S-A-MAODV-JOIN... BSMR-NJOIN... BSMR-JOIN , , =g 0 u 3l -.l!l 150 uell Q g QI..c 50, O ,------r------, i ,-----, , I I0 Speed (m/s) (rnls) ::I---. 'C 11i.=::=:t:=:::::::::::r---""""= g 80 \...-::=:::.:::= f -9: r4 -g 60 QI f 40 (a) Nn-adversarial scenari (grup size E {1O, {,30)) 30}) (b) Attack Scenari (grup size =30, = 2m/s, strategic placement) Fig. 13: BSMR verhead E. Prtcl Overhead We first cmpare the verhead incurred by A-MAODV and BSMR in a nn-adversarial scenari (Fig. 13(a)), fr grup sizes f and 30, under different levels f mbility. BSMR has higher verhead because bth rute request and rute reply are bradcast, and because f the extra MRATE packets bradcast peridically. BSMR's verhead due t duble flding in the rute discvery phase becmes mre nticeable especially at higher levels f mbility, when link breaks ccur mre ften. We then cmpare the verhead under adversarial cnditins. In Fig. l3(b) 13(b) we fcus n ne f the strngest studied attack scenaris: Black hle with strategic adversarial placement. Fr the NJOIN case, BSMR's additinal verhead cmpared t A-MAODV grws slwly as the number f adversaries increases (frm 40 mre packetslsec. packets/sec. fr 0 adversaries t 55 mre packetslsec. packets/sec. fr adversaries). Fr the JOIN case, the additinal verhead des nt grw as we increase the number f adversaries. This indicates that as the number f adversaries increases, BSMR incurs little extra verhead ver the nn-adversarial case. This is nt surprising: the bulk f the additinal verhead is caused by the initial rute discvery phase which leads t the creatin f the multicast tree with avidance f adversaries; Afterwards, BSMR's additinal verhead cnsists nly f peridical MRATE packets and f ccasinal rute discvery in case a link breaks due t mbility. VII. CONCLUSION In this paper we have discussed several aspects that make designing attack-resilient multicast ruting prtcls fr multi-hp wireless netwrks mre challenging when cmpared t their unicast cunterpart. A mre cmplex trust mdel and underlying structure fr the ruting prtcl make slutins tailred fr unicast settings nt applicable fr multicast prtcls. In the absence f defense mechanisms, Byzantine attacks can prevent multicast prtcls t achieve their design gals. We have prpsed BSMR, a ruting prtcl which relies n nvel general mechanisms t mitigate Byzantine attacks. BSMR identifies and avids adversarial links based n a reliability metric assciated with each link and capturing adversarial behavir. Our experimental results shw that BSMR's BSNLR's strategy is

23 effective against strng insider attacks such as black hles and fld rushing. We believe that this strategy can als be effective against wrmhle attacks and defer the experimental validatin fr future wrk. ACKNOWLEDGEMENTS The first authr wuld like t thank R5zvan Razvan Musaliu-E. Mus5liu-E. fr fruitful "cpy-rm" discussins in the early stages f this wrk. This wrk is supprted by Natinal Science Fundatin CyberTrust Award N The views expressed in this research are nt endrsed by the Natinal Science Fundatin. REFERENCES [1] [I] Y. Y B. KO K and N. H. Vaidya, Yaidya, "Flding-based gecasting prtcls fr mbile ad hc netwrks," Mb. Netw. Appl., vl. 7, n. 6, pp , 1480, 02. [2] R. Chandra, V. Ramasubramanian, and K. Birman, Binnan, "Annymus gssip: Imprving multicast reliability in mbile ad-hc netwrks," in Prc. f ICDCS, 01. [3] Y.-B. Y-B. KO K and N. H. Vaidya, Yaidya, "GeTORA: a prtcl fr gecasting in mbile ad hc netwrks," in Prc. f ICNP. IEEE Cmputer Sciety, 00, p. 24 [4] E. L. Madruga and J. J. Garcia-Luna-Aceves, "Scalable multicasting: the cre-assisted mesh prtcl," Mb. Netw. Appl., vl. 6, n. 2, pp , 01. [5] S. J. Lee, W. Su, and M. Gerla, Geria, "On-demand multicast ruting prtcl in multihp wireless mbile netwrks," Mb. Netw. Appl., vl. 7, n. 6, pp , 02. [6] E. Ryer and C. Perkins, "Multicast ad-hc n-demand distance vectr (MAODY) (MAODV) ruting," in Internet Draft, July 0 [7] J. G. Jetcheva and D. B. Jhnsn, "Adaptive demand-driven multicast ruting in multi-hp wireless ad hc netwrks." in Prc. f MbiHc, 01, pp [8] L. Lamprt, R. Shstak, and M. Pease, ''The "The Byzantine generals prblem," in Advances in Ultra-Dependable Distributed Systems. IEEE Cmputer Sciety Press, [Online]. Available: citeseer.nj.nec.cmllamprt82byzantine.html citeseer.nj.nec.cmhmprt82byzantine.html [9] P. Papadimitrats and Z. Haas, "Secure ruting fr mbile ad hc netwrks," in Prc. f CNDS, January 02, pp [] [l] Y.-C. Y-C. Hu, D. B. Jhnsn, and A. Perrig, "SEAD: Secure efficient distance vectr ruting fr mbile wireless ad hc netwrks," in Prc. f WMCSA, June 02. [II] I] Y.-C. Y-c. Hu, A. Perrig, and D. B. Jhnsn, "Ariadne: A secure n-demand ruting prtcl fr ad hc netwrks," in Prc. fmobicom, September 02. [12] [I21 K. Sanzgiri, B. Dahill, B. N. Levine, C. Shields, and E. Belding-Ryer, "A secure ruting prtcl fr ad hc netwrks," in Prc. f ICNP, Nvember 02. [13] S. Marti, T. Giuli, K. Lai, and M. Baker, "Mitigating ruting misbehavir in mbile ad hc netwrks," in Prc. fmobicom, August 0 [14] [I41 P. Papadimitrats and Z. Haas, "Secure data transmissin in mbile ad hc netwrks," in Prc. f WiSe, Wise, 03, pp [15] [I51 B. Awerbuch, D. Hlmer, C. Nita-Rtaru, and H. Rubens, "An n-demand secure ruting prtcl resilient t byzantine failures," in Prc. f WiSe'02. ACM Press, 02. [16] B. Awerbuch, R. Curtmla, D. Hlmer, C. Nita-Rtaru, and H. Rubens, "On the survivability f ruting prtcls in ad hc wireless netwrks," in Prc. f SecureCmm'05. IEEE, 05. [17] [I71 S. Ry, V. G. Addada, S. Setia, and S. Jajdia, "Securing MAODY: MAODV: Attacks and cuntermeasures," in Prc. 2nd IEEE Int'l. Cn! Cn$ SECON. IEEE, 05. [18] [I81 C. Zhang, B. DeCleene, J. Kurse, and D. Twsley, "Cmparisn f inter-area rekeying algrithms fr secure wireless grup cmmunicatins," Pelfrm. Perfrm. Eval., vl. 49, n. 1-4, 02. [19] [I91 K. H. Rhee, Y. H. Park, and G. Tsudik, "An architecture fr key management in hierarchical mbile ad hc netwrks," Jurnal f Cmmunicatin and Netwrks, vl. 6, n. 2, June 04. [] Y.-C. Y-c. Hu, A. Perrig, and D. B. Jhnsn, "Rushing attacks and defense in wireless ad hc netwrk ruting prtcls," in Prc. fwise. Wise. ACM,03. [21] --, -, "Packet leashes: A defense against wrmhle attacks in wireless ad hc netwrks," in Prc. f INFOCOM, April 03. [22] J. Erikssn, S. Krishnamurthy, and M. Falutss, ''Truelink: "Truelink: A practical cuntermeasure t the wrmhle attack in wireless netwrks," in Prc. ficnp'06. lcnp'06. IEEE, 06. [23] L. Hu and D. Evans, "Using directinal antennas t prevent wrmhle attacks," in Prc. fndss, 04. [24] D. Bruschi and E. Rsti, "Secure multicast in wireless netwrks f mbile hsts: prtcls and issues," Mbile Netwrks and Applicatins, vl. 7, n. 6, pp , 02. [25] T. Kaya, G. Lin, G. Nubir, and A. Yilmaz, "Secure multicast grups n ad hc netwrks," in Prc. f SASN'03. ACM Press, 03, pp [26] S. Zhu, S. Setia, S. Xu, and S. Jajdia, "Gkmpan: An efficient grup rekeying scheme fr secure multicast in ad-hc netwrks," in Prc. f Mbiquits'04. IEEE, 04, pp [27] L. Lazs and R. Pvendran, "Pwer prximity based key management fr secure multicast in ad hc netwrks," 05, acm Jurnal n Wireless Netwrks (WINET). [28] R. Balachandran, B. Ramamurthy, X. Zu, and N. Vindchandran, Yindchandran, "CRTDH: an efficient key agreement scheme fr secure grup cmmunicatins in wireless ad hc netwrks," in Prc. f lcc ICC 05, vl. 2, 05, pp