Indonesian Journal of Electrical Engineering and Computer Science V ol. 5, No . 1, J an uar y 2017, pp . 147 158 DOI: 10.11591/ijeecs .v5.i1.pp147-158 147 Attac ks and Secure Geographic Routing in Wireless Sensor Netw orks Y assine Sabri* and Najib El Kamoun Chouaib Doukkali Univ ersity , B .P: 20 . El J adida Morocco *Corresponding author , e-mail: sabr iy assino@gmail.com Abstract Due to open netw or k nature of wireless sensor netw or ks mak e them highly vulner ab le to a v ar iety of secur ity attac ks and easy target f or adv ersar ies , which ma y capture these nodes , analyz e and easily inser t f ak e route inf or mation. Wireless sensor netw or k is an emerging, cost eff ectiv e and unsuper vised solu tion f or collecting this inf or mation from the ph ysical w or ld and sending this inf or mation bac k to centr aliz ed author ity f or fur ther processing. GRPW (Geog r aphic Routing in connected wireless sensor netw or ks based on Multiple Sinks) is one of the basic routing protocols used f or Suppor ting Mobile Sinks in Wireless Sensor Netw or ks . GRPW , a geog r aphical routing protocol f or wireless sensor netw or ks , is based on an architecture par titioned b y logical le v els , on the othe r hand based on a m ultipoint rela ying flooding technique to reduce the n umber of topology broadcast. GRPW -MuS uses per iodic HELLO pac k ets to neighbor detection. As introduced in Ref erence [1, 2], the w or mhole attac k can f or m a ser ious threat in wi reless sensor netw or ks , especially against man y wireless sensor netw or ks routing protocols and location-based wireless secur ity systems . Here , a tr ust model to handle this attac k in GRPW is pro vide d called GRPW -MuS-s . Using OMNET++ sim ulation and the MiXiM fr ame w or k, results sho w that GRPW -MuS-s pro tocol only has v er y small f alse positiv es f or w or mhole detection dur ing the neighbor disco v er y process (less than GRPW). The a v er age energy usage at e ach node f or GRPW -MuS-s protocol dur ing the neighbor disco v er y and route disco v er y is v er y lo w than GRPW -MuS , which is m uch lo w er than the a v ailab le energy at each node . The cost analysis sho ws that GRPW -MuS-s protocol only needs small memor y usage at each node , which is suitab le f or the sensor netw or k. K e yw or ds: Wireless Sensor Netw or k (WSN), Routing, Secur ity , W or mhole attac k Cop yright c 2017 Institute of Ad v anced Engineering and Science . All rights reser ved. 1. Intr oduction Man y sensor netw or k applications , such as emergency response oper ations in a disaster en vironment or battlefield monitor ing, that r un in untr ustw or th y en vironments , require secure com- m unication and routing [3–5] to saf eguard against diff erent types of att ac ks . The attac ks such as b lac khole , w or mhole , misdirection, and rep la y [6, 7] can cause an e xisting route to be brok en or a ne w route to be pre v ented from being estab lished [8, 9]. There are se v er al e xamples of attac ks against routing in sensor netw or ks; a routing pac k et could be captu red and the inf or mation in the pac k et could be tampered with, or the adv ersar y might inser t a spur ious message in the sensor netw or k. T r aditional secur ity protocols are designed f or resource r ich machines to suppor t large computation and are not applicab le to sensor netw or ks due to resource limitations , ad hoc nature , and inter mittent connectivity . Man y sensor netw or k routi ng protocols ha v e been proposed, b ut v er y f e w of them ha v e been designed with secure routing as a goal. Secure routing protocols in sensor n etw or ks present challenges , which do not e xist in tr aditional netw or ks , such as no centr ally administered routers , lo w po w er , and small memor y nodes . A w or mhole is a tunnel which connects tw o remote nodes . In a w or mhole attac k [10], an attac k er receiv es pac k ets at one location in the netw or k, tunnels them to a remote location in the netw or k, an d t hen repla ys t hem into the netw or k from that location. A w or mhole attac k can be easily e x ecuted against routing in sensor netw or ks because it does not need to ph ysically compromise an y sensor node . Thu s , a w or mhole attac k poses a ser ious threat against routing in Receiv ed No v ember 12, 2016; Re vised December 19, 2016; Accepted December 30, 2016 Evaluation Warning : The document was created with Spire.PDF for Python.
148 ISSN: 2502-4752 the sensor netw or k as most of routing protocols do not ha v e an y mechanism to def end against it. A w or mhole attac k can cause the sensor nodes in the target area to b uild a route through an attac k er which can later tamper with the data messages , or selectiv ely f orw ard data messages . Ho w e v er , most of the researchers proposed solutions against a w or mhole attac k dur ing th e neigh- bor disco v er y process with the use of some special hardw are [11–13]. Moreo v er , their approach did not f ocus on ho w to b uild a secure route against the w or mhole attac k without an y additional special hardw are , such as a directional antenna, GPS , and a synchroniz ed cloc k. In a m ultihop wireless ad hoc Netw or k, mobile nodes cooper ate to f or m a Netw or k with- out using an y infr astr ucture such as access points or base stations . Instead, the mobile nodes f orw ard pac k ets f or each other , allo wing comm unication among nodes outside wireless tr ansmis- sion r ange . The nodes’ mobility and the fundamentally limited capacity of the wireless medium, together with wireless tr ansmission eff ects such as atten uation, m ultipath propagation, and in- terf erence , combine to create significant challenges f or routing protocols oper at ing in an ad hoc netw or k. Se v er al routing protocols f or wireless sensor netw or ks ha v e been de v eloped . GRPW - MuS w as proposed in [14], which belongs to the geog r aphical f or wireless sensor netw or ks class of routing prot ocols . GRPW -MuS is an optimization of the classical geog r aphical algor ithm tai- lored to the requirements of a mobile wireless . The k e y concept used in the protocol is m ultle v els rela ys (MLRs). MLRs are nodes selected in charge of f o rw ar ding broadcast messages dur ing the flooding process in each logical le v el. This technique substantially reduces the message o v erhead as compared with a classical flooding mechanism, where e v er y node retr ansmits each message when it receiv es the first cop y of the message . So this protocol is par ticular ly suitab le f or large and dense Netw or k. In Ref erence[15], attac ks on WSNs protocols gener ally f all into one of tw o f ollo w- ing categor ies: routing-disr upt ion attac ks and resource consumption attac ks . W or mhole attac k is classified into routing-disr uption attac ks . In the w or mhole attac k, an attac k er records pac k ets (or bits) at one location in the Netw or k, tunnels them to another location, and rela ys them there . Due to the nature of wireless tr ansmission, the attac k er can create a w or mhole e v en f or pac k ets not addressed to itself , since it can o v erhear them in wireless tr ansmission and tunnel them to the colluding attac k er at the opposite end of the w or mhole . The GRPW -MuS’ s neighbor disco v er y mechanisms rely hea vily on the reception of HELLO pac k ets to neighbor detection, so it is e xtremely vulner ab le to this att ac k. When an attac k er tun- nels through a w or mhole to a colluding attac k er near node B all HELLO pac k ets tr ansmitted b y node A , and lik e wise tunnels bac k to the first attac k er all HELLO pac k ets tr ansmitted b y B , then A and B will belie v e that the y are neighbors , which w ould cause the routing protocol to f ail to find routes when the y are not actually neighbors . Fur ther more , the attac k er is in visib le at higher la y- ers , unlik e a malicious node in a routing protocol, which can often easily be named, the presence of the w or mhole and the tw o colluding attac k ers at either endpoint of the w or mhole are not visib le in the route . The rest of the paper is organiz ed as f ollo ws: Section 2 discusses related w or k. Section 3 descr ibes the prob lem statement. Section 4 pro vides an o v er vie w of GRPW -MuS approach. Section 5 giv es a detailed descr iption of GRPW -MuS-S approach. Section 6 giv es cost analysis . Section 7 giv es perf or mance e v aluations , and Section 8 concludes the paper . 2. RELA TED W ORK AND B A CKGR OUND The impor tant approach f or pre v enting w or mhole attac ks is presented in Ref erences [16]. The main idea is that b y authenticating either an e xtremely precise timestamp or location inf or- mation combined with a loose timestam p , a receiv er can deter mine if the pac k et has tr a v ersed an unrealistic distance f or the specific netw or k technology used. T empor al leashes rely on e xtremely precise time synchronization and timestamps in each pac k et. But to c o nstr uct a tempor al leash, all nodes m ust ha v e tightly synchroniz ed cloc ks , which in f act are not easy to achie v e in MANET . Geog r aphical leashes rely on all nodes kno wing its o wn location and ha vin g loosely synchroniz ed cloc k. In that paper , the authors also point out that in some circumstances , bounding the distance betw een the sender and receiv er , cannot pre v ent w or mhole attac ks . Another method of pre v ent- ing w or mhole tac ks is kno wn as RF w ater mar king- -, which authenticates a wireless tr ansmission IJEECS V ol. 5, No . 1, J an uar y 2017 : 147 158 Evaluation Warning : The document was created with Spire.PDF for Python.
IJEECS ISSN: 2502-4752 149 b y modulating the RF w a v ef or m in a w a y kno wn only to author iz e node . But if the r adio band in which comm unications are taking place is kno wn, then an attac k er can attempt to tunnel the entire signal from one location to another . Some authors also propose using intr usion detection to han- dle the w or mhole at tac k, b ut intr usion detection is difficult to isolate the attac k er in a softw are-only approach. In [17] presented a gener al mechanism, called pac k et leashes , f or detecting and thus def ending against w or mhole attac ks in wireless netw or ks . The y presented tw o types of pac k et leashes: geog r aphic leashes and tempor al leashes . A geog r aphical leash ensures that the recip- ient of the pac k et is within a cer tain distance from the sender . T o co nstr uct a geog r aphical leash, each node m ust kno w its o wn location, and all nodes m ust ha v e loosely synchroniz ed cloc ks . A tempor al leash ensures that the pac k et has an upper bound on its lif etime , which restr icts the maxim um tr a v el distance since the pac k et can tr a v el as f ast as the speed of ligh t. T o constr uct a tempor al leash, all nodes m ust ha v e tightly synchroniz ed cloc ks . The disadv antage of using pac k et leashes is that the y require either location inf or mation f or each node or need t ight cloc k synchronization betw een the nodes . In [18] ha v e presented a solution that uses directional antennas b y mobile nodes to def end against w or mholes . Their assumption is that if there is no w or mhole attac k and if one node sends pac k ets in a giv en direction, then its neighbor will get that pac k et from the opposite direction. The neighbor ing nodes e xamine the directions of the receiv ed signals from each other with a shared witness . Only when the directions of both pairs match, the neighbor ing relation is confir med. The disadv antage is that each node is to be equip ped with the special hardw are called directional antenna, which is not alw a ys possib le . In [19] proposed a g r aph theoretic model f or char acter izing a w or mhole attac k and de- r iv ed the necessar y and sufficient conditions f or an y candidate solut ion to pre v ent w or mholes . In this approach, a small fr action of the nodes needs to be equipped with a GPS receiv er . In [20] proposed a mechanism, MDSV O W , to detect w or mholes in a sensor netw or k. MDS-V O W detects a w or mhole b y visualizing the anomalies introduced b y an attac k. The anomalies , which are caused b y the f ak e connections through the w or mhole , bend the reconstr ucted surf ace to pull the sensors that are f ar a w a y to each other . By detecting this bending f eature , the w or mhole is located and f ak e connections are identified. The disadv antage is that the message o v erhead is high because all of the sensors need to send their neighbor lists to the base station . In [21], the authors proposed tw o mechanisms based on h ypothesis testing f or detecting w or mholes in wireless sensor netw or ks . The first mechanism, called the neighbor n umber test (NNT), detects increases in the n umber of neighbors of the sensors due to ne w links created b y the w or mhole in the netw or k. The second mechanism, called the all distances test (ADT), detects decreases in the lengths of the shor test pat hs betw een all pairs of sensors , which are due to the shor tcut links created b y a w or mhole in the netw or k. Both m echanisms assume that the sensors send their neighbor lists to the base station and the base st ation r uns the algor ithms on the netw or k topology . The disadv antages are (1) the message o v erhead is high because all of the sensors need to send their neighbor lists to the base station and (2) the mechanisms can only detect the presence of a w or mhole , b ut the y do not pinpoint its e xact location. 3. Pr ob lem statement Recall, in a w or mhole attac k, an attac k er receiv es pac k ets at one location in the netw or k, tunnels them to another location, and retr ansmits them there into the netw or k. In the basic route disco v er y process , t he base station star ts the route disco v er y b y broadcasting a routing beacon. Each node which receiv es the routing beacon records the base station’ s identity as its parent. Then, it rebroadcasts the routing beacon. The algor ithm cont in ues recursiv ely with each node mar king the first node from whom it hears a route beacon to be its parent. The basic route disco v er y process f ails if an attac k er receiv es the routing beacon at one point in the netw or k, tunnels it to another point in the netw or k, and then repla ys it into the netw or k from that point. Since the rout ing beacon tunneled b y the w or mhole reaches the targeted area consider ab ly f aster than it nor mally w ould ha v e through the m ulti-hop routing, the nodes near the endpoint of the w or mhole Attac ks and Secure Geog r aphic Routing in Wireless ... (Y assine Sabr i) Evaluation Warning : The document was created with Spire.PDF for Python.
150 ISSN: 2502-4752 will create a large routing sub-tree in the targeted area with themselv es as the root. F or e x emple , the attac k er tunnels the routing beacon from M 1 to M 2 . The nodes in the target area b uild the route through the w or mhole loca ted betw een M 1 and M 2 . All the tr affic in the targeted area will be channeled through the w or mhole . If an attac k er perf or ms this tun neling honestly and reliab ly , no har m is done; the attac k er actually pro vides a useful ser vice in connecting the netw or k more efficiently [22]. Ho w e v er , the w or mhole puts the attac k er in a v er y po w erful position relativ e to other nodes in the netw or k. The attac k er discards r ather than f orw ard ing all the data pac k ets . Thereb y , it creates a per manent denial-of-ser vice attac k, where the base station cannot receiv e an y inf or mation from the target area. Also , the attac k er can e xploit the w or mhole to selectiv ely discard or modify cer tain data pac k ets . System assumption. W e assume that the sensor nodes after deplo yment are not mo v ab le . Each sensor node has the same en ergy at the star t. It has a unique identity (ID) and an initial k e y KI and the r andom function f . W e assume that the initial k e y KI is stored in the memor y , which can be er ased completely [23]. The sensor nodes comm unicate using RF (r adio frequency), so broadcast is the fundamental comm unication pr imitiv e [24]. T w o nodes within each other ’ s tr ansmission r ange are called one-hop neighbors . W e assume that comm unication channels are bidirectional [24], i.e . if a node y can receiv e a message from z , then it can also send a message to z . W e assume that the channel, based on MA C protocols [25], betw een the sensor nodes is reliab le . That is , the signals sent from diff erent sensor nodes across the same channel do not collide . 4. Security sc heme W e use an adaptation of the tr ust model [26] configured b y Marsh f or use in pure ad hoe Netw or ks . Marsh’ s model computes situational tr ust in agents based upon the gener al tr ust in the tr ustor and in t he impor tance and utility of the situation in which an agent finds itself . Gener al tr ust is basically the tr ust that one entity assigns another entity based upon all pre vious tr ans- actions in all situations . In our model each node ha v e a tr ust e v aluator which gathers data from the neighbor’ s e v ents in all states , filters it, assigns w eights to each e v ent and computes diff erent tr ust le v els based upon them. The tr ust e v aluator has three functions: tr ust der iv ation, quantifi- cation, and computation. At first, in GRPW -Mus the tr ust can come from the inf or mation about the successful tr ansmission of an y pac k et that is rela y ed b y the neighbor ing node , such as some ac kno wledgments . Second, the neighbor ing node’ s HELLO pac k et receiv ed on schedule can also conduce to the tr ust. These e v ents can be categor iz ed into data and control pac k et types , and in each e v ent there are tw o states: success and f ail, which record the n umber of successful e v ents and f ailed e v ents respectiv ely . In tr ust quantification process , w e represent tr ust from 1 to 1 sig- nifying a contin uous r ange from complete distr ust to complete tr ust. T r ust computation in v olv es an assignment of w eights to the e v ent that w ere monitored and quantified. W e use the contin uous r ange from 0 to 1 f or representing the significance of a cer tain e v ent from unimpor tant to most impor tant. The higher w eights represent the e v ent more impor tant. W e define the tr ust T to the neighbor ing node y b y the node x , and it is giv en b y the f ollo wing equation: T x ( y ) = n X i =1 [ W x ( i ) T x ( i )] (1) where W x ( i ) is the w eight of the i th tr ust categor y to x and T x ( i ) is the situational tr ust of x in the i th tr ust categor y . The n represents the n umber of categor y . F rom abo v e equation, w e can get the f ollo wing equations : C h = H S H F H S + H F f or H S + H F 6 = 0 el se C h = 0 (2) Negativ e v alues represent that more f ailed e v ents occur than successes . Hence , a v alue of 1 represents complete distr ust, a v alue of 0 implies a non-contr ib uting e v ent and a v alue of +1 means absolute tr ust in a par ticular e v ent. No w the node x can get the whole tr ust T to the neighbor ing node y . IJEECS V ol. 5, No . 1, J an uar y 2017 : 147 158 Evaluation Warning : The document was created with Spire.PDF for Python.
IJEECS ISSN: 2502-4752 151 T x ( y ) = W x ( C h ) T x ( C h ) + W x ( C d ) T x ( C d ) (3) 5. GRPW -MuS re vie w In this section w e will f ocus on introducing the GRPW -MuS Routing approach as this is the f oundation f or our w or k. F or a more elabor ate descr iption to GRPW -Mus please ref er to [14]. GRPW -Mus Is a geog r aphic routing protocol f or wireless sensor netw or ks f or m ultiple sink, Based on a par titioned topology in circular logic le v els ,each node can get its o wn location inf or mation either b y GPS or other location ser vices . The routing of data is inspired b y the pr in- ciple of w ater flo w in a w ashbasin b y creating the vir tual logic le v els as descr ibed in the figure 1 and 2 . After this logical netw or k reconstr uction ,each sink estab lishes its area based on the sink D S position. The routing of captured data be perf or med within each z one belonging to each node using the GRPW -Mus method f or each Area Sink . Lev el 0 Lev el 1 Lev el 2 Lev el 3 Lev el 4 SB ( sink ) Figure 1. Illustr ation of GRPW -MuS routing netw or k le v els Lev el 0 Lev el 1 Lev el 2 Lev el 3 Lev el 4 Source Designated Sink (DS) S I N K secondar y S I N K secondar y inter nal node Bac kbone area sink Area border noeud Area border noeud Figure 2. Illustr ation of GRPW -MuS routing netw or k le v els The procedure of GRPW -MuS is as f ollo ws . Ev er y node broadcasts HELLO messages that contain one-hop neighbor inf or mation per iodically . The TTL of HELLO messages is 1 , so the y should not f orw arded b y its neighbors . With the aid of HELLO messages , e v er y node obtains local Attac ks and Secure Geog r aphic Routing in Wireless ... (Y assine Sabr i) Evaluation Warning : The document was created with Spire.PDF for Python.
152 ISSN: 2502-4752 topology inf or mation. A node (also called D S ) chooses a subset of its neighbors to act as m ulti- point rela ying nodes f or it is based on the local Le v el topology inf or mation, which Le v el specified in the per iodic HELLO messages later . D S nodes perf or m tw o tasks: 1. when the Sink sends or f orw ards a broadcast pac k et, only its D S nodes among all its neigh- bors f orw ard the pac k et 2. the D S nodes per iodically broadcast its selector list . Thus e v er y node in the each le v el kno ws through which D S nodes e v er y other node could be reached. With each le v el’ s topology inf or mation stored and updated at e v er y node , a sh or test path from one node to e v er y other node could be computed with GRPW -Mus algor ithm, which goes along a ser ies of D S node . 6. Extension to GRPW -MuS The fr ame w or k of e xtension to GRPW -Mus is sho wn in Fig.3 Figure 3. F r ame w or k of e xtension to GRPW -MuS When the n ode receiv es a ne w sender’ s HELLO message , it will mak e tw o ne w records ¡node , positiv e , negativ e , e v ent¿ , to record separ ately the e v ent of this sender’ s HELLO mes- sage’ s coming in time or not, and the e v ent of data f orw arding successfully or not. Then in inf or- mation collection there are tw o tab les to record e v er y possib le neighbor’ s e v ents . These tab les are the inputs of tr ust calculation. By tr ust calculation, e v er y possib le neighbor will get a v alue which represents the probability of the neighb or relationship . The tuples neighbor , probability ¿ will be recorded in Neighbor Set. Some GRPW -MuS inf or mation repositor ies and pac k ets’ f or mat should be modified. When the node broadcasts the HELLO message , it contains its neighbor in- f or mation including the recommendation about the probability of neighbor relationships . F rom re- ceiving others HELLO messages , e v er y node obtains local topology inf or mation. When choosing MPR nodes , the node will tak e the nodCs recommendation as an impor tant f actor . When nodes e xchange the Hello messages which contain the inf or mation about the neighbor relationship’ s probability , e v er y node w ould get global topology inf or mation which can constr uct a w eighted di- rected g r aph. The w eight on the edge represents the e v aluation of edge star t point on the link e xistence betw een it self and the end point.Then from the w eighted directed g r aph of the global topology , w e can use Dijkst r a algor ithm to calculate the routing tab le . In this process , the proba- bility of the ”being a neighbor” is considered as the w eight. 7. P erf ormance e v aluations F or perf or mance analysis , w e ha v e sim ulated GRPW -MS-S protocol using OMNET++ discrete e v ent sim ulator [27]. As OMNET++ is not de v eloped f or the sensor netw or k, a sensor netw or k en vironment is created in OMNET++ with the installation of a MiXiM fr ame w or k patch [28]. In this sim ulation, w e sim ulate 1600 sensor nodes . The tr ansmi s sio n r ange f or each sensor node is 40 m . T r ansmit P o w er P t is the po w er with which the signal is tr ansmitted. The T r ansmit IJEECS V ol. 5, No . 1, J an uar y 2017 : 147 158 Evaluation Warning : The document was created with Spire.PDF for Python.
IJEECS ISSN: 2502-4752 153 P o w er P t decides the tr ansmission r ange f or the sensor node . T r ansmit P o w er (txP o w er) is the po w er consumed b y the tr ansceiv er to tr ansmit a data pac k et. Rece iv e P o w er (rxP o w er) is the po w er consumed to receiv e a data pac k et. w e can see that there are some f alse positiv es , which means that some nodes are mistak enly detected to be connected b y the w or mhole since t he y are actually close nodes . In this section, w e sim ulate the f alse positiv es under diff erent deplo yments and diff erent threshold s used . The pur pose of this sim ulation is to control the f alse positiv es to the minim um. W e design f our diff erent types of sensor deplo yment: 1. Random deplo yment within the g r id (RandomGr id): The whole sensor deplo yment area is divided into g r ids with only one sensor node f or each g r id. The position of the sensor node in the g r id is r andom. 2. Random deplo yment in the whole area (RandomArea): All sensor nodes are r andomly de- plo y ed in the whole deplo yment area. 3. Nor mal distr ib ution within the g r id (Nor malGr id): The whole sensor deplo yment area is di- vided into subareas , where each sub-area holds an equal n umber of sensor nodes . More specifically , let the total n umber of sensor nodes be N, the total n umber of divided g r ids be N g r id , then each g r id contains ( N = N g r id ) sensor nodes . Within each g r id, the sensor nodes are deplo y ed using the nor mal distr ib ution. 4. Nor mal distr ib ution in the whole area (Nor malArea): All sensor nodes are deplo y ed in the whole deplo yment area f ollo wing the nor mal distr ib ution. 10 20 30 40 2 4 6 8 10 Density 10 3 =m 2 F alse P ositiv e % RandomGr id RandomArea Nor malArea Nor malGr id Figure 4. F alse positiv e vs . density ( T h = 4 ). Figs . 5 and 4 descr ibe the relationship betw een the f alse positiv es and the density of the sensor netw or k with the diff erent types of sensor deplo yment under a specific threshold T h . W e find that when the threshold T h is equal to or less than 6 , all of th e f our deplo yments ha v e f alse positiv es t hat are less than 10% . F rom Fig 4, w e find that the Nor malGr id has higher f alse positiv es than other deplo yments . Moreo v er , the f alse positiv es f or Nor malGr id deplo yment in- creases when the density of the sensor nodes increases . This is because in Nor malGr id, when the density increases , each g r id co v ers more nodes . Since nodes in each g r id are deplo y ed with the nor mal distr ib ution, the nodes ha v e more chance to become close nodes in each g r id. This causes more f alse positiv es . W e cannot see m uch diff erence in the f alse positiv es f or the other deplo yments , which are RandomGr id, RandomArea, and Nor malArea. Their f alse positiv es are lo w (less than 10% ) if the threshold Th is belo w 12 . Moreo v er , w e find that the f alse positiv es are roughly the same when the density of the sensor netw or k increases . Th is is because in Random- Gr id/RandomArea, the nodes are r andomly deplo y ed. The nodes could be closer b ut the y are still Attac ks and Secure Geog r aphic Routing in Wireless ... (Y assine Sabr i) Evaluation Warning : The document was created with Spire.PDF for Python.
154 ISSN: 2502-4752 10 20 30 40 5 10 15 20 Density 10 3 =m 2 F alse P ositiv e % RandomGr id RandomArea Nor malArea Nor malGr id Figure 5. F alse positiv e vs . density ( T h = 10 ). not close enough according to so called close nodes . So w e cannot see the f alse positiv e g ro wing with increasing density . In Nor malArea, the nodes are deplo y ed with the nor mal distr ib ution in the area. When the density increases , most the nodes or iginally close are still close . Theref ore , the f alse positiv es do not g ro w with increasing density . F rom the abo v e analysis , to minimiz e the f alse positiv es , a good distr ib ution and a good threshold Th m ust be selected. T o k eep the f alse positiv es belo w 10% , the ideal distr ib ution can be RandomGr id, RandomArea, and Nor malArea with a maxim um threshold Th v alue of 12 . 50 100 150 5 10 Sim ulation Time (s) tr ansmission speed ( K b :s 1 ) GRPW -MS GRPW -MS-s Figure 6. W or mhole attac k analysis In the netw or k there are a set of attac king nodes which represents 20% of the netw or k nodes , which are A 1 and A z in the figure . F or e x emple , A 1 and A 2 , which are the tunnel’ s tw o ends , will e x ecute the w or mhole attac k. A 1 will tunnel all i t’ s hear ing HELLO pac k ets to A 2 , A 2 will also tunnel all the hear ing HELLO pac k ets to A 1 , then both of th em will repla y the HELLO pac k ets . W e sim ulate the or iginate GRPW -MS protocol and the re vised protocol GRPW -MS-S under the same condition. The results are sho wn in Fig. 6. F rom the figure , w e can see that the lo w er line is a z ero line which sim ulate the or iginate GRPW -MS protocol. The z ero means that No can not find a r ight route to send the pac k et , so Nu receiv es nothing . All these happened cases are caused b y w or mhole attac k ers making misbelie ving being its neighbor . Th e upper line is the result of sim ulating the re vised GRPW -MS-S protocol, w e can f ound at first node also can not find the r ight route , b ut after e v aluating some neighbor’ s tr ustiness , eache node star t to choose IJEECS V ol. 5, No . 1, J an uar y 2017 : 147 158 Evaluation Warning : The document was created with Spire.PDF for Python.
IJEECS ISSN: 2502-4752 155 another route to send the pac k et, after man y times tr ying and e v aluating, No finally find a stab le route to Nal, so in the figure it sho ws that the tr ansmitting r ate is going to k eep stab le with time , and after 20 s , it k eeps about 10 : 0 k b=s . 50 100 150 200 50 100 Sim ulation Time (s) The n umber of pac k ets on the w or mhole link W or mhole attac k Def end attac k Figure 7. The n umber of pac k ets on the w or mhole link against time f or GPR W -MS-S Finally , w e e v aluate the def ending eff ectiv eness after detecting the w or mhole attac k. When the w or mhole attac k is initiated, the surrounding pac k ets w ould tr ansf er from the or igi- nal route to this highquality w or mhole link. As sho wn in the Fig.7, the dot cur v e indicates that the n umber of pac k ets on the w or mh ole link dr amatically increases after the w or mhole attac k; when the def ending nodes begin to tak e def ensiv e measures , the square cur v e re v eals that the n umber of pac k ets on the w or mhole link g ro ws e xponentially . Gr adually , the w or mhole link be- comes congested and the metr ic of link decreases , which indicates that our algor ithm’ s def ense against w or mhole is eff ectiv e . Theref ore , when the nodes cond uct the neighbor disco v er y , the y will remo v e the malicious nodes from their respectiv e neighbor lists and the w or mhole link gets eliminated. 10 20 30 40 92 94 96 98 Density 10 3 =m 2 W or mhole detection r ate GRPW -MS GRPW -MS-s Figure 8. W or mhole detection r ate against density ( T h = 10 ). w e e v aluate the algor ithm’ s perf or mance on detecting w or mhole b y v ar ying the length of w or mhole link. Fig.8 re v eals the relationship betw een w or mhole detection r ate and the density . In Attac ks and Secure Geog r aphic Routing in Wireless ... (Y assine Sabr i) Evaluation Warning : The document was created with Spire.PDF for Python.
156 ISSN: 2502-4752 Fig.8, w e can find that tw o algor ithms both ha v e a high detection r ate . In GRPW -MS algor ithm, when density v ar ies from 5 to 20 , the detection r ate has a slight do wnw ard trend. When density contin ues to increase , the detection r ate le v els off , maintaining at about 0 : 955 . By contr ast, in our proposal, the detection r ate sho ws an upw ard trend with the increase of density . Moreo v er , when density is g reater than 10 , the detection r ate of our algor ithm GRPW -MS-S is higher than that of GRPW -MS . The reason is that the longer the w or mhole link is , the more hops the pac k ets ha v e to pass from source to destination if pac k ets are tr ansmitted through the nor mal link. But if there e xists a w or mhole link, the hops betw een source and destination w ould dr amatically decrease and thus mak e the w or mhole attac k eff ect m uch more significant. So , according to our algor ithm, w e can easily detect the w or mhole attac k and thus get a high detection r ate . 8. Conc lusions Because of the wireless medium’ s openness , e v er y node can hear the neighbor’ s r adio without being detected. When tw o or more malicious nodes constr uct one or more w or mholes , the y can destro y the entire Netw or k b y disr upting the routing proto col, especially to GRPW -Mus protocols . In this paper w e introduced a tr ust model to e v aluate the tr ustiness of ”a node is the neighbor” in GRPW -Mus protocol. F rom the tr ustiness calculating, the node can get the r ight route instead of choosing the route caused b y w or mhole attac k. This scheme can r un with no need f or netw or k synchronization and GPS de vices . But the scheme is based on tr ust e v aluation, which predicts the future e v ents b y collecting the past e v ents , so the tr ust e v aluated b y the node lags behind the attac ks . In future w or k, w e will w or k on ho w to secure the tr ustiness message tr ans- mission and ho w to get the recommended path in tr ust g r aph. W e also tak e the node’ s mobility into consider ation, because when the netw or k topology changing f ast, the route will change f ast, which means the tr ust model should k eep tr ac k with it. Ref erences [1] S . K. J angir and N. Hemr ajani, “Ev aluation of b lac k hole , w or mhole and sybil attac ks in mobile ad-hoc netw or ks , ” in Proceedings of the Second Inter national Conf erence on Inf or mation and Comm unication T echnology f or Competitiv e Str ategies , ser . ICTCS ’16. Ne w Y or k, NY , USA: A CM, 2016, pp . 74:1–74:6. [O nline]. A v ailab le: http://doi.acm.org/10.1145/2905055.2905133 [2] R. Mudgal and R. Gupta, “An efficient approach f or w or mhole detection in manet, ” in Proceedings of the Second Inter national Conf erence on Inf or mation and Comm unication T echnology f or Competitiv e Str ategies , ser . ICTCS ’16. Ne w Y or k, NY , USA: A CM, 2016, pp . 29:1–29:6. [Online]. A v ailab le: http://doi.acm.org/10.1145/2905055.2905235 [3] M. Cinque , A. Coronato , A. T esta, and C . Di Mar tino , “A sur v e y on resiliency assessment techniques f or wireless sensor netw or ks , ” in Proceedings of the 11th A CM Inter national Symposium on Mobility Management and Wireless Access , ser . MobiW ac ’13. Ne w Y or k, NY , USA: A CM, 2013, pp . 73–80. [Online]. A v ailab le: http://doi.acm.org/10.1145/2508222.2508235 [4] I. Ouaf aa, L. J alal, K. Salah-ddine , and E. H. Said, “The compar ison study of hier archical routing protocols f or ad-hoc and wireless sensor netw or ks: A liter ature sur v e y , ” in Proceedings of the The Inter national Conf erence on Engineer ing & MIS 2015 , ser . ICEMIS ’15. Ne w Y or k, NY , USA: A CM, 2015, pp . 32:1–32:8. [Online]. A v ailab le: http://doi.acm.org/10.1145/2832987.2833039 [5] K. S . Hamza and F . Amir , “Centr aliz ed cluster ing e v olutionar y algor ithms f or wireless sensor netw or ks , ” in Proceedings of the 10th Inter national Conf eren c e on Inf or matics and Systems , ser . INFOS ’16. Ne w Y or k, NY , USA: A CM, 2016, pp . 273–277. [Online]. A v ailab le: http://doi.acm.org/10.1145/2908446.2908493 IJEECS V ol. 5, No . 1, J an uar y 2017 : 147 158 Evaluation Warning : The document was created with Spire.PDF for Python.