Inter national J our nal of Electrical and Computer Engineering (IJECE) V ol. 6, No. 6, December 2016, pp. 3205 3216 ISSN: 2088-8708 3205 Fibonary Spray and W ait Routing in Delay T olerant Netw orks Priyanka Das 1 , Pr osenjit Cho wdhury 2 , Bikash P oudel 3 , and T anmay De 4 1 Department of Computer Science & Engineering, NSHM Kno wledge Campus Dur g apur , India 2 Softw are Engineer , o9 Solutions, Bang alore, India. 3 Softw are Engineer , Meru Netw orks, India 4 Department of Computer Science & Engineering, National Institute of T echnology Dur g apur , India Article Inf o Article history: Recei v ed Mar 2, 2016 Re vised Aug 18, 2016 Accepted Sep 5, 2016 K eyw ord: Delay T olerant Netw ork (DTN) fibonary latenc y deli v ery ratio ABSTRA CT Although there has been a tremendous rise in places being connected through the In- ternet or an y other netw ork protocol, there still lie areas, which remain out of reach due to v arious reasons. F or all such places the ans wer is a Delay T olerant Netw ork (DTN). A DTN is such a netw ork where there is no fix ed or predefined route for messages and no such guarantee whats oe v er of all messages being correctly routed. DTN can be considered as a superset of netw orks wherein other netw orks such as adhoc, mo- bile, v ehicular etc. form the subset. Therefore routing in DTN is a v ery chanc y af f air where one has to maximize on the present netw ork scenarios to get an y fruitful result other than depending on past information. Also protocols here need to be less com- ple x and not increase the already high nodal o v erhead. In this paper we propose a ne w approach, the Fibonary Spray and W ait, which does e xactly this. It forw ards copies of a message in a modified Binary Spray and W ait manner so that it performs well e v en in non independent and identically distrib uted node structure. W e ha v e supported our statements with mathematical as well as simulation analysis. Copyright c 2016 Institute of Advanced Engineering and Science . All rights r eserved. Corresponding A uthor: Name Priyanka Das Af filiation NSHM Kno wledge Campus Dur g apur Address Behind Clinic, Samdi Road, P .O. Rupnarayanpur Bazaar , W est Beng al-713386, India Phone +91 8900596423 Email priyanka.das.2206@gmail.com 1. INTR ODUCTION The idea for a Delay T olerant Netw ork to e xist, e v olv ed while conceptualizing the InterPla Netary (IPN) Internet project (1). The project’ s aim w as t o establish an ef ficient means of communication between planets and satellites k eeping in mind their enormous dif ference in distance as well as una v ailability of message routers (usually satellites) at all times. Hence the concept (and a need) for a ne w netw ork emer ged wherein messages w ould be passed around the netw ork based on scheduled contacts (as satellite contact periods are usually fix ed). On further research, it w as seen that not only interplanetary b ut man y ne w terrestrial netw orks too needed something more than what w as pro vided by the TCP/IP . The TCP/IP w orks on some assumptions (2) which are lik e inability to support long link delay , Necessity of end-to-end routing paths, data pack ets ha v e lo w time-to-li v e etc. Abiding by these strict prerequisites is not feasible for some ne w netw orks lik e T errestrial ci vilian netw orks (connecting mobile wireless de vices, including personal communicators, intelligent high- w ays, and remote Earth outposts), W ireless Military Ad-Hoc battlefield netw orks (connecting troops, aircraft and satellites), Sensor Netw orks (both on land and in w ater), W ar torn or Disas ter struck areas (lik e a place hit by some natural calamity whose connection with the outs ide w orld has disrupted resulting into a netw ork par - tition) etc. These netw orks on the other hand are characterised by Intermittent connecti vity , Po wer limitations, Netw ork partitions, Arbitrarily long periods of link disconnection, High error rates and l ar ge bidirectional data- J ournal Homepage: http://iaesjournal.com/online/inde x.php/IJECE DOI:  10.11591/ijece.v6i6.10361 Evaluation Warning : The document was created with Spire.PDF for Python.
3206 ISSN: 2088-8708 rate asymmetries. The answer to all these is the Delay T olerant Netw ork (DTN) (1). A DTN is an o v erlay on top of re gional netw orks, including the Internet. DTNs support interoperability of re gional netw orks by han- dling enormous propag ation delays between and within re gional netw orks, and by translating between re gional netw ork communication characteristics. In pro viding these functions, DTNs accommodate the mobility and limited po wer of e v olving wireless communication de vices and co v er up for their incon v eniences. T o accom- modate these features onto the e xisting TCP/IP layer , DTN adds the Bundle layer which most importantly does persistent storage. It is an additional layer on top of the transport layer where the ‘delay is tolerated’. Actually it is the layer in between the application layer and heterogeneous re gion specific lo wer layers which will act as a bridge between incompatible netw orks. F or routing messages in such scenarios message replication seems the solution, b ut we need to k eep in mi nd the v arious netw ork limitations of DTN lik e ener gy , cost and lo w bandwidth. More replication means increasing all these f actors and hence pure replication w ould not gi v e a desirable result here. Thus we need to w ork on mechanisms which can pro vide a netw ork with a balanced solution without testing its limitations too se v erely . In this paper we first look into some related w ork in the routing domain. W e then state our mot i v ation for conceptualising a ne w scheme for DTN. Ne xt we put forw ard our proposed approach and then pro v e its soundness by a mathematical analysis. Ne xt we elucidate the simulator we ha v e used for our simulation pur - pose. Then we compar e our algorithm to pre vious other algorithms sho wing results with dif ferent parameters. Lastly we dra w the conclusion. 1.1. Pr e vious W ork Routing protocols in DTN can be broadly classified into tw o groups on the nature of their strate gy chosen to forw ard a messa g e . This strate gy itself can be either replication based (which relies on more number of copies to increase the chances of a message being deli v ered) or kno wledge based (which uses netw ork in- formation to route data). Based on these strate gies the classification are namely , flooding based and forw arding based routing. In flooding based protocols, a node mak es numerous copies (or replicas) of a single message in an attempt to increase the chance of it being deli v ered. One such popular routing protocol (and also one of the earliest protocol for routing in DTN) is the Epidemic routing (3) which w as originally proposed for v ehicular netw orks (4; 5). Epidemic routing simply replicates messages from one node to another as and when a connection is established between them. In this w ay the chances of message deli v ery increases b ut a lot of netw ork resources (lik e bandwidth, b uf fer storage, ener gy and cost) are w asted in return. An impro v ement on this protoc o l is the Credit Based strate gies (6; 7) which assign a credit v alue to e v ery node connected on their time of connection and gi v es them the po wer to replicate based on that credit. These strate gies greatl y reduce number of replicas. Another protocol in this f amily is the Spray & W ait (8). Spray & W ait is composed of tw o phases. The first phase is kno wn as the Spray phase in which the source of the message sprays one cop y to L (predefined) distinct nodes. After a node recei v es the cop y it enters the w ait phase (the second phase) where it holds the message until it meets its destination directly . Spray & W ait has a v ariant kno wn as Binary Spray & W ait (8) in which a node tha t has n > 1 message copies (source or relay), and encounters another node B (with no copies), hands o v er to B ceil [ n= 2] copies and k eeps f l oor [ n= 2] copies for itself; when it is left with only one cop y , it switches to direct transmission. Spray and focus (9) has the same 1 st phase as Spray and W ait (8). In its 2 nd phase a node forw ards the message based on an utility criteria. MaxProp (10) maintains an ordered- queue based on the destination of each message, ordered by the estimated lik elihood of a future transiti v e path to that destination. PRoPHET (11) too uses pre vious performance records of nodes and routes data based on its findings. Incenti v e a w are routing (12) uses the pair -wise tit-for -tat (TFT) mechanism for routing. The Optimal Probabilistic F orw arding Protocol (13), uses an optimal probabilistic forw arding metric deri v ed by modeling each forw arding as an optimal stopping rule problem. The performance of these methods may deteriorate when the netw ork beha v es dif ferently from what the y had anticipated (an usual phenomena in DTNs). The best method to combat increase in numerous replication of messages is the forw arding based routing mechanism of forw arding a single cop y of a me ssage through the netw ork (14). In this approach a node will pass the message it has to only one of the nodes it connects t o according t o some m etric acquired on observing the netw ork. One such algorithm is the Store and F orw ard on First Contact which routes the message to the first node it connects to (15). K eeping in mind the rarity of an y node-to-node connection it is v ery e vident that although this algorithm incurs minimum replication am ong all algorithms, it is v ery dubitable that it will be able to deli v er man y messages. Another v ariant of this is the Direct Deli v ery protocol (16) which forw ards a mes sage only to its d e stination node directly when it comes in contact with it. This algorithm can IJECE V ol. 6, No. 6, December 2016: 3205 3216 Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE ISSN: 2088-8708 3207 be defined as de generati v e cases of both Flooding based as well as F orw arding based schemes. It has the least o v erhead of all b ut suf fers greatly in deli v ery and delay performances due to ob vious reasons. There are man y protocols belonging to this f amily of forw arding based algorithms where the messages are forw arded using some netw ork oracle or where routing completely tak es place according to some opportunistic or probabilistic metric. Practical routing (17) uses observ ed information about the netw ork for calculating minimum estimated e xpected delay (MEED) for routing. Conditional shortest path routing (18) routes messages with the help of a metric called conditional intermeeting time based on t he observ ations about human mobility traces. These techniques mainly thri v e on netw ork oracles such as future contact arri v al time, last encounter time etc and as a result f ail to utilize the opportunity presented by the present netw ork configuration. 1.2. Moti v ation Finding an optimal routing algorithm for DTN in the general case has been pro v ed to be an NP hard problem (19). Thus we can only focus on specific scenarios and objecti v es and find w ays to maximise its performance v alue. Hence we need to do topical optimization. In Binary Spray & W ait the source of a message initially starts with n copies; an y node A that has n > 1 message copies (source or relay), and encounters another node B (with no copies), hands o v er to B ceil [ n= 2] copies and k eeps f l oor [ n= 2] copies for its elf; when it is left with only one cop y , it switches to direct transmission. The follo wing theorem put forw ard in (8), states that Binary Spray and W ait is optimal in terms of minimum e xpected delay , when node mo v ement is IID (Independent and Identically Distrib uted), referring to IID random v ariables, i.e. when each random v ariable has the same probability distrib ution as the others and all are mutually independent. The Theorem: When all nodes mo v e in an IID manner , Binary Spray & W ait routing is optimal, that is, has the minimum e xpected delay among all proposed v ariants of the Spray & W ait routing scheme. The approach for this algorithm can be used in a more ef ficient manner for non IID nodal mo v ement wherein the latenc y and o v erhead performance can be impro v ed upon. It is not a frequent occurrence that all node mo v ement is IID. So we aim for this impro v ement and choose Binary Spray & W ait algorithm for further modification, it being already good in one aspect we just need to de vise a w ay to decrease its o v erhea d and latenc y for non IID node mo v ement. On analysis we find that the loophole of this algorithm in terms of latenc y occurs due to its reaching the w ait phase a bit later (longer tree). It gets totally dependent on the initial v alue of L . More the v alue, more is the delay to reach w ait phase and hence more o v erhead. A high priority message is bound to ha v e lar ger v alue of L b ut it also means with this system (no matter what), its w ait phase will also get delayed. Its true, though, that more replication are being made, hence more chances of message deli v ery is being created b ut can’ t their be a w ay to de vise a method wherein this number of repli cations are not reduced b ut at the same time a chance is created to reach the w ait phase earlier and reduce latenc y and o v erhead. Making this our moti v ation and k eeping these k e y points into considerations we de vise our ne w algorithm, the Fibonary Spray & W ait. 2. THE PR OPOSED METHOD Our proposed routing protocol f alls under the replication based strate gies and do not use an y netw ork oracle for routing. F or this routi ng algorithm, we will w ork with the f amous Fibonacci sequence in mathemat- ics: 0 ; 1 ; 1 ; 2 ; 3 ; 5 ; 8 ; 13 ; 21 ; 34 ; 55 ; 89 ; 144 . By definition, the first tw o numbers in the Fibonacci sequence are 0 and 1 , and each subsequent number is the sum of the pre vious tw o. In mathematical terms, the sequence F n of Fibonacci numbers is defined by the recurrence relation: F n = F n 1 + F n 2 Let us assume, that our starting node had N copies of a message to be gin with. No w , at an y instant, let a node A ha v e n (where n > 1 ) message copies. W e first determine if n is a Fibonacci number or not. So there are tw o cases that may arise: 1. If n , is NO T a Fibonacci number: Then our node A , forw ards m copies to B (a recipient node with no copies), where m is the highes t Fibonacci number not greater than n . That is, m is the immediate predecessor of n in the Fibonacci series. As a result, our parent node A , no w retains ( n m ) copies of the message for itself. 2. If n , is a Fibonacci number: The parent node A , no w simply follo ws Binary Spray & W ait algorithm, i.e., it forw ards ceil ( n= 2) number of copies to B , and k eeps floor ( n= 2) for itself. F ibonary Spr ay and W ait Routing in Delay T oler ant Networks (Priyanka Das) Evaluation Warning : The document was created with Spire.PDF for Python.
3208 ISSN: 2088-8708 5 5 2 3 2 3 1 1 1 2 1 1 1 2 1 1 1 1 1 2 3 4 5 10 Figure 1: Message replica distrib ution through Binary Spray and W ait 8 2 4 3 1 1 4 2 2 1 10 1 1 1 1 1 2 3 5 6 1 1 1 3 4 Figure 2: Message replica distrib ution through Fi- bonary Spray and W ait When a node is left with only 1 cop y , it switches to direct transmission. Thi s tri vial case can be summed up as a consequence of 1 being a Fibonacci number , so in accordance with clause 2 of our Fibonary Spray & W ait algorit h m , direct transmission happens . Thus, our spraying algorithm in Fibonary Spray & W ait routing is actually determined alternately by the Fibonacci series, and Binary Spray & W ait algorithm. One distinct adv antage of Fibonary Spray & W ait o v er Binary Spray & W ait is that, in Fi bonary Spray & W ait, the direct transm ission case is reached earlier than the corresponding Binary Spray & W ait Routing with the same number of copies. In other w ords, Fibonacci distrib ution is more asymmetric compared to the perfectly symmetric (half) di vision of message copies in Binary Spray & W ait. F or e xample, let us consider the case of 10 copies with our starting node: Binary Spray & W ait di vides 10 into 5 and 5 , whereas our Fibonary algorithm di vides it into 8 and 2 number of copies (see Figure 2). No w 8 may be further di vided into 4 and 4 in subsequent iterations; b ut on the other hand 2 gets di vided into 1 and 1 , and hence direct transmission is reached in the 3 r d le v el itself. F or Binary Spray & W ait to reach direct transmiss ion for initial number of copies as 10 , it will need to go do wn to the 4 th le v el in the spray-tree (Figure 1). Ho we v er , the best case may be impro v ed in Fibonary Spray & W ait, when we compare it with Binary for the same initial number of copies. This is actually a consequence of direct transmission being reached earlier , or in other w ords, the spray phase getting o v er earlier in one branch of the spray-tree. The Algorithm for the same is illustrated ne xt. In the algorithm, ttl g = time to li v e of message g . Algorithm 1: AtNode A 1 f or eac h messa g e g with n copies at node A in the network do 2 while ttl g > 0 do 3 if a connection e xists between curr ent node A carrying messa g e g and any other node B then 4 if n is NO T a F ibonacci number then 5 forw ard m copies of g to B 6 k eep n m copies of g for itself 7 if n is a F ibonacci number then 8 forw ard ceil ( n= 2) copies of g to B 9 k eep floor ( n= 2) copies of g for itself 10 if n = 1 then 11 forw ard g to B if f B is the destination of g 12 if g has not been r eceived by node B earlier then 13 accept g 14 mark g as deli v ered 15 del iv er ed = del iv er ed + 1 16 else 17 discard g IJECE V ol. 6, No. 6, December 2016: 3205 3216 Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE ISSN: 2088-8708 3209 2.1. Mathematical Analysis W e assume that there are n nodes randomly distrib uted with a uniform distrib ution. Each message picks one of the remaining n 1 nodes as its destination with an equal probability . F or our algorithm, e v ery message can ha v e a maximum of ` L 0 copies b ut these ` L 0 copies are to be distrib uted am ong other nodes on the occurrence of a connection based on a predefined maxim. This maxim as already discussed earlier states that there are tw o possibilities for message distrib ution at e v ery branching turn. The tw o cases that may arise are: 1. When ` L 0 is a fibonacci number: then tree distrib ution remains same as in Binary Spray & W ait. 2. When ` L 0 is not a fibonacci number: then suppose L = x such that, 0 ; 1 ; 1 ; 2 ; :::a; b; c:::n is the fibonacci series where b < x < c . ) in inte ger series, the position of x with respect to fibonacci numbers a; b and c w ould be :::a; b; x; c::: No w , in Binary Spray & W ait, this x w ould gi v e rise to tw o di visions of ceil [ n= 2] and f l oor [ n= 2] . In Fibonary Spray & W ait, it w ould be b and ( x b ) . T o get a shorter branch we ha v e to sho w that this dif ference ( x b ) is less than x= 2 . Note that, b < x < c ....................(1) also, c b = a ....................(2) Let x b = d ....................(3) W e ha v e to pro v e that this dif ference ( d ) is al w ays less than x= 2 . As ` d 0 is used in Fibonary Spray & W ait and x= 2 is used in Binary Spray & W ait, the one with smaller v alue will gi v e earlier w ait phase. W e kno w that d < a ....................(4) So if we can sho w a 6 x= 2 , d < x= 2 can be deduced. Let us pro v e this using mathematical induction. d has to be greater than 1 for x to be a non fibonacci number . ) d > 1 ) a > 1 ) a is a minimum of 2 , ) b is a minimum of 3 , ) c is a minimum of 5 , ) x is a minimum of 4 , So the first possible v alue of x is 4 . F or x=4, 2 6 4 = 2 ) 2 6 2 (true) ....................(5) Let this be true for x = n , a 6 n= 2 ....................(6) No w for x = ( n + 1) = 2 , we ha v e to sho w that, F ibonary Spr ay and W ait Routing in Delay T oler ant Networks (Priyanka Das) Evaluation Warning : The document was created with Spire.PDF for Python.
3210 ISSN: 2088-8708 a 6 ( n + 1) = 2 W e kno w , a 6 n= 2 ) a 6 n= 2 + 1 = 2 ) a 6 ( n + 1) = 2 ....................(7) Hence pro v ed by mathematical induction. From (4) , d < a , and a 6 ( n + 1) = 2 from (7) ) d < x= 2 . ) W e are bound to get a branch in Fibonary Spray & W ait which renders quick er w ait phase than the corresponding branch in Binary Spray & W ait. 3. RESUL TS AND DISCUSSION 3.1. Simulation Setup W e ha v e done our simulation using the Opportunistic Netw ork En vironment (ONE) simulator (20), which is specifically designed for DTNs. A time step of 0 : 1 s has been considered, i.e. an y e v ent recording is done after a time step of 0 : 1 s . A ne w message is created e v ery 25 to 35 seconds with message sizes v arying between 500 k B and 1 M B . The simulation w orld size is 4500 3400 meter s . The mo v ement model that we ha v e chosen for all hosts in our simulation is the Shortest P ath Map-Based Mo v ement (SPMBM) model. The reason for this choice is that among the v arious mo v ement models present this is the more realistic one wherein instead of a completely random w alk, the nodes choose a random point on the map and then follo w the shortest route to that point from their current location. Further details of t he simulation parameters tak en into consideration are discussed in T able 1. F or f air comparison purposes we ha v e compared our algorithm with dif ferent routing schemes in this genre namely the Spray & W ait (also called the V anilla Spray & W ait) (8), Binary Spray & W ait (8) and Epidemic (3) (these schemes ha v e already been discussed in Section II Pre vious W ork). The results ha v e been obtained after a rigorous simulation by v aryi ng the number of copies ( L ) o v er a period of 2 days. In addition, a comprehensi v e comparison of some of the popular e xisting routing protocols is sho wn in T able 2. The results are based on a simulation run for 24 hours in the ONE Simulator . The simulation parameters used are as discussed in T able 1. The initial number of messages L for Spray & W ait , Binary Spray & W ait and Fibonary Spray & W ait, has been tak en as 12 for the simulations in T able 2. The graphs ha v e been plotted using gnuplot. T able 1: Simulation P arameters P arameters Node T ype Pedestrian Cars T rams No. of Nodes 58 30 2 Speed 0.5 to 1.5 m/s 10 to 50 km/h 7 to 10 m/s Buf fer Size 5 MB 5 MB 50 MB Message T ime T o Li v e 300 minutes 300 minutes 300 minutes 3.2. Effect of V arying Number of Copies ( L ) The number of copies( L ) is by f ar the most important metric to be t ested by v ariation for an y v ariant of the Spray & W ait scheme (including our Fibonary). Hence we ha v e compared the three v ariants of the Spray & W ait (V anilla, Binary and Fibonary) by v arying the v alue of L on a di v erse range. A v alue of 7 for L has been considered as the lo west and 26 the highest. These simulations ha v e been done on a constant time scale IJECE V ol. 6, No. 6, December 2016: 3205 3216 Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE ISSN: 2088-8708 3211 of 1 day i.e. 24 hour s . Since Epidemic Routing has no concept of maximum number of copies ( L ), we omit it in the comparison charts. F ollo wing is the detailed analysis of the ef fect of v arying number of copies ( L ) on some selected performance parameters. 3.2.1. Impact on Message Deli v ery pr obability From the plot of Figure 3 it is prominent that for lo wer v alues of L , Fibonary S pray & W ait gi v es best message deli v ery probability among the considered schemes. F or higher v alues of L , though not the best b ut Fibonary Spray & W ait sho ws consistent performance and is al w ays better than V anilla Spray & W ait (results say the message deli v ery probability of V anilla Spray & W ait is on an a v erage 3% less than that of Fibonary Spray & W ait). All in all it gi v es quite an impressi v e performance and good message deli v ery probability on all v ariations of L . 3.2.2. Impact on Ov erhead Ratio The o v erhead ratio is defined as the ratio of (Number of Messages Relayed - Number of Mes sages Deli v ered) to (Number of Messages Deli v ered). The o v erhead ratio of both Binary Spray & W ait and Fibonary Spray & W ait is understandably more than v anilla as the y are more ambitious schemes, b ut out of the tw o the o v erhead of Fibonary Spray & W ait is less for lo wer v alues of L . Though it increases for higher v alues of L , it still gi v es a respectable performance. Hence this scheme w orks superfine for netw orks with lo wer v alues of L (usually small netw orks) and also for higher v alues of L as o v erhead is just a tad more than that of Binary Spray & W ait (see Figure 4). 3.2.3. Impact on Message Latency The a v erage mess age latenc y for both Binary Spray & W ait and Fibonary Spray & W ait is more than V anilla Spray & W ait on most occasions. From Figure 5, one can deduce that e xcept when L = 7 , a v erage message latenc y in case of Fibonary Spray & W ait is much less than Binary Spray & W ait. On an a v erage, the a v erage message latenc y of Fibonary Spray & W ait is 1 : 2% less than that of Binary Spray & W ait. The results of message latenc y median though tell a dif ferent story . F or lo wer v alues of L , Fibonary Spray & W ait w orks the best (see Figure 6) among all schemes considered. 3.2.4. Impact on Message Hopcount Message hopcount is the number of nodes that a message has to pass by before reaching its desi gnated destination. Hence if hopcount can be k ept to a bare minimum, one can cash on a number of things lik e sa ving b uf fer space, replica, route congestion etc. Simulation results sho w that though V anilla Spray & W ait outperforms yet ag ain, among the other tw o, Fibonary Spray & W ait performs better in most situations (see Figure 7 and Figure 8). Especially for L = 7 , Fibonary Spray & W ait performs as good as V anilla Spray & W ait (see Figure 8). 3.2.5. Impact on Message Buffertime The time a message spends on a b uf fer (source or intermediate) gi v es a clearer and more com pact vie w of the delay incurred as well as the congestion created due to message replications. Thus this metric is of utmost importance and is one of the tw o most important things to be considered for e v ery routing scheme in DTN (the other metric being message deli v ery probability). The plots of Figure 9 and Figure 10 sho w that in both cases of a v erage message b uf fertime and b uf fertime median, Fibonary Spray & W ait performs w ay better than V anilla Spray & W ait. In f act here the V anilla Spray & W ait incurs huge delay , w ay abo v e than the other tw o. Also Fibonary Spray & W ait performs better than Binary Spray & W ait. Speaking in numbers, on an a v erage the a v erage message b uf fertime of Fibonary Spray & W ait is 19% less than that of V anilla Spray & W ait and 2 : 133% less than that of Binary Spray & W ait. Also the a v erage message b uf fertime median of Fibonary Spray & W ait is 30% less than that of V anilla Spray & W ait. So if we consider delay due to b uf fertime then our scheme w orks the best in the Spray & W ait genre. F ibonary Spr ay and W ait Routing in Delay T oler ant Networks (Priyanka Das) Evaluation Warning : The document was created with Spire.PDF for Python.
3212 ISSN: 2088-8708  0.3  0.32  0.34  0.36  0.38  0.4  0.42  0.44  5  10  15  20  25  30 Message Delivery Probability No.of Copies (L) Binary SnW Vanilla SnW Fibonary Figure 3: Message Deli v ery probability (Simula- tion time for each is 24hours)  10  12  14  16  18  20  22  24  5  10  15  20  25  30 Overhead Ratio No.of Copies (L) Binary SnW Vanilla SnW Fibonary Figure 4: Ov erhead ratio (Simulation T ime = 24 hour s )  3000  3100  3200  3300  3400  3500  5  10  15  20  25  30 Average Message Latency(s) No.of Copies (L) Binary SnW Vanilla SnW Fibonary Figure 5: A v erage Message Latenc y (Simulation T ime = 24 hour s )  2400  2450  2500  2550  2600  2650  2700  2750  2800  5  10  15  20  25  30 Message Latency Median(s) No.of Copies (L) Binary SnW Vanilla SnW Fibonary Figure 6: Message Latenc y Median (Simulation T ime = 24 hour s )  1.5  2  2.5  3  3.5  5  10  15  20  25  30 Average Message Hopcount No.of Copies (L) Binary SnW Vanilla SnW Fibonary Figure 7: A v erage Message Hopcount (Simula- tion T ime = 24 hour s )  1.5  2  2.5  3  3.5  5  10  15  20  25  30 Message Hopcount Median No.of Copies (L) Binary SnW Vanilla SnW Fibonary Figure 8: Message Hopcount Median (Simulation T ime = 24 hour s )  2000  2200  2400  2600  2800  3000  3200  5  10  15  20  25  30 Average Message Buffertime(s) No.of Copies (L) Binary SnW Vanilla SnW Fibonary Figure 9: A v erage Message Buf fertime (Simula- tion T ime = 24 hour s )  1600  1800  2000  2200  2400  2600  10  15  20  25  30 Message Buffertime Median(s) No.of Copies (L) Binary SnW Vanilla SnW Fibonary Figure 10: Message Buf fertime Median (Simula- tion T ime = 24 hour s ) 3.3. Effect of V arying Simulation T ime These simulation results sho w a candid performance which ha v e been tested for short as well as long periods of time. F or this ef fect we v aried the time from 24 hour s (lo w) to 48 hour s (high). When v arying the time quotient we ha v e k ept the number of copies ( L ) at a constant v alue of 26 . This we do to sho w that e v en for IJECE V ol. 6, No. 6, December 2016: 3205 3216 Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE ISSN: 2088-8708 3213 high v alue of L , our scheme w orks just as fine as we ha v e already se en ho w we ll it performs for lo wer v alues of L . 3.3.1. Impact on Message Deli v ery pr obability The bar graphs of Figure 11 depicts that Fibonary Spray & W ait has more message deli v ery probabil ity than either V anilla or Epidemic routing (Epidemic has on an a v erage 43 : 1% lo wer deli v ery probability than Fibonary) o v er all periods of time tested. It though lags behind a little o v er Binary Spray & W ait b ut the dif ference is not much considering other metrics. 3.3.2. Impact on Ov erhead Ratio As e xpected, Figure 12 portrays that V anilla Spray & W ait and Epidemic routing form the tw o oppos ite ends in the spectrum for o v erhead ratio. Our algorithm, Fibonary Spray & W ait is mezzo between the tw o techniques and gi v es a pretty impressi v e performance. In f act our scheme creates 97% lesser o v erhead as compared to Epidemic routing. 3.3.3. Impact on Message Latency The a v erage message latenc y readings from the plot of Figure 13 sho ws that in terms of latenc y , Fibonary Spray & W ait performs better than Binary Spray & W ait (it incumbers a v eragely 1 : 232% lesser latenc y)and Epidemic routing (on an a v erage 32 : 243% lesser) and compensates for its slightly lo wer message deli v ery probability than Binary Spray & W ait. Ev en if we consi der the message latenc y median results, (see Figure 14), Fibonary Spray & W ait ag ain sho ws better performance than Binary Spray & W ait and Epidemic routing ( 21 : 83% lesser) especially when running for longer stretches of time. 3.3.4. Impact on Message Hopcount The graphs of Figure 15 and Figure 16 sho ws a v ery interesting result, that at higher v alue of L (= 26) , Binary and Fibonary Spray & W ait almost ha v e the same a v erage hopcount and along with Epidemic ha v e similar hopcount medians for v aried periods of simulation time. Furthermore, Fibonary Spray & W ait requires 24 : 4% lesser hopcounts than Epidemic routing. 3.3.5. Impact on Message Buffertime Figure 17 and Figure 18 ag ain sho w that Fibonary Spray & W ait is a mezzo between the lo w b uf ferti me of Epidemic and high b uf fertime of V anilla Spray & W ait. Fibonary Spray & W ait performs quite decently with respect to both a v erage as well as b uf fertime median gi ving 29 : 3% lesser message b uf ferti me than Epidemic routing and 41 : 003% lesser b uf fertime median than V anilla Spray & W ait. 3.4. Extensi v e Comparison T able 2 and T able 3 pro vide further comparison of Fibonary Spray & W ait with some more e xisting protocols. One can analyse from these tables ho w our proposed algorithm, Fibonary Spray & W ait, stands in comparison to the other established protocols present. W e did not include the lik es of MaxProp, PRoPHET etc in the pre vious detailed comparisons as the y are part of a dif ferent cate gory of protocols and are not af fected by the parameters which af fect Fibonary Spray & W ait ( lik e the v ariable L ). From T able 2 we observ e that among the v arious routing protocols, the Fibonary Spray & W ait maintains a good balance between Deli v ery Probability , Ov erhead Ratio and Latenc y . In f act it comes second only to MaxProp in terms of Deli v ery Prob- ability b ut beats it in all other metrics by a huge mar gin. Its Ov erhead Ratio is also on the lesser side and in Latenc y readings beats all other protocols b ut Spray & W ait. 4. CONCLUSION In this w ork we ha v e delv ed into the propitious routing problem of Delay T olerant Netw orks. Though a considerable amount of research has been done in this field, there still is a huge scope for impro v ement as dif ferent kinds of DTNs can benefit from these schemes. Al so optimal routing in DTNs being an NP hard F ibonary Spr ay and W ait Routing in Delay T oler ant Networks (Priyanka Das) Evaluation Warning : The document was created with Spire.PDF for Python.
3214 ISSN: 2088-8708 T able 2: A Comparison Between Some Existing Routing Protocols and Fibonary Spray & W ait Routing Protocols Simulation Results Deli v ery Probability Ov erhead Ratio Latenc y A v erage Latenc y Median Epidemic Router 0.2491 42.2223 4509.5733 3404.1000 Spray & W ait Router 0.3461 14.0375 3143.7860 2579.0000 Binary Spray & W ait Router 0.3621 17.6189 3264.9631 2580.3000 Fibonary Spray & W ait Router 0.3703 17.1993 3220.5839 2580.3000 MaxProp Router 0.4305 23.0405 6942.4543 5952.8000 PRoPHET Router 0.2518 38.0258 4709.0958 3778.6000 First Contact Router 0.1896 34.7712 4944.5386 3801.8000 Direct Deli v ery Router 0.3184 0.0000 6549.6077 5725.0000  0.2  0.25  0.3  0.35  0.4  24  36  48 Message Delivery Probability Simulation Time(h) Binary SnW Vanilla SnW Fibonary Epidemic Figure 11: Message Deli v ery probability (No. of Copies ( L ) = 26 )  15  20  25  30  35  40  24  36  48 Overhead Ratio Simulation Time(h) Binary SnW Vanilla SnW Fibonary Epidemic Figure 12: Ov erhead Ratio (No. of Copies ( L ) = 26 )  3000  3200  3400  3600  3800  4000  4200  4400  4600  24  36  48 Average Message Latency(s) Simulation Time(h) Binary SnW Vanilla SnW Fibonary Epidemic Figure 13: A v erage Message Latenc y (No. of Copies ( L ) = 26 )  2400  2600  2800  3000  3200  3400  24  36  48 Message Latency Median(s) Simulation Time(h) Binary SnW Vanilla SnW Fibonary Epidemic Figure 14: Message Latenc y Median (No. of Copies ( L ) = 26 )  1  1.5  2  2.5  3  3.5  4  24  36  48 Average Message Hopcount Simulation Time(h) Binary SnW Vanilla SnW Fibonary Epidemic Figure 15: A v erage Message Hopcount (No. of Copies ( L ) = 26 )  1  1.5  2  2.5  3  3.5  4  24  36  48 Message Hopcount Median Simulation Time(h) Binary SnW Vanilla SnW Fibonary Epidemic Figure 16: Message Hopcount Median (No. of Copies ( L ) = 26 ) problem, one can only aim to w ork to w ards a particular specific goal. Hence we proposed our algorithm which w orks on the good points of the alre ady established Binary Spray & W ait and yet impro v es on it by pro viding a IJECE V ol. 6, No. 6, December 2016: 3205 3216 Evaluation Warning : The document was created with Spire.PDF for Python.