BRPI0905542A2 - anycast routing method using genetic algorithms in delay and disconnect tolerant networks - Google Patents

Abstract

<B> ANYCAST ROUTING METHOD USING GENETIC ALGORITHMS IN TOLERANT NETWORKS TO DELAYS AND DISCONNECTIONS. <D> The present invention deals with a method for selecting anycast paths using genetic algorithms (AGs) in networks tolerant to delays and disconnections or DTNs, represented through evolutionary graphs. In the method, the set of nodes that will form the routes between the source-destination pair of each anycast session is defined as a chromosome, which has a representation of constant size. Dijkstra's shortest path algorithm is used to generate potential solutions (routes) for each anycast session. These solutions are combined by the AG-based anycast routing algorithm, which is designed to produce only regular individuals and to allow you to find solutions in a short time. The AG, by assessing the suitability function, analyzes the various combinations of routes in order to find routes that optimize the routing objectives: routes that satisfy a minimum delivery rate and have the shortest delay. The use of the above approaches allows the route selection performed by the routing algorithm to be efficient and to optimize the performance measures of the network.

Description

"ANYCAST ROUTING METHOD USING ALGORITHMOSGENES IN DELAY AND DISCONNECTION NETWORKS".

Field of the Invention

The present invention relates to a anycast routing method for delay and disconnect tolerant networks or DTNs1 more specifically, an anycast routing method for route selection using genetic algorithms (AGs) in DTN networks represented by evolutionary graphs. The method seeks to find appropriate combinations of routes and destinations that allow the network to be able to transmit messages with a total delivery probability above a permissible value and with the minimum delay, based on information obtained from an evolutionary graph that will characterize the NTD.

State of the Art

DTN networks address the problem of heterogeneous networks that must operate in environments subject to long delays and discontinuous end-to-end connectivity. And in this regard, there is a growing effort to enable communication on networks with these challenging characteristics, thus requiring a delay-and-disconnect-tolerant network architecture (V. Cerf et al., Delay-tolerant networking architecture, IETF RFC 4838, informational, 2007).

The DTN architecture foresees the use of the message switching technique and the persistent storage of data defining an overlay called the Bundle Layer. Each location (below the application layer and above the others) should allow applications not to be developed taking into account delay and disconnection issues, and to allow DTN to become fully independent of the various regional networks.

One of the key elements of DTN architecture is routing. Due to the characteristics of DTNs, efficient data routing in these NTNs has become a critical and challenging issue. Routing protocols for NTDs should take into account that establishing an end-to-end path between source and destination nodes is not guaranteed. Thus, traditional routing protocols are not suitable for networks with the characteristics of DTNs. In addition, to operate efficiently in the most diverse environments in which a DTN node can be found, DTN nodes will likely have to support different routing strategies and protocols. Given these problems, routing in DTNs is a challenging area that still remains open in architecture.

Basically, routing protocols in NTDs are divided into two categories: deterministic and stochastic. Deterministic protocols use information that nodes obtain about network conditions and connectivity to make transfer decisions. Stochastic protocols, on the other hand, try to disseminate several copies of messages among several nodes that will assist in the transmission of the message.

In deterministic routing protocols, the network topology may be known over time, that is, it is assumed that movement and future connections are known. To allow the analysis of the behavior of these dynamic networks over time, a modeling is performed through evolutionary graphs (Afonso Ferreira, "Building a combinatorial reference model for MANETS", IEEE Network, vol. 18, no. 5, pp. 24-29 , 2004). Briefly, an evolutionary graph is an indexed sequence of subgraphs, where the subgraph associated with a given index corresponds to the network topology in the time interval indicated by that index. As a result, various parameters, such as cost and link capacity, that characterize the connection will vary during each time interval.

The time modeling of the networks obtained by the graphs gives rise to several matrices that can be used as objective functions in routing strategies, such as shorter time to reach one or all destinations, or path with the least hops, and this It is an important feature of using the evolutionary graph model: the possibility of considering different metrics as objectives. In addition, using evolutionary graphs makes it easy to visualize the dynamic behavior of DTNs.

An important factor in demonstrating the challenges of routing in DTTs is the definition of node characteristics and edges in the graph. Each edge between two nodes can be represented by mobile devices moving from one node to another (Y. Gong et al., "Anycast Routing in Delay Tolerant Networks", IEEE Global Telecommunications Conference1 pp. 1-5, 2006). Thus, graphfoTTN will be characterized by the limited storage capacity of nodes (disconnected regions) and mobile devices (allowing communication between disconnected regions), as well as the start time and propagation delay of each mobile device. Another important feature of this representation is that network nodes are stationary and generate messages, while mobile devices move from node to node, carrying messages, with themselves not generating messages.

The routing algorithm must take into account the defragmentation / reassembly capability. According to the DTN architecture description, there are two forms of fragmentation / reassembly:

Proactive Fragmentation: A DTN node can split an application data block into multiple smaller blocks and transmit each of these blocks as an independent aggregate. This approximation is used essentially when contact volumes (the product of transmission capacity by contact, ie, the period of time during which the capacity is positive and we are able to exchange data) are known (or predictable) as a back-up;

- Reactive Fragmentation: The previous hop sender can learn (through convergence layer protocols) that only a portion of the aggregate has been delivered to the next hop, and send the remaining portion (s) when subsequent contacts become available (possibly using different hops if routing changes). This scheme is called reactive fragmentation because the defragmentation process occurs after a transmission attempt has occurred.

Reactive fragmentation capability is not required to be available in every DTN implementation, however, the ability to regroup at the target is required to support fragmentation.

The delivery semantics of a message are classified according to the message recipient (s). When the message is sent to a specific node, it has unicast semantics. In cases where the message is intended for a receiver group, two delivery options become available: when the message is to be sent to anyone, and only one of the nodes of the receiver group is semantics anycast. On the other hand, when the message is sent to many or all nodes of the receiving group the semantics is multicast.

Because it is an important service that should be further explored, anycast is useful in situations where a host, application, or user wants to locate a server that supports a particular service. However, if multiple servers support the service, any one of these servers may be used. Anycast on DTNs can be applied in resource discovery cases, for example, people looking for a particular service, such as a doctor or firefighter, without knowing the specific location or ID. It also applies to battlefields, where a soldier wants to deliver a particular message to any command center, or a particular command center wants to send a message to any soldier in a battalion. Anycast can even be used in education, in which students wish to obtain a particular file from any of the members of a student group.

Due to the long and / or variable delays of DTNs, the traditional anycast does not clearly specify nodes that can receive a message when the environment presents the challenges of a DTN network. The definition of the planned message receivers, which must be fixed (although the target group members can join or leave the group), allows the specification of new semantic models (Y.Gong et al., "Anycast routing in delay tolerant"). networks ", IEEE Global Telecommunications Conference1 pp. 1-5, 2006).

As a DTN can have many types of contacts, and in addition, the transmission opportunity, that is, the contact can be very rare and difficult, the routing algorithm must be efficient to take advantage of the transmission opportunities and to use the available information. network to optimize routing.

The routing algorithm is responsible for defining the route and destination of each anycast session (with each anycast session indicating the intention of each source to send messages to the respective receiver group), and this precision influences many network parameters simultaneously (NP-type matching problem). complete). Because genetic algorithms (AGs) are appropriate for solving complex optimization problems, an AG-based anycast delivery routing algorithm will look for the appropriate combination of routes and destinations of each anycast session to optimize message delivery rate, delay, and speed. same distribution in the network.

US20070133504A1 utilizes AGs for broadcast broadcast selection in Ad-hoc networks for efficient data transmission. However, DTN networks have more challenging features than Ad-hoc networks, such as frequently disconnected paths and long, variable paths, with the possibility of never having end-to-end connectivity between source and destination in a given time period.

US20030050902A1 proposes the use of Applied AGs for optimizing the choice of individual sensors to be used in monitoring a particular purpose of a network desenser, with each chromosome representing a sensor. The chromosome decoding scheme in the present invention differs from the previous patent and is based on the nodes that make up the paths of each NetworkDTN anycast session.

In addition to the above cited patents, there are many others that use AGs to aid in routing: EP0921661A2, US7092378B1 and W02002082371A2. However, none addresses a method of routing oncast DTNs to minimize message delivery delay and achieve high delivery rates in order to efficiently utilize transmission opportunities on DTNs. The anycast routing method is based on GAs and the concept of subpopulation (CC Lo and WH Chang, "AmuItiobjective hybrid genetic algorithm for the capacitated multipoint network design problem," IEEE Transaction on System, Man, and Cybernetics, vol. 30, no. 3, pp. 794-470, 2000), as well as limitation of the algorithm search space (LS Randaccio and L. Atzori, "Group multicast routing problem: agenetic algorithms based approach", Computer Networks, vol. 51, no. 4, pp.3989-4004, 2007) are techniques used in the present invention to increase the efficiency of the routing algorithm, being strategies absent in the previous patents. In addition, each chromosome is represented by the network nodes that form the anycast paths.

In addition to the strategies mentioned above, the present invention differs from the others found for dealing with the anycast service, which should be further explored, in deterministic NTDs. In these networks, as the contacts are predictable, the routing algorithm uses fragmentation at the source, that is, blocks of data will be split only at the source and reassembled at the destination, which must have this reassembly capability.

Description of the Figures

For a better understanding of the present invention reference is made to the accompanying figures where:

FIGURE 1 shows the general AG flow diagram used for routing in the present invention.

FIGURE 2 shows the coding scheme used to represent each AG individual.

FIGURE 3 describes how the AGsc-based routing algorithm places individuals in order of suitability in the present invention.

FIGURE 4A shows two individuals to be applied to genetic operators.

FIGURE 4B shows the crossing scheme used.

FIGURE 4C shows the mutation scheme used.

Description of the invention

In the present invention, the routing algorithm will perform anycast routing on DTN networks using the information obtained from graphsevolutionary. Since the definition of the routes and destinations of each anycast session influence many parameters simultaneously, we applied GAs that are search algorithms based on natural selection and evolution mechanisms for routing optimization, as they are suitable for solving complex problems. In addition, the AGs-based routing algorithm allows the use of different anycast semantic models, depending on the application needs.

First, a set of potential solutions for ananycast routing are considered in isolation by combining the paths that connect each source-destination pair of a session. Potential assumptions are calculated in isolation using the Dijkstra domenor path algorithm based on the number of hops. To demonstrate how potential isolation insolutions are obtained, suppose a simplexcount of 9 nodes (# 1 through # 9) and an anycast session with the source / Ii node wishing to send messages to any of three possible receivers (also called planned receivers) n5, n8 and no. Applying the shortest path Dijkstra algorithm according to the network topology, the route with the smallest number of hops could be ni-n4-n5. Applying the algorithm again, the second shortest path could be ni-n2-n4-n5, followed by nrn3-n6-n9, and so on, the shortest path algorithm is applied until the number of potential solutions in isolation is found. wanted. These solutions will form an array of routes for each anycast session of the network.

The matrix F shows the solutions that could be obtained for the previous session described above. Depending on the number of nodes, links and sessions the number of potential solutions in isolation can be greatly increased. The ellipsis in the matrix indicates two ways to limit the size of the matrix: by the number of hops (columns) or by the number of solutions found (rows). Therefore, the number of potential solutions for anycast registration, which will be the shortest possible paths between the source node pair and one of the possible anycast destinations of each session, was limited to five, calculated using the shortest path Dijkstra algorithm. In this case, we would only consider the first five lines of matrix F. The presence of numbers in the matrix is to make all lines of the matrix the same size. We will see that this representation is useful for the routing algorithm that uses AG for route decision.

<formula> formula see original document page 9 </formula>

In the present invention the AGs-based routing algorithm is applied to anycast routing in DTNs1 using a node representation, a method of calculating potential solutions in isolation, and a criterion for limiting the number of potential solutions in isolation, as described for F matrix formation. .

Because network resources on a DTN are extremely scarce, and the opportunity for transmission can be rare and very disputed, one goal is to design a routing algorithm that can accommodate network traffic so that one traffic does not disrupt the other. This way, the anycast routing algorithm on AG-based DTNs will optimize network performance by searching for the appropriate combination of routes and destinations from each session. As input, the AG receives the potential solutions in isolation, ie the matrix F of each anycast session. FIGURE 1 depicts the operation of the GA of the present invention, having the following steps:

(a) definition of a method for encoding (2) potential solutions on chromosomes using potential solutions in isolation (1);

b) Creation of the initial population (3);

c) Assessment of the aptitude (5) of individuals of the population;

d) Verification of GA closure conditions:

d.1) if satisfied (6), execution is terminated and solution is the current generation population in order of suitability;

d.2) if they are not met (7), the following steps are performed.

e) Selection of individuals for reproduction (8) and maintenance of the best individuals of the generation to form two subpopulations: reserve elitism strategy (11) and alteration of routes in the best individuals (12);

f) For individuals selected for reproduction, genetic operators apply (9);

(g) The next generation or new population (4) shall be formed by four subpopulations:

g.1) Elitism reserve strategy (11) g.2) Change of routes in the best individuals (12);

g.3) Universal stochastic sampling (13) using individuals resulting from reproduction (8) and the application of genetic operators (9);

g.4) Completely random method (14) using potential isolation solutions (10) for generating individuals.

h) After the new population (4) is formed, the algorithm returns to the fitness assessment step (5) of individuals and continues to perform steps (8), (9), (11), (12), (13), (14), (4) and (5) until the closing conditions are met (6).

The following will detail each of the previous steps. Step one (a), coding (2), involves defining a method for coding potential chromosome resolutions.

The chromosome is formed by the combination of routes of eachanycast session (potential isolation solutions (1)), which represents a possible solution, and its length is proportional to the number of anycast sessions. The representation scheme used is shown in FIGURE 2 (based on the association of each gene at each node of the route between the destination source). Forming an individual from the population, one route is selected for the first session (z = 1), another for the second (z = 2), until the session z (z = z), where s are the source nodes, n nodes intermediaries and k the chosen destinations. In FIGURE 2 the source nodes s (s1, s2, s ^ z) have subindices related to anycast session number z. The subindexes for the intermediate nodes η (n ^ i, n ^ j, n ^ m, n ^ n, n ^ o) are defined by the route chosen by the routing algorithm (they will be the intermediate nodes of matrix F). Finally, for destination nodes k, doissub-indices (k12, k23, k ^ z1) are used: the first indicates the anycast session number z for which the destination was chosen, and the second index indicates which of the Planned receivers was chosen, spi to spz are the starting positions of each individual, and are defined based on the number of columns of the potential solution matrix in isolation from each anycast session. To keep these positions constant for all individuals, some genes are filled with 0 values, which justifies the placement of zeros in the matrix F described earlier. This addition of redundancy is to prevent the emergence of undesirable individuals, ie only regular individuals are created. This avoids the need to use repair and verification functions to test whether individuals are valid.

The second step (b) shown in the GA flowchart used is the creation of the initial population (3). The strategy of using a common population number of individuals greater than the number of nodes in the DTN network was adopted. For example, for a 40-node NTD, an initial population of 50 individuals can be used. For initial population formation, randomly select routes using the matrix of potential solutions in isolation from each anycast session. Next, to assess the fitness (5) of individuals in the population, network performance measures are used. These measurements are calculated based on the traffic generated (number of anycast sessions, number of messages in each session, among others) and representation information through evolutionary graphs: node storage capacity (£> (/)), mobile device storage capacity (c (/, J)) moving from one node / to one node /, time divided by mobile devices (w (i, j)) at each edge, and motion delay of mobile devices (md (ij) ). More precisely, two performance measures are considered:

1) Probability of delivery of messages

<formula> formula see original document page 12 </formula>

where mk (i) is the total number of unique anycast messages received by a member of the anycast receiver group for each anycast session ζ and ms (i) is the total number of messages transmitted by each anycast source. 2) Weighted average delay D:

<formula> formula see original document page 13 </formula>

where faith is the number of fragments, Wi is the weight of each fragment proportional to the number of messages in the current fragment and Dz (Z) is the delay of each fragment, that is, the sum of the average delay d (i, j) of each jump forming arota source-destination:

The average delay d (i, j) is the sum of the starting time of mobile devices w (JJ) and the motion delay md (i, j):

<formula> formula see original document page 13 </formula>

The fitness of the individual is determined according to the probability of delivery of Equation (1) and the delivery delay of Equation (2). In addition, a feature of the AG-based anycast routing algorithm is the calculation of routes with a delivery probability above a threshold (PErmn) and the shortest delay. FIGURE 3 shows a flowchart describing how the AG puts individuals in order of suitability. First, the current generation population (15) are placed in decreasing order of delivery probability (16). Then individuals who have a probability of delivery greater than a minimum PEmin value are verified. Individuals who have a probability of delivery greater than PEmin (18) are arranged in increasing order of delay (19), and together with those who have no probability of delivery greater than PEmin (17), will form the population of individuals in order of suitability (20). ). You can adjust the value of PEmin according to the needs of the application.

The termination condition of the GA used is the number of generations. If this condition is not met (7), the new population (4) must be formed. Thus, GA stores individuals with the best aptitude values through the reserve elitism strategy (11) for the formation of the first subpopulation. Route alteration in the best individuals (12), that is, in the chromosomes with the best aptitude values, forms the second subpopulation, with the purpose of this subpopulation being to obtain better individuals with small alterations (only the route of one nocromosome session) over the best of the previous generation. The third subpopulation is formed after the reproduction process (8). For selection of individuals for reproduction, the paired tournament selection technique is used, that is, two individuals from the population are randomly chosen, and the one with the best aptitude value will be used for reproduction.

The genetic operators (9) used are those of point crossing and mutation. To illustrate the operation of the genetic operators, the individuals X1 and X2 shown in FIGURE 4A are considered. The crossing is performed by randomly choosing the starting position of one of the routes represented by sp1, sp2, sp3, and spz. It should be remembered that letter autilization is for ease of viewing only, as each chromosome is represented by network nodes, and that, when necessary, is filled with zero to keep the chromosome size and each anycast session constant. For the crossing a random position is chosen, for example sp3 which will indicate the position from which the genetic material of the individuals will be exchanged. FIGURE 4B shows the individuals Y1 and Y2 resulting from crossover of individuals X1 and X2 at sp3. The mutation is performed by randomly changing one of the individual's pathways (for this the matrix of potential solutions in isolation corresponding to the position to which the mutation is desired) is used. FIGURE 4C illustrates X 'resulting from mutation of the second route of subject X1. To delimit the mutation, the sp2 and sp3-1 start position vectors are used, ie the starting position of the current route and the initial position of the subsequent route minus one. The crossing and mutation probabilities are pc = 0.8 and pm = 0.03, respectively. These values are slightly higher than the traditional ones to increase population diversity.

Of the two remaining subpopulations, only the third (universal stochastic sampling (13)) uses individuals resulting from reproduction (8) and the application of genetic operators (9) with probability pc and pm. Thus, a universal stochastic sampling (13) is performed, with this method also known as the roulette method for forming a third subpopulation. The last subpopulation is formed by a completely random method (14), that is, individuals are randomly generated using the matrix of potential solutions in isolation of anycast record. These two subpopulations, together with the subpopulations formed by the reserve elitism strategy (11) and the change of route in the best individuals (12) based on aptitude values, will form the new population (4). The concept of subpopulation was used to be efficient and to use different values of selection pressure.

When the selection pressure is high, it means that the chance of choosing more fit individuals is very high. An increase in deselection pressure decreases population diversity and vice versa. The universal stochastic sampling method (13) increases the selection pressure, which may cause premature GA convergence. To decrease the deselection pressure, the completely random method (14) is used. However, the best chromosomes of the present generation may be lost. This problem is solved by including the reserve elitism strategy (11). Finally, the route alteration strategy in the best individuals (12) is added to the selection process in order to assist in the search for the optimal global solution. The main differences in the present invention are the application of genetic operators before the formation of the third subpopulation, representation of the network-based nodes and the new population (4) formed by different amounts of each subpopulation: Subpopulation Number of individuals

1 Elitism reserve strategy 10%

2 Route change in the best individuals 10%

3 Universal Stochastic Sampling 20%

4 Completely random method 60%

The values shown indicate that the strategy used was to maintain population diversity through two subpopulations, universal stochastic sampling (13) and completely random method (14), representing 20% and 60%, respectively, of the formation of the new population (4). . In addition, the other two subpopulations (reserve elitism strategy (11) and route change in the best individuals (12)) will account for 10% each in the formation of the new population (4), so as to avoid individuals with good fitness values are lost.

After the new population is formed, each individual's fitness is assessed and ordered as described above. The stages of reproduction, application of genetic operators, formation of the four subpopulations for the formation of the new population, and evaluation of it, are performed until the closure condition of the GA is satisfied. The termination condition of the algorithm used was the number of generations, however other termination conditions may be used, such as the mobile device starting time that forms the first hop of each anycast session.

The number of generations required to achieve suboptimal routes is dependent on network conditions such as number of sessions, amount of traffic, node density, number of messages, and so on. GA-based routinganycast three definitions are used:

- For any number of anycast sessions ζ less than 4, 60generations are used;

- For a number of anycast sessions between 4 and 9, 120generations are used;

- For a number of anycast z sessions greater than 9, the following expression was used to define the number of generations (nger):

As an example, for anycast number of z sessions equal to 20, the number of generations (nger) used is 600.

The anycast routing algorithm on AG-based DTNs has the advantage of performing an appropriate combination of routes for routing decisions. Another positive feature of the algorithm is the use of alternative routes, preventing multiple traffic from using and competing on the same link in a graph. Thus, the algorithm is able to deal with the dynamic conditions of DTNs.

In addition, simulation-based experiments show that the strategy of limiting the number of potential solutions in isolation helps the AG-based routing algorithm converge considerably faster and produce similar results than when not limiting the number of solutions to be combined. by the AG. This behavior can be explained by the fact that when using a number of potential solutions in limited isolation, the algorithm search space is smaller, decreasing the time required for the algorithm to converge. Also, when not limiting the number of potential isolation solutions, containing large numbers of hops, which probably will not be part of the optimal solution, are used and combined by the algorithm, causing a high number of generations to be used. Another strategy used that proved to be efficient was the concept of subpopulation, which helped algorithm converging faster. Therefore, the present invention is based on the use of these two strategies together and the application of GA to anycast routing in DTNs, together with the coding scheme used, allows the anycast routing algorithm based on GAs to be efficient and respond well to the dynamism and challenges present. DTNs1 by finding suitable combination combinations for routing.

The present invention contributes by providing an efficient AG-basedanycast routing method on DTNs. The method is intended to be applied in developing regions lacking infrastructure, enabling the information technology revolution to be brought to populations in such restricted locations that have no communication at all. In addition, in many situations, a host, application or user may require a particular service or specific information. However, if several nodes, servers, or users support the service or have the required information, either one may be used. In this scenario, considering unicast service, that is, transmitting a source to a defined destination, this choice of only one destination would probably make communication impossible, since other possible destinations would be dropped. This opportunity to transmit to one, possibly the destination that presents the best communication conditions, among a group of possible nodes, allowing communication in scenarios in which the unicast service would often be unviable. Thus, NTDs represented by evolutionary graphs, coupled with the benefits of anycast service and the use of an efficient AGs1-based anycast routing algorithm are an attractive solution for providing communication in remote and rural regions. These unconnected regions, with no possibility of communication, can use messengers in cars, buses, motorcycles and / or trucks, equipped with storage devices, to act as carriers to deliver messages, with the routing algorithm being responsible for defining which messengers to forward messages to. In the representation through graphs used, the (fixed) nodes would be the regions without communication, and the edges (mobile devices) would be the messengers. Thus, this example described as an example, in addition to the potential applications of anycast in NTDNs and the efficient route selection AG-based routing algorithm, are a promising solution in using the ancast service to provide communication in rural and remote, underserved regions. communication infrastructure.

Claims (6)

1. Anycast routing method for path selection using genetic algorithm or AG in a delay and disconnection tolerant network or DTTs, comprising the steps of: a) communication of disconnected nodes (regions lacking infrastructure) using mobile messengers; (c) creation and limitation of potential isolation solutions for anycast registration obtained using algorithm that calculates the path as the least number of hops available to be combined by the AG-based anycast routing algorithm c) definition of DTN network nodes as a chromosome, and representation of the chromosome, using the potential solutions in isolation, by the nodes that form the route of each anycast session and indicate the mobile messengers through which the message is passed, d) assessing the suitability of individuals within the population with a minimum likelihood of delivering messages. and creation of the new population using the (e) definition of the mobile messengers to which messages are transmitted using information from the evolutionary graph; and f) selection of anycast routing paths based on measures of delay performance and message delivery probability. Anycast routing method according to claim 1, characterized in that step b) comprises the steps of: - using a modeled evolutionary graph to represent DTN network characteristics for calculating potential isolation solutions using the shortest path algorithm with baseline number of hops; e- limiting the number of potential solutions to five, reducing the search space of the algorithm in order to reduce the response time of the algorithm. Anycast routing method according to claim 1, characterized in that step c) comprises the steps of: - encoding the set of nodes that form the path between the source and destination of each anycast session on a single chromosome using assolutions. potentials of each session; e- representation may use zero fills to maintain constant chromosome size facilitating the generation of only regular individuals. Anycast routing method according to claim 1, characterized in that step d) comprises the steps of: - defining a minimum message delivery probability (PEm / n) (defined according to application needs) which the AG-based anycast routing algorithm will look for: - assessing the suitability and ordering of individuals in the population based on the calculation of the delay, the probability of delivery of ePEmin messages, the formation of the new population using four subpopulations (formed by different amounts of each subpopulation): strategy of elitism (10%), change of route in the best individuals (10%), universal stochastic sampling (20%) and completely random method (60%). Anycast routing method according to claim 1, characterized in that step e) comprises a step of determining the mobile messengers for the route of each anycast session, which will indicate to which nodes the message is forwarded, based on the paths represented in the chromosome. Anycast routing method according to claim 1, characterized in that step f) comprises an anycast routing path selection step based on the paths represented on the chromosome, in which a minimum delivery probability (PEmin) is sought and the delay It is minimal.