WO2009047439A2

WO2009047439A2 - Method for the failure tolerant distributed routing of data in a network

Info

Publication number: WO2009047439A2
Application number: PCT/FR2008/051654
Authority: WO
Inventors: François ELEOUET
Original assignee: France Telecom
Priority date: 2007-09-21
Filing date: 2008-09-16
Publication date: 2009-04-16
Also published as: WO2009047439A3; FR2921533A1

Abstract

The invention relates to a method for routing at least one piece of information between a source node and a target node in a network including a plurality of N nodes each connected to at least one neighbouring node by a link representative of a connection. According to the invention, the method comprises the step (10) of supervising the state of the connection of each node in said network with each of its neighbouring nodes, respectively, said method further comprising an iterative step for each of the nodes i (i ? [1..N]) of said plurality of nodes, for the sequential transmission to at least one neighbouring node of at least one message enabling said at least one neighbouring node to progressively build after iterations of said iterative step (12) a so-called path tree that comprises at least one path that does not run through said at least one neighbouring node and that can lead to the at least one target node, respectively.

Description

Distributed data routing method in a network, fault tolerant

The present invention relates to the field of telecommunications networks. More specifically, the invention relates to a distributed routing method and system for rapidly re-routing traffic (for example, data streams, information packets) during transit in a telecommunications network in the form of a plurality of interconnected nodes, in case of failure in the network. The invention can in particular be implemented in applications for routing and re-routing of data in networks of the packet-switched type.

Many routing protocols rely on defined routing algorithms to facilitate routing of data in packet-switched networks.

In general, such routing algorithms are designed to answer the same problem: in a graph composed of nodes connected by edges, each stop (and not node) having a certain cost (or weight), it is necessary to determine a path minimizing the total cost of stops crossed. This cost may depend on various parameters, such as bandwidth, reliability, delay, etc., or more generally a combination of these different parameters.

Reported at the node level, this is to know which neighbor to send data to another node to minimize the cost of traversing the network, to route data from a source node to destination a target node in the network.

In networks of the IP type (for "Internet Protocol" in English, or

"Internet Protocol" in French), RIP protocols are mainly used (for

Routing Information Protocol in English, OSPF (for Open Shortest Path First), ISIS (for

Intermediate System to Intermediate System "in English), BGP (for" Border Gateway Protocol "in English or") and EIGRP (for "Enhanced Interior Gateway Protocol").

These different protocols can be grouped into two major families: link state protocols (OSPF & ISIS) and distributed computing protocols (RIP, BGP, STP, EIGRP).

Link state protocols use a broadcast technique that allows different nodes to gain a complete view of the topology of a network and thus run the shortest path algorithm of Djikstra, better known as SPF algorithm, for "Short Path First" in English. This algorithm allows each constituent node of a network architecture to be able to determine which route it must choose to join the other nodes of the network. Such link state protocols provide fast convergence, typically of the order of a few seconds. They are deployed in most operators' telecommunications networks. However, a disadvantage of this category of protocols lies in the fact that they do not always make it possible to ensure the routing of the information packets during the periods of convergence.

In addition, a second disadvantage of the link state routing protocols arises from the fact that the information dissemination relating to changes in network topology is independent of the calculation of the routes that an information packet can take between a source node and a network. a target node, micro-loops can be formed during these information dissemination phases, which consequently implies a degradation of the service, which degradation can cover the form of interruption of the routing of the data in the network on certain periods of time. However, if such interruption times remained tolerable in the so-called "best effort" use of IP-type networks, they are no longer possible in the context of the implementation of services of the IP telephony type (or ToIP, for "Telephony over IP" in English) or the type of television over IP (or IPTV, for "Internet Protocol Television" in English), in which the continuity of the service provided to users must be guaranteed.

In an attempt to overcome the aforementioned drawbacks, the MPLS protocol (for "Multi Protocol Label Switching" in English) has introduced the notion of a path in IP networks, so as to allow routing packets on alternative paths in situations of topology change of the communication network in use thereof.

However, a disadvantage of such a technical solution is related to its complexity of implementation. It is therefore poorly suited for collection networks, in which the number of hardware devices to be managed is generally important, such as those covered by the present invention.

For distributed computing routing protocols, routers do not have a global view of the network. They are content to announce to their neighbors information on the best known path to reach each destination in a given communication network, which allows each router constituting the topology of the communication network to get closer to each other. knowledge of the best route (or path) to reach each destination within the network. In the algorithm of the prior art of Bellmann-Ford, from which the protocol

RIPs, routers exchange vectors indicating the cost of each destination in the network. This is called a routing protocol with distance vectors.

A disadvantage of such protocols is that they nevertheless suffer from an infinite counting problem because of their looping distribution of invalid routes (or paths) which substantially penalizes their ability to converge in the event of a failure in the network.

To overcome this last disadvantage, some approaches known as the path vector algorithm and from which the BGP protocol originated, consisted in indicating the routes only possible to reach each destination in a given network. These notably make it possible to invalidate all the routes that would be detected as being in a loop.

Specifically, such path vector protocols rely on the consistent approach, rather than associating with each route the entire network link

(connection between two nodes of a network) traversed, to associate with each node located on the road taken, the characteristics of the last link used to join said node. Thus, once put end to end, the set of different links to determine the distance of the route that a packet of information will have to travel on the network between a source node and a target node of the network.

However, if such path vector or link routing algorithms provide convergence times substantially equivalent to those of the link state algorithms, while providing reduced implementation complexity and loopless routing at all times , they do not allow the re-routing of information packets during their convergence phases in the event of a network failure, which is a significant drawback with regard to the implementation of services of the ToIP or IPTV type, as mentioned above .

The present invention provides a solution that does not have the drawbacks mentioned above

The invention aims to solve the aforementioned drawbacks by proposing a distributed routing method, taking advantage of the simplicity of implementation of the link vector protocols, but also, which is adapted to ensure the re-routing of the information packets. in a network architecture, including in case of failure in the latter.

It is particularly by means of the present invention to be able to control and ensure continuity of service, especially for services of the ToIP or IPTV type, and therefore to propose a new network routing protocol that is fault-tolerant. For this purpose, the invention relates to a method of routing at least one piece of information between a source node and at least one target node in a network comprising a plurality of N nodes each connected to at least one neighboring node by means of a node. representative link of a connection. According to the invention, said method comprising a step of supervising a state of the connection of each node of said network with each of its neighboring nodes, respectively, it also advantageously comprises an iterative step, for each of the nodes i, i G [1..N]) of said plurality of nodes, of transmission step by step, at least one neighboring node, of at least one message enabling said at least one neighboring node to build as iterations of said iterative step, a tree, said path tree, comprising at least one passing path not by said at least one neighboring node and making it possible to reach said at least one target node, respectively.

Path tree means here any related acyclic graph

Thus, the specificity of the routing method according to the invention resides first and foremost in its ability to allow the detection of network connection drops between constituent nodes of a network architecture, the supervision step making it possible to detect any anomaly concerning a network connection between two nodes constituting the latter, for example a fall of the network connection, or even a failure of a hardware device forming a node of the network (a router or a server, for example).

This is particularly interesting in that such a detection authorized by the implementation of a supervision of the state of the network connections of each node of a network architecture with each of its neighbors, makes it possible to anticipate the re-routing data or packets of information on another route (or path) than that impacted by the fall of a network connection. Such an anticipation possibility renders the routing method according to the invention fault-tolerant and confers on it an ability to ensure and guarantee the continuity of the service provided to the users, even in the event of breakdown of certain connections and / or network equipment ( temporary or definitive) constitutive (fs) of a given network architecture. Another advantage of the method according to the invention also lies in the selection of paths (a path consisting of a succession of network links to ensure the connection between two nodes) transmitted by a node to its different neighbors.

Indeed, a node advantageously uses the advertisements or messages received from each of its neighboring nodes, respectively, to construct, respectively for each of said neighboring nodes, a tree of paths comprising a predetermined number of paths not passing through said at least one a neighboring node and to reach said at least one target node.

Also, each node of the network thus maintains in a routing table of its own as many path trees as it has neighboring nodes, which advantageously allows it to have several routes to reach the same destination, and so to have at least one valid alternative path in case of failure in the network (drop of a connection, temporary or permanent failure of a hardware device, for example).

In a preferred embodiment of the invention, a command relationship being defined relative to the different paths contained in said path tree, said transmitted message contains at least one information representative of a shorter path to reach said path. least one target node, from said source node.

By order relation, is meant any information relating to a predetermined metric and for evaluating and / or determining that one path will be better than another path. Such information may advantageously correspond, by way of illustrative and nonlimiting example, to a serial number of a router, or even the cost of a network link between two nodes of a network, such a cost being able to for example depending on a bandwidth level, jitter and / or a percentage of loss of information packets on such a link. Moreover, the information representative of a shorter path advantageously makes it possible to discriminate relevantly in the choice of an alternative path to be used preferentially when an anomaly is detected in the network, which was not possible. until now not possible to obtain with the solutions of the prior art. Preferably, each of said nodes of said plurality of nodes contains, for each of said neighboring nodes, at least one path tree in the form of a routing table able to direct at least one message received by said node on a path to reach said target node from said source node

Thus, each node of the network, once its routing table filled, can then rebroadcast to each of its neighbors the best route not passing through this neighbor, which advantageously avoids unnecessary information transmission between neighboring nodes in the network, but above all, lead to the dissemination of alternative paths that may be used in the event of a network connection being broken between nodes of the network. Such a technical approach according to the invention then makes it possible to avoid any loopback situation or rebound phenomenon as in the known solutions of the prior art. In a particular embodiment of the invention, said supervision step is performed by a device for supervising the connectivity of each node of said plurality of nodes of the network with each of said neighboring nodes, respectively. Such a supervision device will also be connected to the network and will have an ability to detect any anomaly in the state of the network connection between the constituent nodes of the latter, or any operating anomaly of a device forming a constituent node. of such a network.

An advantage of such a particular embodiment of the invention resides in particular in the possibility of making the routing method according to the invention benefit from already existing network architectures which would be devoid of any fault tolerance mechanism. .

In a preferred embodiment of the invention, said supervision step is performed by each node of said plurality of nodes. A first advantage of the implementation of said supervision step by each node of the network resides in the ability of each node to be able to detect directly and in an optimal manner, in particular with regard to the detection delay, any anomaly likely to concern its own network connections with each of its neighbor nodes. As a result of this first advantage, each node that has detected an anomaly of the state of the connection with at least one of its neighboring nodes can advantageously anticipate the re-routing of data or information packets on a route other than that impacted by the fall of the network connection by informing prior to this re-routing, each of its nodes see said anomaly detected, the latter can then take into account the latter by updating accordingly their respective routing tables. Such an approach will further promote fault tolerance and will ensure and ensure continuity of service.

Advantageously, when at least one node of said plurality of nodes detects in said supervision step an anomaly of a state of the network connection with one of its neighbors, a verification step is performed to determine if said anomaly is related to a malfunction of said neighbor node and / or a malfunction of the network link between said node and said neighbor node.

Such a verification proves to be interesting in the sense that it makes it possible to distinguish initially between untimely failures and definitive failures that may occur on the network and, consequently, to decide to trace relevant information to the other nodes of the network in question. taking into account the type of failure that has occurred and is detected by a network node.

Thus, by way of illustrative and nonlimiting example of the foregoing, if a node detects an anomaly relating to its network connection with a first neighbor node, it will check at said verification step if this anomaly corresponds to a failure hardware of said neighbor node, or if it corresponds to a temporary breakdown of the connection link with the node.

Assuming that it is not a hardware failure of the neighboring node, but an anomaly of connectivity with the latter, then no information will be sent step by step by the node to the other nodes of the network which would have a link with the first node involved in said anomaly, these other nodes can a priori continue to exchange information with this first common neighbor node.

In the opposite hypothesis, information would be sent back step by step by the node at the origin of the detection of said anomaly towards the other nodes of the network also having as neighboring node this same first neighboring node, so that these latter can update their respective routing tables, so as to prohibit any subsequent rebroadcasting of information packets on paths that would pass through this first neighbor node.

Preferably, if at said verification step it is determined that said anomaly concerns a malfunction of said neighbor node, then, the following steps are performed:

"identification and removal in the routing table associated with said node of all the paths passing through said neighbor node impacted by said anomaly and no longer being able to be used to transmit said at least one information item to said at least one target node; "transmission by said node to at least one of its other neighboring nodes distinct from said neighbor node impacted by said anomaly of at least one notification message of said anomaly;

said neighbor nodes update their respective routing tables, to indicate the paths impacted by said anomaly and not being able to be rebroadcast to the other nodes of said plurality of nodes; "rebroadcasting said at least one notification message from said plurality of nodes; said anomaly, step by step by each of said other neighboring nodes, to at least one of their respective neighboring nodes and distinct from said node impacted by said anomaly; said updating and redistributing steps being iterated until all the routing tables of each of the nodes of said plurality of nodes are updated.

Thus, in said updating step, the whole of the part of the tree starting from the node concerned by said anomaly is removed (for example by truncation) from the routing table and on the other trees, it is ensured that not repost links that are no longer valid.

A technical advantage of such an approach lies in the fact that the alternative paths broadcast step by step to the different nodes of a network will make it possible specifically to deal with the network connection drops that may occur between nodes of the network, which allows to make the network routing according to the invention fault-tolerant, which is an essential condition, particularly with regard to the guarantee of routing data / information packets, for example and without limitation, in services of ToIP or IPTV (IPTV).

Thus, when a node detects a fall of the network connection, see more generally an anomaly in the network communication, with one of its neighboring nodes, it is not able to determine the origin of this anomaly, namely for example distinguish between the fall of the network connection or a failure of the network equipment (a router or a communication terminal equipping the neighboring node).

The node having detected the anomaly will then indicate in its routing table which is the neighboring node impacted by the anomaly detected, as well as all the routes of its routing table passing through this neighboring node, so as to avoid redistributing it to its other neighbors, any road passing through the neighboring knot impacted by the anomaly. The node having detected the anomaly will then propagate to these other neighbors a notification message of the anomaly, which will then update their own routing table taking account of the latter, before rebroadcasting this notification message step by step. to their own neighboring nodes, so that at the end of the process, all the nodes of the network have modified their own view of the network topology taking into account the fall of the detected network connection, until the links and / or the node impacted by the network connection drop has disappeared from the set of routing tables of all the nodes constituting the network architecture considered (or possibly that a node informs the other nodes that the node initially detected as impacted by this anomaly has become operational again).

In a preferred embodiment of the invention, said method comprises a step of updating by said source node of a transmission table containing a set of weakest weight paths transmitted by said source node to each of its nodes. so-called neighboring nodes. Such a table makes it possible to keep track of the advertisements issued to the different neighboring nodes, which firstly makes it possible to avoid the unnecessary transmission of packets (when the entry of the routing table to be announced has not changed) but also to ensure the switching of the packets.

Advantageously, when a node of said network receives from a first neighbor node an information packet intended for a target node of said network, said node executes a step of switching said information packet to a second node distinct from said first neighboring node and located on a shorter path leading to said target node, said shorter path having been transmitted by said second node to said first neighbor node at said transmitting step. Since the paths are always acyclic, such a technical approach makes it possible to switch the packets of information exchanged in the network that is loop-free at any time, even if the routing of the packets is not done according to a plus. short path (lowest weight path). This makes it possible to ensure continuity and consistency of quality of service in the routing of data, even for routing information in networks with variable quality of service. The invention also relates to a communication terminal intended to be used in a node of a network for routing at least one piece of information between a source node and a target node, said network comprising a plurality of nodes each connected to a node. least one neighbor node by means of a connection. Advantageously, such a terminal (for example and without limitation, a router or server equipment) comprising means for monitoring a connection state of each node of said network with each of its neighboring nodes, respectively, said terminal comprises in besides means for transmitting, step by step, iteratively for each of the nodes i, i G [1..N] of said plurality of nodes, and to at least one neighboring node, at least one message enabling said auditing at least one neighbor node to build a tree, said path tree, comprising at least one path not passing through said at least one neighboring node and making it possible to reach said at least one target node, respectively.

The invention also relates to a communication system comprising a plurality of nodes each connected to at least one neighboring node by means of a connection link in a communication network adapted for the transport of at least one piece of information between a source node and a target node, the system being characterized in that said nodes are each equipped with a communication terminal as mentioned above. The invention also relates to a computer program product downloadable from a communication network and / or stored on a computer readable and / or executable information medium by a microprocessor, characterized in that it comprises code instructions for the execution of a routing method according to the routing method according to the invention described above. It also relates to a recording medium readable by a computer on which is recorded a computer program comprising instructions for performing the steps of the routing method according to the invention.

The characteristics and technical advantages of the present invention will become more apparent from the following description, given by way of non-limiting indication, with reference to the appended drawings in which:

FIG. 1 is a flow chart summarizing the major implementation steps of the routing method according to the invention; FIG. 2 serves as a basis for the description of the processing of the increase of the cost of the paths in a communication network by the routing method according to the invention;

FIG. 3 gives an exemplary implementation of the routing method for routing information packets in the event of a drop in the connection with a node of the network of FIG. 2;

FIG. 4 gives an example of implementation of the information packet switching of the routing method according to the invention following the detection of a network connection anomaly with a node of the network of FIG. 2; FIG. 5 presents a flow chart detailing the packet switching principle implemented in the switching step 18 according to the invention, presented in FIG.

The present invention therefore proposes a new distributed routing technique, based on link state broadcasting, which allows, on the one hand, to ensure the re-routing of the traffic transiting in a communication network, even in the presence of breakdowns. (drop or loss of the connection between nodes of the network, failure of a device in a node of the network, for example) likely to occur in the network and which, on the other hand is free of loops, at any time, what allows maintaining quality of service in the routing of data, essential point, especially with regard to the implementation of ToIP services, or IP TV, including te but not limited to.

The distribution of the calculation of the roads makes it possible to limit the necessary power at the level of each node.

In addition, since nodes only need a partial view of the network, topology changes do not always affect the entire network, which helps to increase the scalability of this process. routing according to the invention.

The latter is also free of any timer or parameter, which simplifies its implementation. The algorithm proposed through the method according to the invention can serve as a basis for the development of new simple protocols, making it possible to share the extended load, while ensuring the distribution of labels and the maintenance of connectivity in the network, even in the event of faults or anomalies in the network.

A possible embodiment of the routing method according to the invention is described below. In the remainder of this description, a network is likened to a set of nodes, connected by point-to-point links, with at most one link between two same nodes. Each node has a unique identifier (that we note ID).

We assume that these nodes exchange messages according to a method to ensure their routing without errors, respecting their order and in a finite time as long as the link is operational.

We also assume that the nodes have a mechanism to detect any changes in the operational state or characteristics of a link within a finite time period.

Nodes exchange ads composed of links used in their routing tables, each ad issued by a node K to its neighbor node /, noted

V ^κ is a vector composed of a variable number of links Vf, ordered according to their distance to the transmitter, comprising the following information:

"the type of message, specifying whether it is an addition, an update, a deletion or a notification of a link drop;" the characteristics of the link, namely: node J for which the neighbor announces a route (ie is the head of the link); o the predecessor node P of this node J in the routing table of the issuer (ie the neighbor node of J by the link considered having initiated the diffusion of this path); o the weight of this link or the cost of the node for the neighbor. These two pieces of information are indeed equivalent because, by putting the cost of the different links received end to end, the receiver can deduce the cost of the path leading to the node J, and conversely, if it receives the cost of the path leading to J, he can deduce the cost of the link J - P by deducting the cost path leading to P. To ensure the operation of the routing method according to the invention, each node maintains a link table, a routing table and a transmission table.

The links table, denoted L in the rest of the description, contains the various operational links to which a node / is connected, as well as their respective costs. It also contains the neighboring nodes to which the / node is connected by means of these different links (network connection).

Subsequently, L ^κ represents the characteristics of a link / <-> K (link between the node / and a neighboring node K) as stored in the node /. In the routing table, denoted T, is stored for each node a set of links corresponding to the different announcements received from its neighbors.

For each node J of the network, the node / may receive an announcement from each of its neighbors K containing a route that it stores in its routing table in the form of an entry Rf containing the following information: "the cost of the advertised route;

"the characteristics of the link, received in V ^κ : o the considered node o the predecessor node of the considered node o the cost of the link between the considered node and its predecessor neighbor node;

"the neighbor node from which this announcement originates (ie the K node);" the state of this link; o enabled (or "Active"); o busy (or "pending"); o invalid (or "invalid").

These different informations are referenced in two ways:

"first of all, for each node J of the network, we order the Rf in the table

C _} according to their increasing weight (in the case of a tie, we separate the ads in favor of the one learned from the neighbor with the smallest ID). At this table are also associated two parameters: o the state of this node, namely "active" or "pending"; o a marker to determine if this node has undergone a change, in which case the announcements to be sent to the neighbors for this node must be calculated; "Then, for each of the neighbors K of /, the different Rfs chained together constitute a tree G ^κ (each entry has a pointer allowing access to the entry corresponding to the predecessor, as well as a set of pointers allowing access to the successors of this node, ie the different nodes having this entry as predecessor.) Note Gf the sub-tree of G ^κ consisting of Rf and all of its successors (the departing branch of

The send table makes it possible to keep track of the announcements made to the different neighbors, which firstly makes it possible to avoid the unnecessary transmission of packets (when the entry of the routing table to be announced has not changed ) but also to ensure the switching of the packets. Whenever the node / sends a route Rf for a node J to a neighbor node K, an entry Sf comprising the following information is created:

"the characteristics of the link, contained in Rf: o the considered node o its predecessor node o the cost of the link;

"the neighbor from which this announcement originates (ie node X);" a marker to force the update broadcast even if the characteristics have not changed. At the initialization of the routing method (or algorithm) according to the invention, a node adds its own entry to its routing table and sends an empty announcement to each of its neighbors.

On receiving an announcement from a neighbor node K, the node / checks whether the entry Rf corresponding to the link L ^κ is present in C _κ . If it is not the case, it adds it (the weight of this road being equal to the weight of the link). On receipt of a message V ^κ , the node / adds or updates the links Rf entries of its routing table corresponding to different Vf.

It then calculates the weight of the road by adding the weight of the link of Rf to that of the route Rf, where P is the predecessor of J in Rf. Note that this iterative approach requires a scheduling of Vf according to their respective costs, because one could not add Rf to the routing table if the latter does not contain Rf.

The node / then updates the cost of each of the nodes of Gf.

Once the various messages have been processed, the node calculates the different announcements V ^κ that it must send to each of its neighbors K in its table of links L, by calculating for each node J impacted by the previous update which entry Rf he must send to this neighbor.

The node / must send to a neighbor the entry Rf having the least weight, such that K is not in the set of predecessor nodes of Rf connecting this entry to the node / (ie that the node K must not to find yourself on the connecting path

The different V / are then ordered according to the increasing cost of the corresponding Rf before being sent to the neighboring node K.

This convergence phase finally uses a mechanism similar to those of other link vector / path algorithms. It differs nevertheless in the use of the notion of path. Indeed, in the algorithms of the prior art, a node checked at the reception if an ad had made a loop, examining if the path went through. These roads did not bring any information to the receiver. By adding this additional constraint of taking into account the transmission of the paths in the communication network, thus forcing a node / to send to its neighbor node K links belonging to the tree of the shortest path of the network not passing by K, which, step by step, will cause the nodes to broadcast alternative paths. In the routing method according to the invention, it is assumed that all the nodes are initialized simultaneously. They first emit empty advertisements to their different neighbors who can then add their different links in their routing tables. At the next iteration, the nodes can diffuse the links connecting them to their different neighbors.

At the end of the intelligence phase of the routing tables of the different nodes of a network, each of the nodes of the network has a plurality of different paths allowing it to join each of the other nodes of the network, which by means of a method appropriate switching, will allow it to ensure routing to each of these nodes in case of drop of link (network connection for example) with one of its neighbors.

Indeed, when a node loses connectivity with one of its neighbors, it can not detect whether it is only a link drop, or a complete fall of the node, in which case it must not to broadcast paths passing through this node to its different neighboring nodes.

To do this, the routing method according to the invention makes it possible to check the state of a node, while ensuring that the paths rediffused during this verification are valid. In more detail, when a node / receives a link drop notification, it puts in its routing table the node J impacted by this fall (the head of the link which has fallen) to the "pending" state, which means that the state of this neighbor node is unknown at this point. It also indicates in its routing table all the entries whose path passes through this "pending" node, thus preventing their rediffusion (for any K in ¹ L, and for each node N of the network, have puts

Rf in the "pending" state if Rf e Gf).

However, if a node J is connected to a node / by an operational link, / will keep this neighbor in the "activated" state, which will mean that the node has not fallen.

On the other hand, when a node / receives a link drop notification from a neighbor node K, it marks the entry Sf to allow confirmation of the link already issued. Thus, if the node J goes back to the "activated" (or "active") state in the routing table of the node / and this node informs it step by step of its other neighboring nodes which will consequently update their routing tables , respectively.

Then, when / emits an update to its neighbors, it relays this link drop notification only to the neighbor who has announced the best route to the node / (the first entry of C ₇ ).

If it no longer has a route to this node, it issues this notification to all the nodes to which it had previously issued an announcement.

For successors S J, also affected by this link drop, it rebroadcasts Rf <≡ C _s that if the entry is not in the "pending" in their routing table.

If / does not have an entry satisfying these criteria (ie if all the paths to a target node S pass through the node J), it indicates to its neighbor node K having announced to it the best route to the target node S of delete the entry he had previously sent to him.

Finally, for the algorithm to converge, it is also necessary that the node

/ can return node J to the active state to start broadcasting again paths that pass through it (for example if it were only a link drop, or if the node is active again). This transition can be done if it receives a link announcement for this node such that the neighbor K that announces this link is on the shortest path to the predecessor node P of this link (or if R _p is the best entry of C _P ).

To limit the number of messages exchanged during this phase, it is possible to implement different methods, it can be verified, when receiving a link drop notification, if the predecessor indicated in this link is not already active, to which case it is a link drop, which allows to reactivate the two nodes. One can also check that this notification comes from the neighbor having announced the best way for this link.

The example in Figure 3 illustrates this mechanism. In Figure 3, it is assumed that the bold links have a cost of one and the links a cost of ten. When the node A falls, the node D detects the fall of the link A <- D and puts this node in the state "pending" in its routing table, as well as all the roads which cross it.

It then sends a notification to its neighbor node E, which is on the shortest alternative path to A.

Simultaneously, it also broadcasts the link B <- C because the old route B <- A <- D is no longer available, while the new B <- C <- D does not pass through the node A.

It also tells E to remove the paths to G, H, /, J, K, L, and M (which he previously sent) from his routing table, because E is now on the best alternate path of D for these nodes.

When node E receives this update from node D, it also puts node A in the "pending" state in its routing table, before updating its route for node B and removing in its table of routing the routes for the nodes G, H, I, J, K, L and M passing through the node D.

It then rebroadcasts the link drop notification message (connection anomaly) to the / node, but it does not rebroadcast it to the F node, because this node is not located on the best path to the A node. It sends only the update for node B. Meanwhile, J has also sent a notification to / along with an update for the M node. The node / therefore takes all the paths leading to the node A and removes this node from its routing table. In doing so, it sends a notification to the nodes to which it had previously sent links leading to the node A, namely J, H and E, and so on ... It will be noted that in this example, the various nodes of the network keep permanently a route to the other nodes of the network, which ensures routing even during the convergence period.

In order to guarantee and maintain a continuity in the quality of the service offered to the users of the communication network, in particular with regard to services of the ToIP or IPTV type, even in the event of a fall of a connection between nodes of a considered network, it is also important to address the problem of increasing paths when a network anomaly is detected, as shown in the example of network of FIG. 2 in which an anomaly occurring at node B is considered.

In this example and only for simplification reasons, each link is considered to have a weight of one. In this example, the node C, when it detects the failure, can send an update to the node E by telling it to use the link A <- I to join the node A (this entry is not set to the "active" state in its routing table, since the path leading to it does not go through B).

However, node C does not send a link drop notification because node D is not located on the shortest path to B.

The node D thus has in its routing table an invalid path leading to the node A and passing through the node B, which it can rebroadcast to the rest of the network because this path has a cost of 5, lower than that of the alternative path 2 announced by node C (cost of 6). This is because D has not yet received the update from E indicating the link A <- /. However, he can easily infer that the link A <- B is not usable, since C has replaced the path 1 by the 2 (of higher cost) in his routing table, which indicates that the path 1 is no longer usable due to detected anomaly impacting network node B. When the cost of a route between a source node and a target node of the network increases, this indicates that the path to that target node is no longer valid.

Consequently, if a source node / receives an update of the path to a target node J of a neighbor node K involving an increase in the cost of the route Rf between the nodes J and K, then it must invalidate all the entries of its routing table Rf <≡ C ₇ having a higher cost and using the same link as

Conversely, if after updating an entry in its routing table, there are other higher cost entries having the same characteristics as the latter (same link weight and same predecessor), these entries must be reactivated if they are invalid. An invalid Rf entry of a routing table will therefore systematically have an infinite cost and can not be used for switching or rebroadcast to other nodes of the network.

To take full advantage of the routing method according to the invention, it is necessary to switch packets specifically, as described below.

Firstly, a node / must switch the packets to J according to their input interface, according to the following rule: a packet received from a neighbor node K must be sent to the node (the interface) specified in the vector Sf, that is the node by which the node / learned the path it has rebroadcast to the neighbor node K. Since paths are always acyclic, this method ensures loop-free switching at all times, even if routing is not done in the shortest way.

It can be noted that this notion of path can be easily used to implement label switching. One could indeed add in each vector of links V / label, that the different nodes would then use to switch packets.

However, during the convergence phases, it may be that when a node receives a packet from a neighbor node K to a target node J, it no longer has an entry Sf. In the example of Figure 4, which assumes that the source node

F sends packets to the target node M, when the node A falls, it is a certain time before the link M <- L is announced to the node F.

During this time this path will not be established in the nodes of the network (they did not all emit a vector of links V ^ corresponding to this link). To ensure routing during this phase, it is necessary to switch the packets to other criteria.

Thus, at the time of the fall of the node A, the node F sends the packets to M according to the path No. 1.

When these packets arrive at the node D, the node can not route them normally since at first the output interface corresponding to Sf ₁ has fallen, and this entry is then suppressed on the emission of V ^E

During this period, however, node D has two routes to reach target node M (which are in the "pending" state), so it could use path # 2 (the best of these two routes) to re-route the traffic.

However, when the information packets arrive at the node J, the problem is the same since the entry S _M 'no longer exists, the node J must then redirect the packets on the path n ° 3, which will lead them finally to the target node M. However, if nodes are allowed to forward packets in this way, micro-loops will appear during the convergence phase. Indeed, if the path n ° 3 did not exist, the node J would in turn reroute the packets towards the node /, which would have led to the formation of a loop.

To prevent the formation of this type of loop, the routing method must limit the number of re-routings along roads with a "pending" state. The number of times that packets can be forwarded on a "pending" route (when it can not be switched according to the S table) will be a compromise between the ability of the algorithm to route packets to the destination. fault in the network and its tendency to retransmit packets unnecessarily if there is no alternative path.

To measure the number of reroutes, it will be necessary to introduce a method for measuring the number of times a packet has been rerouted on a road in the "pending" state, for example by incrementing a counter in the packets, or by associating different tunnels to the same road, each of these tunnels corresponding to a number of re-routing.

Thus, it is ensured that the packets are sent on paths that are free of loops, while guaranteeing the maintenance of the information packet routing service between a source node and a target node of the network.

FIG. 1 summarizes the major implementation steps of the routing method according to the invention, namely: a step 10 of supervising a state of the connection of each node of said network with each of its neighboring nodes, respectively;

"a verification step 11 to determine whether said anomaly is related to a malfunction of said neighbor node and / or a malfunction of the network link between said node and said neighbor node;

an iterative step 12, for each of the nodes i, i G [1..N]) of said plurality of nodes, of transmission, step by step, to at least one neighboring node, of at least one message allowing said least one neighbor node to construct as iterations of said iterative step 12, a path tree comprising a plurality of paths not passing through said at least one neighboring node and making it possible to reach said at least one node target, respectively

If in said verification step (11) it is determined that said anomaly relates to a malfunction of said neighbor node, then, the following steps are performed:

13i identification and withdrawal 13 ₂ in the routing table associated with said node of all the paths passing through said neighbor node impacted by said anomaly and can no longer be used to transmit said at least one information to said at least one target node; "transmission 14 by said node to at least one of its other neighboring nodes distinct from said neighbor node impacted by said anomaly of at least one notification message of said anomaly;

updated by said other neighboring nodes of their respective routing tables, to indicate the paths impacted by said anomaly and not being able to be rebroadcast to the other nodes of said plurality of nodes;

repeating said at least one notification message of said anomaly, step by step by each of said other neighboring nodes, to at least one of their respective neighboring nodes and distinct from said node impacted by said anomaly; update and 16 are iterated until all the routing tables of each of the nodes of said plurality of nodes are updated. The routing method according to the invention further comprises:

"step 17 of update by said source node of a transmission table containing a set of paths of the lowest weight transmitted by said source node to each of said neighboring nodes;" a step 18 of switching said packet of information to a second node distinct from said first neighbor node and located on a shorter path leading to said target node, said shorter path having been transmitted by said second node to said first neighbor node at said transmitting step 12. As mentioned above, the distributed routing method according to the invention, based on the link state broadcast, makes it possible to ensure a re-routing of the traffic (information packets, for example), even in the face of the occurrence faults or failures that can occur in the network, while being free of loops at any time, which allows to maintain a continuity in the services based on the communication network and to offer a constant quality of service to the users , which is essential for certain services, for example and in a non-limiting manner of the ToIP or IPTV type.

The distribution of the calculation of the roads makes it possible to limit the necessary power at the level of each node. In addition, since nodes only need a partial view of the network, topology changes do not always affect the entire network, which helps to increase the scalability of this process. distributed routing according to the invention. In addition, the latter being free from any timer or timing parameter, its implementation in a communication network architecture is very simple and can be expensive.

Such a method could therefore also be used to develop new simple protocols, allowing for extensive load sharing, ensuring the dissemination of labels and to maintain connectivity in the network in case of fault.

Claims

claims

A method of routing at least one piece of information between a source node and at least one target node in a network comprising a plurality of N nodes each connected to at least one neighboring node by means of a link representative of a connection, characterized in that said method comprising a step (10) of monitoring a state of the connection of each node of said network with each of its neighboring nodes, respectively, it further comprises a step (12) iterative, for each of the nodes i, i G [1..N]) of said plurality of nodes, of transmission, step by step, to at least one neighboring node, of at least one message enabling said at least one neighboring node to build as and when measuring iterations of said iterative step (12), a tree, said path tree, comprising at least one path not passing through said at least one neighboring node and making it possible to reach said at least one target node, respectively.

2. Method according to claim 1, characterized in that a relation of order being defined relative to the different paths contained in said path tree, said transmitted message contains at least one information representative of a shorter path to reach said at least one target node, from said source node.

3. Method according to any one of claims 1 and 2, characterized in that each of said nodes of said plurality of nodes contains, for each of said neighboring nodes at least one tree of paths in the form of a table routing adapted to direct at least one message received by said node on a path to reach said target node from said source node.

4. routing method according to any one of claims 1 to 3, characterized in that said step (10) of supervision is performed by a device for monitoring the connectivity of each node of said plurality of nodes of the network with each of its so-called neighboring nodes, respectively.

5. Routing method according to any one of claims 1 to 3, characterized in that said step (10) supervision is performed by each node of said plurality of nodes.

6. Routing method according to claim 5, characterized in that when at least one node of said plurality of nodes detects at said step (10) of supervision an anomaly of a state of the network connection with one of its neighbors, a verification step (11) is executed to determine whether said anomaly is related to a malfunction of said neighboring node and / or to a malfunction of the link network between said node and said neighbor node.

7. Routing method according to claim 6, characterized in that if said verification step (11) is determined that said anomaly relates to a malfunction of said neighbor node, then, the following steps are performed:

identification (130 ^and withdrawal (13 ₂ ) in the routing table associated with said node of all the paths passing through said neighboring node impacted by said anomaly and no longer being able to be used to transmit said at least one information item to said at least one target node;

"transmission (14) by said node to at least one of its other distinct neighbor nodes of said neighbor node affected by said anomaly of at least one notification message of said anomaly;" update (15) by said other nodes neighbors of their respective routing tables, to indicate the paths impacted by said anomaly and can no longer be rebroadcast to other nodes of said plurality of nodes;

"rediffusion (16) of said at least one notification message of said anomaly, step by step by each of said other neighboring nodes, to at least one of their respective neighboring nodes and distinct from said node impacted by said anomaly; 15) and (16) replay is iterated until all the routing tables of each of the nodes of said plurality of nodes are updated.

8. Routing method according to any one of claims 1 to 7, characterized in that it comprises a step (17) for updating by said source node of a transmission table containing a set of weight paths. the weakest transmitted by said source node to each of said neighboring nodes.

9. A routing method according to any one of claims 1 to 8, characterized in that when a node of said network receives from a first neighbor node an information packet for a target node of said network, said node performs a step (18) of switching said information packet to a second distinct node of said first neighbor node and located on a shorter path leading to said target node, said shorter path having been transmitted by said second node to said first neighbor node at said transmission step (12).

A communication terminal for use in a node of a network for routing at least one information between a source node and a target node, said network comprising a plurality of nodes each connected to at least one neighboring node by means of a connection, characterized in that said terminal comprising means for supervising a state of the connection of each node of said network with each of its neighboring nodes, respectively, the terminal further comprises means for transmitting near in near, iteratively for each of the nodes i, i G [1..N] of said plurality of nodes, and to at least one neighboring node, at least one message enabling said at least one neighboring node to construct a tree, said path tree, comprising at least one path not passing through said at least one neighboring node and making it possible to reach said at least one target node, respectively.

A communication system comprising a plurality of nodes each connected to at least one neighboring node by means of a connection link in a communication network adapted for transporting at least one information between a source node and a target node, the system being characterized in that said nodes are each equipped with a communication terminal according to claim 10.

12. Computer program product downloadable from a communication network and / or stored on a computer-readable and / or executable information medium by a microprocessor, characterized in that it comprises code instructions for the execution of a routing method according to any one of claims 1 to 9.

A computer-readable recording medium on which is recorded a computer program comprising instructions for performing the steps of the routing method according to any one of claims 1 to 9.