JP5892545B2 - Dynamic tiering method in network - Google Patents

Description

The present invention relates to a dynamic tiering method in a network that can cope with a sudden increase or decrease in traffic.

A wireless mesh network (Non-Patent Document 1) is a network in which nodes are connected in a mesh shape to construct a wireless backbone. The cost of wireless technology has been reduced, and wireless communication devices can support a plurality of wireless communication as mesh nodes, and can provide functions such as client access and interactive service.

A wireless mesh network, such as a mobile ad-hoc network (hereinafter abbreviated as MANET), is an autonomously distributed network that does not depend on a base station or a fixed network but has mobile terminals (hereinafter referred to as nodes) as components. Yes, the node performs data transfer in a bucket-relay manner (multi-hop). Hierarchical routing Hi-AODV (Ad hoc On-demand Distance Vector ad hoc on) using autonomous distributed clustering and autonomous distributed clustering as a protocol to efficiently manage node movement and improve data packet arrival performance in MANET environment A demand distance vector, Non-Patent Document 2, Non-Patent Document 3) has been proposed. Hi-AODV is a hierarchical routing in which AODV is applied to a set of clusters configured by autonomous distributed clustering. Autonomous distributed clustering is a technique in which a large-scale network is divided into a plurality of small sub-networks called clusters, and a management node called a cluster head manages each cluster. In this way, by dividing the network into a plurality of clusters and making them hierarchical, efficient routing such as reduction of control packets can be performed. Such a layering method can also be applied to a wireless mesh network.

Further, in an ad hoc network, a communication method that can configure a cluster in an appropriate shape is also known (Patent Document 1).

JP 2006-186446 A

Shiro Sakata, Hidenori Aoki, Kenichi Mase, “Ad-hoc network and wireless LAN mesh network” IEICE Japanese Journal B, J89-B (6), pp.811-823, 2006. T. Ohta, S. Inoue, Y. Kakuda, and K. Ishida, “An adaptive multihop clustering scheme for ad hoc networks with high mobility,” IEICE Trans. Fundamentals, vol. E86-A, no. 7, pp. 1689 -1697 2003. Ryotaro Oda, “Research on efficient routing using autonomous distributed clustering in ad hoc networks” 2009 Hiroshima City University Doctoral Dissertation, pp.19-29, 2010. L. Atzori, A.lera, G. Morabito, “The Internet of Things: A survey” Computer networks, Volume 54, Issue 15, 28 pp.2787-2805, Oct. 2010. Study meeting on how to secure communications in emergencies such as large-scale disasters, Ministry of Internal Affairs and Communications, “About securing communications in emergencies such as large-scale disasters” p.67, Nov. 2011.

The number of users who use services requiring high bandwidth on smartphones and tablet terminals is increasing, and traffic flowing through the nodes is increasing. Furthermore, communication between things (sensors, wireless tags, etc.) represented by the Internet of Things (Non-patent Document 4) and things, and concentration of communication for failure and safety confirmation of base stations during disasters ( Non-patent document 5) may cause a sudden increase in traffic, and it is difficult to predict the traffic flowing in the node. When the hierarchization method is used in such an environment, if the operation is always performed in a cluster of the same size, the load on the cluster head increases when traffic increases. As a result, packet loss occurs at the cluster head, leading to a reduction in throughput.

The communication method described in Patent Document 1 is intended to reduce the number of gateways and to cope with a sudden increase in traffic, taking into account the geographical shape, and having a rounded cluster shape. Is different.

The present invention proposes a method in which a network is divided into a plurality of clusters and managed using the concept of autonomous distributed clustering, and a cluster is dynamically changed by traffic flowing in the cluster. In the method of the present invention, traffic information is acquired and managed in real time using control packets transmitted periodically. When the cluster head determines that the traffic is high, the clusters are dynamically divided and combined. As a result, the load on the cluster head is distributed to prevent a decrease in throughput.

In the dynamic tiering method in the network according to the present invention, the network is divided into a plurality of clusters, and one cluster has one cluster head for managing the cluster, and the cluster head is the cluster head. Is periodically broadcast to the cluster member, and the received cluster member forwards it to the downstream node, and the leaf node of the cluster that has received the member packet is directed to the upstream node toward the cluster head. Each node that has received the member packet MEP builds a tree with the source node of the member packet as a parent, and each member in the cluster is controlled by a control packet consisting of a member acknowledgment packet to be transmitted, Always record the amount of passing data packets per hour, and Is recorded in the member acknowledgment packet and is sent to the cluster head periodically as data for each transmission of the member acknowledgment packet together with its own node information. The gateway which is an adjacent node transmits node information and traffic information of the adjacent cluster in addition to its own information on the member confirmation response packet, and the cluster head transmits the member confirmation response packet from the cluster member. based generates a cluster member list, and manages the cluster, collect the passing packet amount of each member, the sum as a total traffic in the cluster, a total traffic of the cluster head in regularly the cluster And if the preset threshold is exceeded Divides the cluster, if the lower limit of the threshold value set, after confirming the traffic information of neighboring clusters, with a join with an adjacent cluster if they meet a predetermined condition, when the division of the cluster Is a node that is ½ of the average number of hops to each member in the cluster and there are a plurality of nodes that satisfy this condition, the adjacent cluster list among the nodes with the determined number of hops A node that knows a lot of information is a new cluster head candidate .

In addition, the cluster head transmits a packet RNCP designating an arbitrary node in the cluster as a new cluster head candidate after division, and the arbitrary node that has received the RNCP becomes a cluster head candidate. A packet CNCP with its own node ID added is sent to the surrounding nodes, and other nodes in the cluster set the node ID included in the cluster ID change packet received earlier as their own cluster ID. The division is completed.

If the total traffic in the cluster is below the lower limit of the threshold, check the adjacent cluster list, check the total intra-cluster traffic of the adjacent cluster, and the sum of the traffic in its own cluster and the total traffic in the adjacent cluster is the threshold. If the upper limit of the cluster is not exceeded, the cluster that has undergone the split will not join for a certain period of time in order to join the cluster with the smallest intra-cluster traffic and prevent rejoining before the path is built after splitting the cluster It is characterized by this.

As a result of comparison with fixed hierarchical routing in an environment with high traffic, we confirmed that the data arrival rate was improved and the amount of dropped packets was reduced. In addition, a decrease in the amount of packets passing through the cluster was confirmed as an evaluation value of load distribution.

It is a figure for demonstrating the cluster construction by the control packet concerning embodiment of this invention, (a) is a figure which shows control packet transmission by a cluster head, (b) is a figure which shows control packet transmission from a cluster member. is there. FIG. 6 is a diagram for explaining a dividing operation using RNCP and CNCP in the same embodiment. It is a figure which shows an experiment field. It is a curve figure which shows the number of SD pairs for every time. It is a wave form diagram which shows the amount of drop packets of the cluster head for every time. It is a graph which shows the amount of dropped packets at the time of buffering. It is a graph which shows the data arrival rate of the packet for evaluation. It is a graph which shows the passage packet amount at the time of cluster head passage of the packet for evaluation.

In the following, autonomous distributed clustering and hierarchical routing Hi-AODV will be described as examples.

Autonomous distributed clustering Autonomous distributed clustering is a method of managing a network by dividing it into a plurality of sub-networks in a MANET environment. Each subnetwork is called a cluster. In order to limit the number of nodes in one cluster, a lower limit value L and an upper limit value U are set in advance. These set values are used as threshold values for the number of members in the cluster when the clusters are joined or divided. Each cluster consists of one cluster head that manages the cluster and multiple cluster members, and the node ID of the cluster head is the cluster ID of that cluster. At this time, all the nodes of the cluster are maintained so as to be connected by nodes having the same cluster ID. Autonomous distributed clustering has the feature that clusters do not overlap.

Figure 1 shows an example of control packet transmission by the cluster head and cluster members. Cluster head C broadcasts control packets in the cluster as shown in Fig. 1 (a) in order to efficiently collect information on cluster members (nodes indicated by circles). Broadcast refers to sending information to all nodes in the cluster. The control packet forms a tree rooted in itself. Then, the cluster head C collects member information by control packets from the members as shown in FIG. 1 (b). A node adjacent to a node having a cluster ID different from its own cluster ID is called a gateway. Further, when gateways of different clusters exist within the communication range, it is said that the clusters are adjacent. If the number of cluster members is less than the lower limit L, join with one of the adjacent clusters. If the number of cluster members exceeds the upper limit U, the own cluster is divided into two.

When transmitting a data packet, if the source node and the destination node belong to the same cluster, the data packet is transferred using a tree rooted at the cluster head C. In autonomous distributed clustering, cluster head C centrally manages the entire cluster, reducing the load on other cluster members.

Hierarchical routing Hi-AODV
Hierarchical routing Hi-AODV performs intra-cluster routing and inter-cluster routing in networks divided by clusters. Intra-cluster routing is performed using a cluster head base tree. In inter-cluster routing, the cluster is regarded as one virtual node and routing is performed based on the AODV concept. In intra-cluster routing in Hi-AODV, when a node transmits a data packet, the data packet is transmitted to the cluster head based on the cluster head base tree. When the cluster head receives a data packet, the destination node is searched from its own cluster member list, and when the destination node is in the cluster, the data packet is transmitted to that node. If the destination node does not exist in the cluster member list, the data packet is buffered and the inter-cluster routing process is executed. In inter-cluster routing in Hi-AODV, a cluster is regarded as one virtual node, and a route is constructed by RREQ (Route REQuest) and RREP (Route REPly). The route table used for inter-cluster routing is managed only by the cluster head of each cluster, and the route search / maintenance is performed on a cluster basis. Further, only the cluster head processes the control packet, and the other nodes only perform forwarding, thereby suppressing overhead due to routing. However, when the traffic in the cluster increases, there is a problem that the load on the cluster head increases.

A method of dynamically changing the size of the cluster according to the amount of traffic is used so that the load on the cluster head does not increase even if the intra-cluster traffic increases. In this method, the network is divided into multiple clusters and managed like autonomous distributed clustering in MANET, and data communication is performed by hierarchical routing. In the autonomous distributed clustering in the MANET environment, the cluster is divided or joined using the upper limit U and the lower limit L of the number of nodes. However, in the method of the present invention, the change is performed according to the traffic volume in order to cope with a sudden increase or decrease in traffic. When a route construction request is generated from the transmission source node, the route construction between clusters is performed by the hierarchical routing Hi-AODV. Hereinafter, cluster construction, traffic information collection, cluster division, combination, and data packet transfer will be described in order.

Cluster construction and traffic information collection Cluster management and traffic information acquisition according to the method of the present invention are performed by using control packets called cluster member packets (MEP) and cluster member acknowledge packets (MAP). To do. The cluster head periodically broadcasts the MEP to each member in the cluster. Each node that receives the MEP constructs a tree with the MEP source node as the parent. Each member in the cluster always keeps track of the amount of transit data packets per hour. The amount of passing packets recorded in the cluster per time is recorded in the MAP and periodically sent to the cluster head as data for each MAP transmission together with its own node information. At this time, the gateway, which is a node adjacent to another cluster, transmits the node information and traffic information of the adjacent cluster on the MAP in addition to its own information. The cluster head generates a cluster member list based on the MAP from the cluster member and manages the cluster. Then, the amount of passing packets of each member is collected, and the sum is defined as the total traffic in the cluster.

The split and join cluster head periodically checks the total traffic in the cluster, and splits the cluster if it exceeds the preset upper limit U. If it falls below the lower limit L of the set threshold, after confirming the traffic information of the adjacent cluster, if the condition is satisfied, it joins with the adjacent cluster.

When the traffic information in the cluster collected by the divided cluster head exceeds the upper limit U of the threshold, the cluster is divided. Figure 2 shows the division procedure. At the time of splitting, cluster head A designates node B in the cluster as a new cluster head candidate after splitting, and transmits a packet called RNCP (Recommendation for NEw Cluster head Packet) as shown in FIG. 2 (a). Node B that receives the RNCP becomes a cluster head candidate. Then, the cluster head A and the node B that has newly become a cluster head candidate transmit a CNCP (Candidate New Cluster Packet) with its own node ID added to the surrounding nodes as shown in FIG. The other nodes in the cluster set the node ID included in the cluster ID change packet received earlier as their own cluster ID, and the cluster division is completed.

The method of determining cluster head candidates at the time of division is as follows. It is desirable to divide the cluster so that the traffic in the cluster is halved as much as possible. For this purpose, a method is used in which the average number of hops to each member in the cluster is 1/2. If there are a plurality of nodes that satisfy the condition, a node that knows a large number of adjacent cluster lists among the determined number of hops is determined as a new cluster head candidate.

If the total traffic in the combined cluster is less than or equal to the lower threshold L, check the adjacent cluster list and check the total intra-cluster traffic of the adjacent cluster. If the sum of the traffic in the cluster and the total traffic in the adjacent cluster does not exceed the upper limit U of the threshold, the cluster is joined to the cluster having the smallest intra-cluster traffic. Further, in order to prevent reconnection before the path is constructed after dividing the cluster, the cluster in which the division has occurred is not connected for a certain period of time.

Data packet transfer Data packets are transferred using Hi-AODV based on clustering. Inter-cluster communication, like Hi-AODV inter-cluster communication, treats a cluster as one virtual node and performs routing based on the AODV concept using RREQ. If the cluster ID on the route changes due to the splitting or joining of clusters, route disconnection occurs. In that case, route search is performed by RREQ.

Simulation experiment The performance of the method of the present invention is evaluated by a simulation experiment. In the hierarchical routing (hereinafter referred to as the conventional method) in which the method of the present invention is not divided and combined, the background traffic is changed by increasing the number of SD pairs and a simulation is performed for comparison. Assuming that traffic in the network increases rapidly in the experiment, prepare multiple SD pairs with a short transmission interval so that the traffic at one location increases, and use this as the background SD pair. Prepare an evaluation SD pair in addition to the background SD pair. In order to evaluate the load of the cluster head through which the arrival data packet passes, the passing packet amount per unit time is recorded in the packet every time it passes through the cluster head. The total of the recorded passing packet amounts divided by the number of packets reaching the destination node is used as the passing data packet amount of the cluster head. The evaluation items are the number of dropped packets that occur when the buffer capacity is maximum when buffering data packets, the data arrival rate of the evaluation SD pair, and the amount of packets passing through the cluster head with respect to the arrival data packet of the evaluation SD pair.

Table 1 shows the environmental parameters in the experimental environment simulation experiment. In order to evaluate packet loss in the network with a communication bandwidth of 11 [Mbps], simulation was performed in an environment where loss due to packet collision did not occur. The amount of passing data packets, which is the threshold value of the method of the present invention, was set to an upper limit U 250 [Kbyte] and a lower limit L 50 [Kbyte], and the cluster size threshold value of the conventional method was set to an upper limit U 20 and a lower limit L 10. After the cluster configuration is stabilized, it is considered that the value exceeds the lower limit. Assume that the background traffic increases in the mesh network. Traffic is increased by increasing the number of CBR transmission / reception pairs over time.

Figure 3 shows the fields used in the experiment. The upper circle in FIG. 3 indicates a node, and the filled nodes indicate a source node and a destination node. Nodes are arranged in a grid pattern on this field. Figure 4 shows the number of SD pairs during data transfer. The number of SD pairs was two ways: Case 1 and Case 2. Communication with 5 evaluation SD pairs and always transfer data packets. As time goes on, increase or decrease the number of SD pairs during background data transfer. The number of SD pairs shown in FIG. 4 indicates the total of the number of background SD pairs and the number of evaluation packet SD pairs. The CBR transmission interval was set to 0.1 [s] for the background SD pair and 0.25 [S] for the evaluation SD pair.

Experimental results and discussion Drop packet volume during buffering Case 1 shows the amount of cluster head dropped packets per hour in Fig. 5, and Fig. 6 shows the total dropped packets of all nodes in cases 1 and 2. It can be seen that the amount of dropped packets can be reduced in case 2 compared to case 1 in the SD pair. Also, when looking at the amount of dropped packets per time, the graph according to the method of the present invention shows that the amount of dropped packets increases around 115 seconds, but thereafter the amount of dropped packets is reduced compared to the conventional method. This is considered to be because the amount of dropped packets increased because the packets were temporarily overflowed from the buffer until the path construction was completed because the cluster was divided when the traffic increased. However, it is considered that the amount of dropped packets can be reduced thereafter by the effect of the method of the present invention. As shown in FIG. 6, the total amount of dropped packets is also reduced. It can be said that the method of the present invention is more effective when the number of SD pairs in Case 2 is large.

Data arrival rate The data arrival rate may be decreased due to radio wave interference or buffer overflow that occurs when a data packet is buffered when the buffer size is the maximum value. In the present invention, since an experiment was performed in an environment in which no packet loss due to radio wave interference occurred, it is considered that the buffer overflow most affected the data arrival rate.

Figure 7 shows the data arrival rate of the evaluation packet. In the method of the present invention, the data packet arrival rate is improved by about 8% in case 1 and by about 23% in case 2. This is probably because the reduction in the amount of dropped packets has led to an improvement in the data arrival rate as shown in FIG.

As an evaluation of the load of the cluster head when passing through the cluster head, it can be evaluated that the smaller amount of packets passing through the cluster head is distributed and transferred to a plurality of clusters. Figure 8 shows the amount of packets passing through the cluster head. The graph of the method of the present invention has a higher data arrival rate but a smaller amount of packets passing through the cluster head. Therefore, it can be said that the load on the cluster head can be distributed by dynamically adjusting the cluster size. There is not much difference between Case 1 and Case 2. This is probably because the MEP and MAP may not reach when there is a high load, so the tree cannot be constructed and the amount of passing packets does not increase.

As described above, the present invention employs a dynamic tiering method that dynamically changes the size of the cluster to reduce the load on the cluster head when the background traffic increases, particularly in a wireless mesh network. As a result of the simulation experiment, it was confirmed that the data arrival rate was improved and the amount of dropped packets was reduced, and the decrease in the amount of packets passing through the cluster when passing through the cluster was confirmed as an evaluation value of load distribution.

The present invention works effectively in a network such as a backbone in which a large amount of traffic flows while fluctuating spatially and temporally. The target network is wide, such as wired, wireless and mobile. It can also be applied to ad hoc networks. For example, it can be used as a means for ensuring communication in the case where traffic rapidly increases during a disaster such as a fire, earthquake, typhoon, or tsunami. It can also be used when traffic rapidly increases due to events such as network events. There is no limitation on the type of traffic that flows through the network. This is particularly effective in a flowing network.

C, A, B cluster head

Claims (3)

The network is divided into a plurality of clusters, and one cluster has one cluster head that manages the cluster.
A member packet periodically broadcast from the cluster head to the cluster member, and the received cluster member forwards it to the downstream node;
Controlled by a control packet consisting of a member acknowledgment packet transmitted from the leaf node that received the member packet of the cluster to the upstream node toward the cluster head,
Each node that receives the member packet constructs a tree with the source node of the member packet as a parent,
Each member in the cluster always records the amount of transit data packets per hour,
It said amount passing packets per recorded time in the cluster is described in the member acknowledgment packets, sent with its own node information periodically to the cluster head as a data for each transmission the member acknowledgment packet,
A gateway that is a node adjacent to another cluster transmits node information and traffic information of the adjacent cluster in addition to its own information in the member acknowledgment packet,
The cluster head on the basis of the member acknowledgment packet from the cluster member, generates a cluster member list, and manages the cluster, collect the passing packet amount of each member, total traffic in the sum the cluster age,
The cluster head periodically checks the total traffic in the cluster, and if the upper limit of a preset threshold is exceeded, the cluster is divided,
If it falls below the lower limit of the set threshold, after confirming the traffic information of the adjacent cluster, if it meets the predetermined condition, it will join with the adjacent cluster ,
When the cluster is divided, if there are a plurality of nodes satisfying this condition, which are half of the average number of hops to each member in the cluster, the node of the calculated number of hops Dynamic layering method in a network characterized in that a node that knows many neighboring cluster lists among them is a new cluster head candidate The cluster head transmits a packet RNCP designating a new cluster head candidate after dividing any node in the cluster,
The arbitrary node that has received the RNCP becomes a cluster head candidate, the node transmits a packet CNCP with its own node ID added to surrounding nodes,
2. The network operation according to claim 1, wherein another node in the cluster sets the node ID included in the cluster ID change packet received earlier as its own cluster ID, and completes the division of the cluster. Stratification method If the total traffic in the cluster is below the lower limit of the threshold, check the adjacent cluster list, check the total intra-cluster traffic of the adjacent cluster,
If the sum of the traffic in your cluster and the total traffic in adjacent clusters does not exceed the threshold limit, it joins the cluster with the lowest intra-cluster traffic,
2. The dynamic layering method in a network according to claim 1, wherein, in order to prevent reconnection before the path is constructed after dividing the cluster, the cluster in which the division has occurred does not join for a certain period of time.

Images