JP4548792B2 - Communication route control method, communication route control system and program for overlay network - Google Patents

Description

  The present invention relates to a technology for improving end-to-end communication quality in the Internet, and more particularly to end-to-end communication quality in a communication path using an overlay network logically formed on an IP network. The present invention relates to a technique suitable for efficient improvement.

  The Internet, which represents an IP network, has been developed and spread to accommodate various applications. Nowadays, the real time is sensitive to VoIP (Voice over IP) and QoS (Quality of Service) represented by streaming. The containment of applications etc. is also developing rapidly.

  Along with this, it has become an important issue to realize technology ("end-to-end QoS management technology") on the Internet to avoid end-to-end congestion and improve quality. . However, there are the following problems in realizing such a technique.

  (1) The Internet has already become a social infrastructure, and it is difficult to expand new functions at the network layer that greatly change the existing network structure.

  (2) The Internet is formed by a plurality of ASs (Autonomous Systems) having different management entities, and it is difficult to extend new functions to all ASs simultaneously.

  Under such circumstances, as an effective technique that can improve end-to-end QoS without changing a lower network layer, a QoS management technique using an overlay network described in Non-Patent Document 1, for example, has attracted attention.

  As described in Non-Patent Document 2, for example, an overlay network is a network that uses an existing link and forms a logical (virtual) link in an upper layer according to the purpose.

  The basic concept of QoS management by such an overlay network is illustrated in FIG. In FIG. 1, 1 to 3 are nodes (overlay nodes a, b, and c) constituting the overlay network 11, and 4 to 8 are IP routers constituting the IP network 12, and are broken lines from x to y. It is assumed that traffic is flowing along the route indicated by the arrow. Further, it is assumed that there is a congested IP router on this route, and as a result, the QoS between x and y is lowered.

  At this time, using the overlay network 11 formed by the overlay nodes a1, b2, and c3, the traffic is detoured along a route (x → overlay node a1 → overlay node b2 → overlay node c3 → y) represented by a solid line arrow. If possible, the above congestion can be avoided.

  In fact, Non-Patent Documents 3, 4, and 5 show that there are a number of such detour routes in the real network based on actual measurements. However, the results of Non-Patent Document 3 and Non-Patent Document 5 show that the overlay network topology is a full mesh and evaluation is performed when ideal communication path calculation is performed using quality information measured between all overlay nodes. Thus, no consideration is given to a decrease in scalability (system extensibility) when the total number of overlay nodes increases.

  That is, when the total number of overlay nodes is large, there is a problem that it is practically impossible to calculate communication paths using quality measurement information between all overlay nodes.

  In order to avoid such a problem, Non-Patent Document 4 evaluates a case where the number of relay node candidates that provide a detour route is limited. However, the relay node candidates are simply selected at random, and the relay node candidate selection method has not been studied.

L. Zhi and P. Mohapatra, “QRON: QoS-aware routing in overlay net-works,” IEEE J. Select. Areas Commun., Vol. 22, pp. 29-40, January 2004. WIDE Project, “Integrated Distributed Environment with Overlay Network,” WIDE Project Research Report, Part 17, 2002. Kamei, Kawahara, “Traffic engineering by end-host overlay network and its effectiveness,” Shingaku Sodai, BS-5-3, 2004. S. Rewaskar and J. Kaur, “Testing the Scalability of Overlay Routing Infrastructures,” Proc. PAM 2004. April 2004. S. Banerjee, T. G. Grifin and M. Pias, “The Interdomain Connectivity of PlanetLab Nodes,” Proc. PAM 2004, April 2004.

  On the other hand, the present inventors (in “Japanese Patent Application No. 2005-197060”) appropriately limit the relay node candidates based on the measurement data, so that the route search is performed with all nodes as relay candidates. It shows that almost the same QoS improvement can be achieved.

  In this technology, for all node pairs, the optimal route is calculated when all nodes are relay node candidates, and the “frequency” at which each node is selected as the relay node that provides the optimal route is calculated. In addition, the top M nodes having the highest “frequency” are extracted, and only M relay candidate nodes are used in the subsequent route calculation with the above preparation.

  In this technique, it was necessary to search all routes once, but the present inventors further described (“Kawara, Kamei, Uchida, Abe,“ QoS overlay route when the number of alternative route candidates is limited. Selection algorithm and its evaluation, “Science Technical Report IN2005-195, pp. 231-236 (March 2006)”), always keep the number of relay node candidates below a certain number, A technique has been proposed in which a relay node capable of providing a high-quality route through a selection action is learned, and such a node is selected with a high probability to improve the QoS almost equivalent to the optimum case.

  However, in this technology, it is assumed that the frequency with which each relay node can provide an optimal route is constant and does not change, and the network conditions change, and the order of the frequency of being the optimal relay node changes. In some cases, there was a problem that it could not be handled.

  For example, until now, a certain node has a poor quality because the network where the node is installed has insufficient line bandwidth, and the frequency of being an optimal node was very small. Assume the case where the quality has been improved by expansion.

  In such a case, in the above-described two technologies, the top M nodes having high quality are determined based on the network conditions so far, and a relay node is selected from the top M nodes. The problem that it cannot be selected arises.

  On the other hand, when the network conditions around a node that has been ranked as the top M relay node candidates so far change and the quality deteriorates, the node is selected as a relay candidate. Arise.

  The problem with the prior art is that it is not possible to properly select a relay node that can appropriately follow a change in network conditions in the overlay network and provide a high-quality route according to the network conditions at that time. .

  An object of the present invention is to solve these problems of the prior art and efficiently improve the end-to-end communication quality in the overlay network.

  In order to achieve the above object, the present invention provides a relay node capable of appropriately following a change in network conditions and providing a high-quality route according to the network conditions at the time of performing communication path control in an overlay network. It is characterized by having a mechanism that can correct the selection probability of each node so that can be appropriately selected. That is, in the overlay network, which is a logical network composed of N overlay nodes connected to the IP network, each node i (i = 1) belonging to the overlay network is determined when determining the communication path from the source node to the destination node. ~ N) measure the communication quality between the source node i and the destination node j, with the local node as the source node and the other node j (all except i) as the destination node j. To select M nodes from among the N nodes as relay node candidates, obtain a communication quality measurement result between the relay candidate node k and the destination node j from the relay candidate node, and receive from the own node i. The communication quality between the destination node j and the communication quality when reaching the destination node j via the relay candidate node k from its own node i are compared. In the case where better quality is given, the route for direct transfer to the destination node j without using the relay node is determined as a communication route from the own node i to the destination node j, and the latter gives good quality. In this case, the node k * that provides the highest quality is used as the relay node, and the communication path is determined as the local node i → the relay node k * → the destination node j. M ′ relay node candidates are selected according to the frequency selected as the relay node, and the remaining MM ′ candidates select the relay node candidates with equal probability without depending on the frequency. Changes in network conditions are monitored, and when the changes are detected, the selection probability of each node is changed according to the situation.

  According to the present invention, it is possible to improve the communication quality while following the fluctuation of the network condition and reducing the cost associated with the communication path calculation in the overlay network.

  The best mode for carrying out the present invention will be described below with reference to the drawings. FIG. 2 is a block diagram showing a configuration example of an overlay network provided with a communication path control system according to the present invention. The overlay network includes overlay nodes 21 to 25 (hereinafter simply referred to as “nodes 21 to 21”). 25 ”and a management server 20 are provided.

  Each node such as the nodes 21 to 25 and the management server 20 have a computer configuration including a CPU (Central Processing Unit), a main memory, a display device, an input device, and an external storage device, and a CD-ROM via an optical disk drive device or the like. After the programs and data recorded in the storage medium such as the above are installed in the external storage device, they are read from the external storage device into the main memory and processed by the CPU, thereby executing the functions of the respective processing units according to the present invention.

  Each node such as the nodes 21 to 25 is logically connected to other nodes. That is, it knows the IP addresses of other nodes in the overlay network and is in a communicable state.

  The management server 20 has a function of managing any one of the nodes including the nodes 21 to 25 as a relay node candidate.

  In such a configuration, in this example, when performing communication path control in the overlay network, a relay node that can appropriately follow a change in network conditions and provide a high-quality path according to the network conditions at that time is provided. A mechanism (first to fifth techniques) that can correct the selection probability of each node including the nodes 21 to 25 so as to be appropriately selected is provided.

  The first technique is to measure the communication quality between the nodes constituting the overlay network and to determine whether or not to make a detour using a relay node. When determining a communication path from a certain node to a destination node, each node i (i = 1 to N) has its own node as a source node and another node j (all except i) as a destination node j. The communication quality between the source node i and the destination node j is measured, while M nodes are selected from the N nodes as relay node candidates and received from the relay candidate node k. The communication quality measurement result between the destination node j is obtained from the relay candidate node, the communication quality between the own node i and the destination node j, and the destination destination node k from the own node i via the relay candidate node k. Compare the communication quality when arriving at the node j, and if the former gives better quality, the route from the local node i to the destination node j is directly transferred to the destination node j without using the relay node. If the latter is determined as a communication path and the latter gives good quality, the node k * providing the highest quality among them is set as a relay node, and the communication path is determined from the own node i → the relay node k * → the destination node j. At the next selection of relay node candidates, M ′ relay node candidates are selected according to the frequency of selection as a relay node in the past, and the remaining MM ′ numbers do not depend on the frequency. In addition, relay node candidates are selected with equal probability, and further, changes in network conditions are monitored. When such changes are detected, the selection probability of each node is changed according to the situation.

  The technical part for selecting relay node candidates according to the frequency of selection as a relay node in the past is the above-mentioned “Kawara, Kamei, Uchida, Abe,“ QoS overlay route selection when the number of alternative route candidates is limited. Algorithm and its evaluation, “Technology Report, IN2005-195, pp.231-236, (March 2006)”.

  In this example, in addition to that, a function of selecting with equal probability without depending on the frequency is added. By doing this, a node is selected without depending on the past situation. For example, a node that was low quality in the past but is now of high quality can be selected with a certain probability, and once selected Then, the selection probability gradually increases, and selection according to the network status at that time becomes possible.

  Next, the second technique will be described. This second technique selects a relay candidate node. In this example, the management server 20 exists, and a relay candidate server selection technique using this management server 20 will be described.

  The management server 20 manages information on relay node candidates in the overlay network. The management server 20 manages the score C (n, k) in the nth measurement period (time point n) of each node including the nodes 21 to 25 shown in the figure.

At the time point n, the source node i reads the score C (n, k) of each node k (k = 1 to N) from the management server 20 when determining the route to the destination node j, and the probability p_k = A node k (k is other than i) is selected as a relay candidate node by C (n, k) / Σ _k C (n, k), and this is repeated until MM ′ relay candidate nodes are determined.

  In addition, relay candidate nodes are selected from the remaining node groups with equal probability, and this is repeated until M 'items are selected. Then, the source node i determines the route to the destination node j as described above.

  At that time, if the node k * is determined as the relay node, the node k * notifies the management server 20 of the fact. A similar procedure is performed for all source and destination node pairs.

  Then, the management server 20 receives from each node that it has been determined as a relay node, and counts the number of times C ′ (n, k) at which each node k has become a relay node at time n. Using the value, the score of each node k at the next time point n + 1 is C (n + 1, k) = max {C_low, (1-1 / t ′) C (n, k) + 1 / t′C ′ (n , K)}.

  Here, C_low is a lower limit value of a predetermined score, 1 / t ′ is a smoothing parameter, and t ′ = min {n + β, c using predetermined parameters β (> 0) and c (> 1). }, And the initial value of C (n, k) is C (0, k) = 1 (all k).

  Here, the score C (n, k) of the node is updated so as to increase when the node k provides a high-quality route, and to decrease otherwise. Therefore, by selecting a relay node candidate in proportion to the score, a relay node that provides a high-quality route is easily selected at the next relay node candidate selection.

  Also, by setting the lower limit value C_low in the score so that the score does not become excessively small, when the node whose current score is lower becomes the higher node in the future when the network condition changes, the score becomes lower. I am trying to recover.

  Next, the third technique will be described. This third technique, like the first technique described above, measures the communication quality between the nodes and determines whether to make a detour using a relay node. In this example, the third technique As the communication quality between i and the destination node j, the above-mentioned “Kawahara, Kamei, Uchida, Abe,“ QoS overlay route selection algorithm when the number of alternative route candidates is limited and its evaluation, ”IEICE IN2005 -195, pp.231-236, (March 2006) ", and M relays are used when determining the route from the local node i to the destination node j using the delay time d (i, j). A technique for determining whether to make a detour using any of the candidate nodes will be described.

  In this case, using the delay time d (i, k) from the node i to the relay candidate node k and the delay time d (k, j) between the relay candidate node k and the destination node j, d (i, j )> D (i, k) + d (k, j) is examined to determine whether or not k exists. If not, the direct communication from node i to node j is determined as the communication path. K that minimizes the right side is set as the relay node k *, and the communication path is determined as self node i → relay node k * → destination node j.

  Next, the fourth technique and the fifth technique will be described. These fourth and fifth techniques change the selection probability of a relay node candidate in accordance with the fluctuation condition of the network condition in the first technique described above. In this example, the node used for selecting a relay node A technique for changing the score will be described as an example.

In the fourth technique, the management server 20 counts the number of times C ′ (n, k) at which each node k becomes a relay node at time n, and calculates the total value Σ _k C ′ (n, k). Calculated and divided by the total number of node pairs N (N−1) is the frequency F (n) with better quality of the detour path, that is, F (n) = Σ _k C ′ (n, k) / {n (n-1) } and, if the current time n is "n>T" F_T meaning a moving average to the T period before the (n) _{"F_T (n) = Σ x =} 1 _˜TF (n−x) / T ”and when“ F (n) / F_T (n) ”is smaller than a predetermined threshold value, the score C (n, k) is reinitialized to C (0, k) (k = 1 to N), and time n = 0.

  This is because when the network condition is in a steady state without changing, it is considered that the frequency F (n) for which the detour is better is not greatly changed and takes a substantially constant value. In the current relay node selection, an appropriate detour cannot be found, and as a result, the value of F (n) is considered to be small.

  Therefore, by monitoring the behavior of F (n) and determining that the value of F (n) has changed, it is possible to follow a new situation by reinitializing the score of each node and learning again. Become.

In the fifth technique, each node i measures the delay d (i, j) from its own node i to the destination node j, and the average delay time d_avg from the current node of the current n to all other nodes. (N) = Σ _j d (i, j) / (N−1) is calculated.

  On the other hand, the exponential weighted moving average d_avg_s (n−1) at the time point n−1 is calculated by “d_avg_s (n−1) = (1−α) d_avg_s (n−2) + αd_avg (n−1)”. Here, α is a predetermined smoothing parameter, and 0 <α <1.

  If “d_avg (n) <d_avg_s (n−1) −th_d” is satisfied, the management server 20 is notified accordingly. Here, th_d is a threshold value, which is predetermined or lower y percent from the history of delay measurement results “d_avg_s (n−2), d_avg_s (n−3), d_avg (n−4),. A value is obtained and the difference between the value and d_avg_s (n−1) is set to th_d.

  The management server 20 that has received the change in delay from the node i has the node i in the upper X node with respect to the score C (n, k) of the node k (k = 1 to N) managed by the own server. If it is not entered, the score C (n, i) of the node i is increased by C (n, i) ← C (n, i) + C_add. Here, C_add is a predetermined score increase amount.

  By doing this, when the quality (delay time) of a node currently ranked lower is changed and improved, the node is appropriately ranked higher by increasing the score of that node. It becomes possible.

  Hereinafter, in this way, the communication quality between nodes is measured by either the first technique or the third technique, and the function for determining whether to make a detour by using the communication quality, the relay is performed by the second technique. FIG. 3 and FIG. 4 are used for each means for executing the function of selecting a candidate node and the function of changing the score of a node used when selecting a relay node by either the fourth technique or the fifth technique. I will explain.

  FIG. 3 is a block diagram illustrating a configuration example of the overlay node in FIG. 2, and FIG. 4 is a block diagram illustrating a configuration example of the management server in FIG.

  Each node 21 to 25 in FIG. 2 updates a communication path with another node at regular intervals as follows. Here, the description will be given focusing on the node 21 in FIG. However, the other nodes 22 to 25 behave similarly in the same manner.

  A node (denoted as “overlay node” in FIG. 3) 21 in FIG. 3 selects a relay candidate node by the above-described second technique, measures the communication quality between the nodes by the third technique, and uses it. The communication quality measuring unit 21a, the communication quality acquiring unit 21b, the relay node determining unit 21c, the relay candidate selecting unit 21d, and the detour route setting unit 21e are provided as functions for determining whether or not to detour. Therefore, the management server 20 in FIG. 4 includes a relay node selection count management unit 20a and a node score update / management unit 20b.

  In FIG. 3, the communication quality measuring unit 21a measures the delay time d (1, j) between itself (node 21) and another node j (j = 22 to 25) every measurement period τ. Similarly, the node i (i = 22 to 25) also measures the delay time d (i, j) with another node j (any of j = 21 to 25 excluding j = i). And if this delay time measurement is implemented, the relay candidate selection part 21d will be notified to that effect.

The relay candidate selection unit 21d obtains the score C (n, k) at the n-th measurement period (hereinafter referred to as time point n) of each node k from the node score update / management unit 20b of the management server 20 in FIG. Download it and use it to select node k (k is other than i) as a relay candidate node with probability p_k = C (n, k) / Σ _k C (n, k), and select this as M−M ′ Repeat until a relay candidate node is selected.

  In addition, relay candidate nodes are selected from the remaining node groups with equal probability, and this is repeated until M 'items are selected. The total M relay candidate nodes thus selected are notified to the relay node determination unit 21c.

  Upon receiving the notification, the relay node determination unit 21c instructs the communication quality acquisition unit 21b to acquire the delay time measurement result d (k, j) of the nodes k and j from the relay candidate node k. .

  Upon receiving the instruction, the communication quality acquisition unit 21b acquires the delay time measurement result d (k, j) of the nodes k and j from the communication quality measurement unit 21a of the relay candidate node k, and sends it to the relay node determination unit 21c. To be notified. The above procedure is performed for all relay candidate nodes.

  The relay node determination unit 21c acquires the delay time d (1, k) of the relay candidate node k from its own node 1 from the communication quality measurement unit 21a, and the relay candidate node k notified from the communication quality acquisition unit 21b. Using delay time d (k, j) between the destination nodes j, k satisfying “d (21, j)> d (21, k) + d (k, j)” is M candidate candidates. If it does not exist, the direct path from the node 21 to the node j is determined as the communication path. If it exists, k that minimizes the right side is set as the relay node k *. decide.

  After determining in this way, the detour route setting unit 21e of the node k * is requested to set the communication route as “own node 21 → relay node k * → destination node j”. The above procedure is performed for all destination nodes j (j = 22 to 25).

  The detour route setting unit 21e of the relay node k * sets a route of “node 21 → relay node k * → destination node j”, and then the relay node of the management server 20 indicates that it has been selected as the relay node. Notify the selection count management unit 20a.

  When receiving the notification from the relay node k *, the relay node selection count management unit 20a of the management server 20 in FIG. 4 calculates the number of times C ′ (n, k) at which the node k * has become a relay node as “C ′ ( n, k) ← C ′ (n, k) +1 ”.

  The above is repeated until the n-th measurement cycle is completed, and the number of times of becoming a relay node of each node is counted. The result is notified to the node score update / management unit 20b.

  In the node score update / management unit 20b, when the (n + 1) th measurement cycle is reached, the score C (n + 1, k) of all nodes is set to “C (n + 1, k) = max {C_low, (1-1 / t ′) C”. (N, k) + 1 / t′C ′ (n, k)} ”.

  As described above, C_low is a lower limit value of a predetermined score, 1 / t ′ is a smoothing parameter, and “t ′ = min” using predetermined parameters β (> 0) and c (> 1). {N + β, c} ”, and the initial value of C (n, k) is“ C (0, k) = 1 (all k) ”.

Further, the node score update / management unit 20b counts the number of times C ′ (n, k) at which each node k has become a relay node at time n, and calculates the total value Σ _k C ′ (n, k). The frequency F (n) = Σ _k C ′ (n, k) / {N (N−1)} is calculated and divided by the total number of node pairs N (N−1). If the current time point n is n> T, F_T (n), which means a moving average up to the previous T period, is expressed as “F_T (n) _{= Σx = 1 to} TF ( _nx ) / T When “F (n) / F_T (n)” is smaller than a predetermined threshold value, the score C (n, k) of each node is reinitialized and “C (0, k) "(k = 1 to N), and" time n = 0 ".

3 measures the delay d (21, j) from the own node 21 to the destination node j in the communication quality measuring unit 21a, and moves from the current n own node to all other nodes. The average delay time d_avg (n) = Σ _j d (21, j) / (N−1) is calculated.

  On the other hand, the exponential weighted moving average d_avg_s (n−1) at the time point n−1 is calculated by “d_avg_s (n−1) = (1−α) d_avg_s (n−2) + αd_avg (n−1)”. As described above, α is a predetermined smoothing parameter, and 0 <α <1.

  If “d_avg (n) <d_avg_s (n−1) −th_d” is satisfied, the fact is notified to the node score update / management unit 20b in the management server 20. As described above, th_d is a threshold value, which is predetermined or lower than the delay measurement result history d_avg_s (n−2), d_avg_s (n−3), d_avg (n−4),. The y percent value is obtained, and the difference between the value and d_avg_s (n−1) is set to th_d.

  The node score update / management unit 20b in the management server 20 that has received the change in delay from the node 21 receives the score C (k = 1 to N) of the node k (k = 1 to N) managed by the node score update / management unit 20b. For n, k), it is checked whether or not the node 21 is entered in the upper X node. If not, the score C (n, i) of the node 21 is expressed as “C (n, i) ← C (n , I) + C_add ”. Here, C_add is a score increase amount determined in advance as described above.

  Next, a communication path control method in the overlay network according to the present invention will be described with reference to FIGS. FIG. 5 is a flowchart showing a first processing operation example according to the present invention of the communication path control system including the overlay node and the management server in FIG. 2, and FIG. 6 is a sequence diagram showing the second processing operation example. FIG. 7 is a flowchart showing a third processing operation example, FIG. 8 is a flowchart showing a fourth processing operation example, and FIG. 9 is a sequence diagram showing a fifth processing operation example.

  The example shown in FIG. 5 corresponds to the first technique described above, and in the overlay network that is a logical network composed of N overlay nodes connected to the IP network, the communication path from the source node to the destination node is shown. This is a communication path control procedure that is controlled according to network conditions, and each node i (i = 1 to N) belonging to the overlay network receives the other node j (all except i) with its own node as the source node. As the destination node j, the communication quality between the source node i and the destination node j is measured (step S501), M nodes are selected from the N nodes as relay node candidates, and the selected relay is selected. The communication quality measurement result between the relay candidate node k and the destination node j is acquired from the candidate node k, and the local node i passes through the relay candidate node k. And it calculates the communication quality when reaching the destination node j Te (step S502).

  The calculated communication quality is compared with the measured communication quality. If the measured communication quality is good, a direct communication path from the own node i to the destination node j is determined. If the calculated communication quality is good, the highest quality is obtained. Is determined (step S503), and the frequency of determining the node k * is stored in the storage device (step S504).

  Then, when the next relay node candidate is selected (step S505), M ′ relay node candidates are selected with a probability corresponding to the stored frequency (that is, it becomes easier to be selected in order of increasing frequency), and the remaining M− M ′ selects relay node candidates with equal probability without depending on the frequency (step S506), and further monitors changes in network conditions in which MM ′ relay node candidates are installed (step S506). S507) When a change is detected (step S508), the selection probability of the relay node candidate is changed according to the situation (step S509).

  The example shown in FIG. 6 corresponds to the second technique described above, and each node k determined as a relay node is set in the overlay network to the management server that manages information on candidate relay nodes, and the source node i Information indicating that the own node k has been determined as the relay node is notified as a communication path of a pair of the destination node j and the destination node j (step S601).

  The management server that has received the information from each node k counts the number of times C ′ (n, k) at which the node k has become a relay node at time n, and uses this count value for each time at the next time n + 1. The score of node k is expressed by the expression “C (n + 1, k) = max {C_low, (1-1 / t ′) C (n, k) + 1 / t′C ′ (n, k)}” (“C_low” Is a lower limit value of a predetermined score, “1 / t ′” is a smoothing parameter, and “t ′ = min {n + β, c}” using predetermined parameters β (> 0) and c (> 1). Given, the initial value of C (n, k) is updated by C (0, k) = 1 (all k)) (step S602).

When determining the communication path to the destination node j at the time point n (step S603), the source node i reads the score C (n, k) of each node k from the management server (step S604), and the probability p_k = The process of selecting a node k (k is other than i) as a relay candidate node in C (n, k) / Σ _k C (n, k) is repeated until MM ′ relay candidate nodes are determined (step In addition, the process of selecting relay candidate nodes from the remaining node group with equal probability is repeated until M ′ pieces are selected (steps S607 and S608).

  The example shown in FIG. 7 corresponds to the third technique described above, and the source node i uses one of the M relay candidate nodes as a detour for the communication path from the own node i to the destination node j. When deciding whether to do so, the delay time d (i, k) from the node i to the relay candidate node k and the delay time d (k, j) between the relay candidate node k and the destination node j are used. , Whether or not k satisfying “d (i, j)> d (i, k) + d (k, j)” is present (step S701). A direct communication path to j is determined (step S702), and if it exists, k that minimizes the right side is determined as a relay node k * (step S703).

The example shown in FIG. 8 corresponds to the fourth technique described above, and the management server counts the number of times C ′ (n, k) each node k has become a relay node at time n, The total value Σ _k C ′ (n, k) is calculated (step S801), and the calculated total value Σ _k C ′ (n, k) is divided by the total number of node pairs N (N−1) F (n) ( = Σ _k C 'a (n, k) / {n (n-1)}), towards the detour path is determined as a good frequency quality (step S802).

If the current time point n is n> T, F_T (n), which means a moving average up to the period before T, is expressed as “F_T (n) _{= Σx = 1 to} TF ( _nx ) / T”. If “F (n) / F_T (n)” is smaller than a predetermined threshold value (step S804), the score C (n, k) of each node is calculated. Reinitialization is performed to set C (0, k) (k = 1 to N) and time point n = 0 (step S805).

The example shown in FIG. 9 corresponds to the fifth technique described above, and each overlay node i measures the delay d (i, j) from its own node i to the destination node j (step S901), and the current n The average delay time da “= d_avg (n) = Σ _j d (i, j) / (N−1)” from the own node i to all other nodes is calculated (step S902).

Also, the exponential weighted moving average d _{n−1 at time n−1} (= d_avg_s (n−1)) “d_avg_s (n−1) = (1−α) d_avg_s (n−2) + αd_avg (n−1) ,..., Α is a predetermined smoothing parameter, and is calculated using an equation of 0 <α <1 (step S903).

  Then, “d_avg (n) <d_avg_s (n−1) −th_d,... Th_d is a threshold value and is predetermined or a history of delay measurement results d_avg_s (n−2), d_avg_s (n−3) , D_avg (n−4),..., And obtains the lower y percentage value and sets the difference between that value and d_avg_s (n−1) to th_d ”(step S904), information indicating the change in delay Is notified to the management server (step S905).

  When the management server receives the information indicating the change in delay from the node i (step S906), the management server within the upper X node regarding the score C (n, k) of the node k (k = 1 to N) managed by the server itself Node i is entered (step S907). If not, the score C (n, i) of node I is expressed as “C (n, i) ← C (n, i) + C_add,... C_add” Is a predetermined score increase amount "using the substitution formula (step S908).

  Hereinafter, FIGS. 10 and 11 show the evaluation results when the above-described first technique is executed. In this evaluation, at a computer terminal contracted as a user with 18 ISPs (Internet Service Providers) that are appropriately geographically separated in Japan, 1 packet per second is transmitted for 3 minutes between each terminal. Data obtained by measuring the delay time (maximum delay in 3 minutes) was used (measurement period was 48 hours).

  Using the delay time d (n, i, j) between the terminal i and the terminal j at the n-th time (n = 1, 2,..., 48), 18 terminals are regarded as overlay nodes. Using the above data, the delay between each node was simulated from the time of n = 1 to the 24th hour as measured data, and the following network fluctuations were simulated after the time of n = 24th hour.

  At the time of n = 24 hours, the delay times of the first-ranked node and the 18th-ranked node with the score C (n, k) were switched, and the second-ranked node and the 17th-ranked node were also switched. In other words, when the delay time of the upper node virtually changes to the delay time of the lower node (that is, when the delay time of the upper node deteriorates), the delay time of the lower node is similar to that of the upper node. Simulated the case where the delay time changed and changed.

  Note that the number of candidate relay nodes M = 4, M ′ = 1, “β = 0.5”, “c = 4”, and “C_low = 0.1”.

  FIG. 10 shows a time series of the nth delay time at this time. Here, the 95th percentile of the delay time of all node pairs at the nth time is plotted. Learned in the figure is the result when this example is applied. For comparison, when route control is not performed (that is, when all the direct paths are selected, default in the figure), N = 18 all nodes can be used as relay candidates without limiting the number of relay node candidates. The results for cases (optimal in the figure) are also shown.

  As a result, it can be seen that the present invention is able to improve the quality almost at the same time as the entire route search (optimal) after the n = 24th hour after the change of the network condition.

  For reference, FIG. 11 shows changes in scores of the first, second, third, fourth, 17th, and 18th nodes in the present invention. From this, it can be confirmed that the score of the node that was originally lower can be raised appropriately.

  As described above with reference to FIGS. 2 to 11, according to this example, the communication quality is improved while following the fluctuation of the network condition and reducing the cost associated with the communication path calculation in the overlay network. It becomes possible.

  In addition, this invention is not limited to the example demonstrated using FIGS. 2-11, In the range which does not deviate from the summary, various changes are possible. For example, the computer configuration of the overlay node or the management server may be a computer configuration without a keyboard or optical disk drive. In this example, an optical disk is used as a recording medium, but an FD (Flexible Disk) or the like may be used as a recording medium. As for the program installation, the program may be downloaded and installed via a network via a communication device.

It is explanatory drawing which shows the concept of the route control by an overlay network. It is a block diagram which shows the structural example of the overlay network provided with the communication path control system which concerns on this invention. FIG. 3 is a block diagram illustrating a configuration example of an overlay node in FIG. 2. It is a block diagram which shows the structural example of the management server in FIG. 3 is a flowchart showing a first processing operation example according to the present invention of a communication path control system including an overlay node and a management server in FIG. 2. It is a sequence diagram which shows the 2nd processing operation example which concerns on this invention of the communication path control system which consists of an overlay node in FIG. 2, and a management server. It is a flowchart which shows the 3rd processing operation example which concerns on this invention of the communication path control system which consists of an overlay node in FIG. 2, and a management server. 10 is a flowchart showing a fourth processing operation example according to the present invention of the communication path control system including the overlay node and the management server in FIG. 2. It is a sequence diagram which shows the 5th example of a process operation based on this invention of the communication path control system which consists of an overlay node and a management server in FIG. It is explanatory drawing which shows the effect of this example. It is explanatory drawing which shows transition of a score when this example is applied.

Explanation of symbols

  1-3: Overlay node, 4-8: IP router, 11: Overlay network, 12: IP network, 20: Management server, 20a: Relay node selection frequency management unit, 20b: Node score update / management unit, 21-25 : Overlay node, 21a: communication quality measurement unit, 21b: communication quality acquisition unit, 21c: relay node determination unit, 21d: relay candidate selection unit, 21e: detour route setting unit.

Claims (7)

In an overlay network, which is a logical network composed of N overlay nodes connected to an IP network, a communication path control method for controlling a communication path from a source node to a destination node according to network conditions,
Each node i (i = 1 to N) belonging to the overlay network has its own node as the source node and other nodes j (all except i) as the destination node j, and the source node i and destination node j Measuring the communication quality between,
M nodes are selected from among the N nodes as relay node candidates, and a communication quality measurement result between the relay candidate node k and the destination node j is acquired from the selected relay candidate node k, and Calculating communication quality when reaching the destination node j from the node i via the relay candidate node k;
The calculated communication quality is compared with the measured communication quality. If the measured communication quality is good, a direct communication path from the own node i to the destination node j is determined. If the calculated communication quality is good, Determining a communication path to a destination node j that relays a node k * providing high quality;
Storing the frequency of determining the node k * in a storage device;
When the next relay node candidate is selected, M ′ relay node candidates are selected in descending order of the stored frequency, and the remaining MM ′ candidates are selected with the same probability without depending on the frequency. A step to choose;
Monitoring a change in the network condition where the MM ′ relay node candidates are installed, and changing the selection probability of the relay node candidate according to the situation when the change is detected. An overlay network communication path control method characterized by the above. The overlay network communication path control method according to claim 1,
Each node k determined as the relay node is relayed to a server that is installed in the overlay network and manages information on candidate relay nodes as a communication path of a pair of a source node i and a destination node j. Execute a step of notifying information indicating that it has been determined as a node;
The server that has received the information from each node k counts the number of times C ′ (n, k) at which the node k has become a relay node at time n, and uses each count value at each of the next time n + 1. The score of node k is expressed by the expression “C (n + 1, k) = max {C_low, (1-1 / t ′) C (n, k) + 1 / t′C ′ (n, k)}” (“C_low” Is a lower limit value of a predetermined score, “1 / t ′” is a smoothing parameter, and “t ′ = min {n + β, c}” using predetermined parameters β (> 0) and c (> 1). The initial value of C (n, k) is updated by C (0, k) = 1 (all k)),
When the source node i determines the communication path to the destination node j at the time point n, the score C (n, k) of each node k is read from the server, and the probability p_k = C (n, k) Repeating the process of selecting a node k (k is other than i) as a relay candidate node at / Σ _k C (n, k) until MM ′ relay candidate nodes are determined;
An overlay network communication path control method comprising: performing a process of selecting relay candidate nodes from the remaining node group with equal probability until M ′ is selected. The overlay network communication path control method according to claim 1,
The source node i is
When determining whether or not to bypass using any of the M relay candidate nodes in the communication path from the own node i to the destination node j,
Using the delay time d (i, k) from the node i to the relay candidate node k and the delay time d (k, j) between the relay candidate node k and the destination node j, “d (i, j)> checking whether or not k satisfying “d (i, k) + d (k, j)” exists;
Determining the direct communication path from the local node i to the destination node j if it does not exist, and determining k that minimizes the right side as the relay node k * if it exists. Communication method for overlay network. The overlay network communication path control method according to claim 2,
The management server
Counting the number of times C ′ (n, k) each node k has become a relay node at time n and calculating the total value Σ _k C ′ (n, k) of the count values;
A value F (n) (= Σ _k C ′ (n, k) / {N (N−1)} obtained by dividing the calculated total value Σ _k C ′ (n, k) by the total number N (N−1) of node pairs. ) As a frequency that the detour route is better in quality,
If the current time point n is n> T, F_T (n), which means a moving average up to the period before T, is expressed as “F_T (n) _{= Σx = 1 to} TF ( _nx ) / T”. If “F (n) / F_T (n)” is smaller than a predetermined threshold value, the score C (n, k) of each node is reinitialized and C (0, k) (K = 1 to N) and a step of setting time n = 0. A communication path control method for an overlay network according to claim 2,
Each overlay node i is
Measuring a delay d (i, j) from the local node i to the destination node j;
Calculating an average delay time d_avg (n) = Σ _j d (i, j) / (N−1) from the current node i of the current node n to all other nodes;
The exponential weighted moving average d_avg_s (n−1) at time point n−1 is expressed as “d_avg_s (n−1) = (1−α) d_avg_s (n−2) + αd_avg (n−1),. And calculating using the formula 0 <α <1;
“D_avg (n) <d_avg_s (n−1) −th_d,... Th_d is a threshold value and is predetermined or a history of delay measurement results d_avg_s (n−2), d_avg_s (n−3), d_avg (N-4),... Is obtained, and the difference between the value and d_avg_s (n−1) is set to th_d ”, the information indicating the delay change is notified to the management server. And a step of
When the management server receives the information indicating the change in delay from the node i, the node i is included in the upper X node with respect to the score C (n, k) of the node k (k = 1 to N) managed by the server. If not, the score C (n, i) of the node i is expressed as “C (n, i) ← C (n, i) + C_add,... C_add is a predetermined score increase amount. A communication path control method for an overlay network, comprising a step of increasing using an assignment expression of “some”. A means for executing the steps in the communication route control method for an overlay network according to any one of claims 1 or 3 to measure communication quality between nodes and to determine whether or not to bypass using a relay node; ,
Means for selecting relay candidate nodes by executing each step in the communication route control method for an overlay network according to claim 2;
A means for executing each step in the communication route control method for an overlay network according to claim 4 or 5 to change a score of a node used when selecting a relay node, Communication path control system in overlay network.   The program for making a computer perform each step in the communication path control method of the overlay network in any one of Claims 1-5.

Images