JP2013098646A - Packet transfer system and method, and node device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To probabilistically transfer a packet in a router without routing-processing using a target address described in the packet.SOLUTION: In a node device positioned between adjacent nodes according to the present invention, information relating to the adjacent nodes is repeatedly exchanged between the adjacent nodes, so as to grasp a topology of an entire NW. An allowable hop length upper limit value defined from a shortest hop distance between arbitrary nodes is calculated, and a probability r(n) at which a packet reached from each input port i is transferred to each output port j, and a probability R(n) at which a packet using a node n as a start is transferred to each output port j are calculated. The allowable hop length upper limit value is written in a header of the packet, and the packet is transferred to each output port according to a probability of transfer probability storage means.

Description

  The present invention relates to a packet transfer system and method, and node equipment, and more particularly, from a given node (originating node) in a network (NW) in which a plurality of nodes having a plurality of input ports and output ports are connected in any shape. The present invention relates to a packet transfer system and method for performing packet transfer stochastically without using an address of a destination node in a transit node when performing packet transfer to an arbitrary node (destination node), and a node device.

  The current Internet has a form in which a plurality of NWs operated by ISP (Internet Services Protocol) or the like are connected to each other. Each ISP's NW is composed of one or more autonomous NWs (AS), and each AS can be configured arbitrarily using OSPF (Open Shortest Path First) and between ASs using BGP (Border Gateway Protocol). Packet transfer from the originating terminal to any destination terminal is realized.

  In OSPF, routers exchange the cost information of connecting links with adjacent nodes to grasp the topology structure including link cost information of the entire NW, and use Dijkstra's algorithm to minimize the cost path between each node. Is calculated. Then, each node calculates in advance the output port to be transferred next and manages it as a routing table when a packet having each destination address arrives so that each packet is transferred on the minimum cost path. Such creation / update of the routing table is autonomously performed at each node.

  When a packet arrives at a router, the router searches the routing table using the packet's target address as a key, obtains information about the next router to be forwarded described in the corresponding item, and is connected to that router. Forward incoming packets to the current output port. In some cases, the address of the last-arriving router is written in the routing table, or only a part of the address including the addresses of a plurality of routers is written. The router needs to search for an entry on the routing table having the largest matching area with respect to the target address described in the packet (see, for example, Non-Patent Document 1).

"BUFFALO: Bloom Filter Forwarding Architecture for Large Organizations," ACM CoNEXT 2009.

  However, a method widely used as a method for the router to find an entry in the routing table that most closely matches the destination address of the arrival packet is a method using a hash value obtained using the destination address as a key. When hash values collide, it is necessary to use a measure such as preparing another address separately.

  Also, studies are being made to reduce the processing load required for address lookup using a Bloom filter. However, all of these methods determine the output port to which the packet is to be transferred based on the target address. Therefore, as the NW scale (number of target addresses) and the amount of traffic flowing through the NW increase, it is necessary for packet routing processing. There is a problem that the load increases.

  The present invention has been made in view of the above points, and a packet transfer system and method capable of performing packet transfer stochastically in a router without performing routing processing using a target address described in the packet, and The purpose is to provide node equipment.

In order to solve the above problems, the present invention (Claim 1) is a packet transfer system in a network (NW) in which a plurality of nodes having a plurality of input ports and output ports are connected in an arbitrary shape,
When packet transfer from any node (originating node) to any node (destination node),
The transit node is
There is a transfer processing means for transferring packets by probabilistically selecting the link of the next node to be transferred based on a preset logic without using the destination node address.

The present invention (Claim 2) provides the transfer processing means,
By exchanging information about adjacent nodes with adjacent nodes, the connection structure (hereinafter referred to as “topology”) of the entire network is grasped, and the allowable hop length upper limit value determined from the shortest hop distance between arbitrary nodes Shortest hop route calculation means for calculating and storing the allowable hop upper limit storage means,
A packet transfer means for transferring packets;
Have
The packet transfer means includes
For a packet having itself as a source node and an arbitrary node as a destination node, an allowable hop length upper limit value of the allowable hop upper limit storage means is set as a residual allowable hop length for the set of itself and the destination node, Means for inserting the ID of the node into the packet header and sending it out;
When a packet arrives from an arbitrary input port, confirmation means for confirming whether it matches the destination node ID entered in the header;
If it is determined by the checking means that the packet arrives, the packet that arrived is not transferred. If the packet does not match, the residual allowable hop length is decreased by 1, and if the residual allowable hop length is 2 or more, And means for transferring the packet to another output port excluding an adjacent node to which the packet has been transferred.

Further, according to the present invention (Claim 3), in the transfer processing means,
Transfer probability storage means for storing transfer probability r _ij (n) and R _j (n) of a predetermined packet;
In the packet transfer means,
Using the r _ij (n) and the R _j (n), the node n forwards the arrival packet from the adjacent node i to the adjacent node j with the probability r _ij (n), and also issues the node n. Means for transferring a packet as a node from node n to adjacent node j with probability R _j (n).

Further, according to the present invention (Claim 4), traffic other than the adjacent node j can be obtained under a condition that the traffic amount flowing through the link from the node n to the adjacent node j is equal to or less than a predetermined target value B _nj . The transfer probability r _ij (n) to the adjacent node j for the arrival packets from all the adjacent nodes i is the same as the transfer probability R _j (n) of the packet having the node n as the originating node to the adjacent node j. Transfer probability calculation means for autonomously determining the value of the value at random time intervals and storing it in the transfer probability storage means.

  The present invention (Claim 5) is the transfer probability calculation means according to Claim 4,

をノードｓを発ノード、ノードｄを着ノードとするパケット（以下、「パケットｓｄ」と記す）で、該ノードｄに一つ以上到達する確率とし、また、前記ノードｎにおいて、前記隣接ノードｉからの到着パケットｓｄを前記隣接ノードｊに対して転送した場合と転送しない場合の

Is a packet having the node s as the originating node and the node d as the destination node (hereinafter referred to as “packet sd”), and is a probability of reaching one or more of the nodes d. When the arrival packet sd from is transferred to the adjacent node j and when it is not transferred

の差分を

The difference of

とし、さらに、

And then

が最小となるノードｓとノードｄのペアをｓ*，ｄ*とし、同様に該ノードｎを発ノードとするパケットに対しても、ｉ＝０として定義し、ｒ_ij(n)＞０を満たす隣接ノードｉの中で、

S *, d * is a pair of a node s and a node d that minimizes s, and similarly, i = 0 is defined for a packet having the node n as a source node, and r _ij (n)> 0 Among the neighboring nodes i that satisfy,

が最大の隣接ノードｉをｋ_２とし、ｒ_ij(n)＜１、

Let k ₂ be the adjacent node i with the largest value r _{2 ij} (n) <1,

を満たす隣接ノードｉの中で、

Among adjacent nodes i satisfying

が最小の隣接ノードｉをｋ_１とするとき、該隣接ノードｋ_２からの（ｋ_２の場合はノードｎを発ノードとする）トラヒックの該隣接ノードｊへの転送量を減少させ、逆に隣接ノードｋ_１からの（ｋ_１＝の場合はノードｎを発ノードとする）トラヒックの隣接ノードｊへの転送量を増加させることで、該ノードｎから該ノードｊへのリンク上を流れるトラヒック量を前記目標値Ｂ_njに維持したまま、

When k ₁ is the smallest adjacent node i, the amount of traffic from the adjacent node k ₂ to the adjacent node j (in the case of k ₂ , the node n is the originating node) is reduced, and vice versa. The traffic flowing on the link from the node n to the node j is increased by increasing the transfer amount of the traffic from the adjacent node k ₁ (the node n is the source node in the case of k ₁ =) to the adjacent node j. While maintaining the amount at the target value B _nj ,

となるよう

To be

（もしくはＲ_j(n)）と

(Or R _j (n)) and

（もしくはＲ_j(n)）とを、ノードｎにおいてランダムな時間で更新する手段を含む。

(Or R _j (n)) is updated at a random time in the node n.

  According to the present invention, when a packet is transferred from an arbitrary node (originating node) to an arbitrary node (destination node) in a network in which a plurality of nodes having a plurality of input ports and output ports are connected in an arbitrary shape. In addition, the packet can be transferred stochastically by selecting and transferring the link of the next node to be transferred based on a certain logic without using the destination node address in the transit node.

1 is a system configuration diagram according to an embodiment of the present invention. It is a sequence diagram of a stochastic routing-free packet transfer method in an embodiment of the present invention. It is a figure which shows the minimum arrival probability in the equal allocation method of this invention. It is a figure which shows the minimum arrival probability improvement effect of the maximum minimum arrival rate method of this invention. It is a figure which shows the minimum arrival probability in the maximum minimum arrival rate method of this invention. It is a figure which shows the average arrival probability in the maximum minimum arrival rate method of this invention.

  Embodiments of the present invention will be described below with reference to the drawings.

  In the present invention, when a packet is transferred from an arbitrary node (originating node) to an arbitrary node (destination node) in a network in which a plurality of nodes having a plurality of input ports and output ports are connected in an arbitrary shape, In the transit node, the packet is transferred without using the destination node address by probabilistically selecting the link of the next node to be transferred based on a certain logic.

  FIG. 1 shows a system configuration according to an embodiment of the present invention.

  The probabilistic routing-free packet transfer system shown in FIG. 1 includes an adjacent node information exchange unit 101, a topology information storage unit 102, a shortest pop route calculation unit 103, an allowable hop upper limit storage unit 104, a transfer probability calculation unit 105, and a transfer probability storage. Unit 106 and packet transfer processing unit 107. The topology information storage unit 102, the allowable hop upper limit storage unit 104, and the transfer probability storage unit 106 are storage media such as a hard disk.

  The adjacent node information processing unit 101 notifies the adjacent node of information related to a node to which the adjacent node information processing unit 101 is connected and similar information notified from the adjacent node. By repeating such information exchange between adjacent nodes, the topology of the entire NW is grasped and stored in the topology information storage unit 102.

  The shortest hop route calculation unit 103 calculates the shortest hop route between the arrival and departure nodes using the topology information in the topology information storage unit 102, and calculates the upper limit value of the allowable number of hops from the shortest hop length for each departure and arrival node pair. And stored in the allowable hop upper limit storage unit 104.

Transmission probability calculation unit 105, a probability r _ij (n) to forward packets arriving from the input port i to the output port j, the probability R _j to forward packets to the node n and originating in each output port j ( n) is calculated and stored in the transfer probability storage unit 106.

Each time a packet arrives at node n, the packet transfer processing unit 107 transfers the packet to each output port according to the transfer probability R _j (n) stored in the transfer probability storage unit 106. Further, when transferring a packet originating from the node n, the packet transfer processing unit 107 writes the hop upper limit value stored in the allowable hop upper limit storage unit 104 in the header of the packet, and a transfer probability storage unit According to the transfer probability accommodated in 106, the packet is transferred to each output port.

  FIG. 2 shows a sequence diagram of the probabilistic routing-free packet transfer scheme in one embodiment of the present invention.

  Prior to the start of service operation, the adjacent node information exchanging unit 101 of each node exchanges information regarding the connected node with the adjacent node (step 1), and the shortest hop route calculating unit 103 accepts the allowable hop length for each pair of incoming and outgoing nodes. An upper limit is set in the allowable hop information storage unit 104 (step 2). Then, the transfer probability calculation unit 105 sets the packet transfer probability (step 3).

  During service operation (step 4), every time a packet arrives from the input port or generates a packet having itself as a source node, the packet transfer processing unit 107 sets a predetermined packet transfer probability in the transfer probability storage unit 106. Then, the packet is transferred to each output port (step 5). The transfer probability calculation unit 105 updates the packet transfer probability in the transfer probability storage unit 106 autonomously at further random time intervals (step 6).

  Next, the stochastic routing-free packet transfer method of the present invention will be described in detail.

[Packet transfer method]
Allowed for transfer of a packet (hereinafter referred to as “packet sd”) in which the adjacent node set of node n is A ′ _n , the number of adjacent nodes is A _n , node s is the originating node, and node d is the destination node _Let M _sd be the upper limit of the hop length. It is assumed that the topology information of A ′ _n , A _n , and nodes s and d is stored in the topology information storage unit 102, and M _sd is stored in the allowable hop upper limit storage unit 104, and the transfer probability R _j (n) Is stored in the transfer probability storage unit 106. In the header of the packet, a calling address, a calling auto number, and an allowable residual hop length value (TTL: time to live) are described as in the current Internet.

In response to a packet transfer request generated from a lower-level NW or terminal accommodated in node n and having any other node j as the destination node, the packet transfer processing unit 107 of node n sets the value of M _nd It is described in the TTL of the packet, and is transferred with a probability of R _j (n) for each output port j with j∈A ′ _n . Therefore, there is a possibility that the same packet is output to two or more output ports, and there is a possibility that a packet is not output to any output port.

  The packet transfer processing unit 107 checks whether the destination address d described in the header of the packet is the node n when the packet arrives at the node n from the input port i of i∈A′n, When d = n, packet transfer is not performed.

On the other hand, when d ≠ n, the packet is not transferred when the TTL is checked and 1 is set. In the case of 2 or more, after the value of TTL is decreased by 1, it is transferred with a probability of r _ij (n) to each output port j where j ≠ A′n and j ≠ i. Therefore, the same packet may be output to two or more output ports, or the packet may not be output to any output port.

The transfer probability calculation unit 105 of each node n performs packet transfer using one of the following two methods (i) equal allocation method and (ii) maximum / minimum arrival rate method at autonomously random time intervals. Probabilities R _j (n) and r _ij (n) are set.

(I) Equal allocation method:
Arrival from all adjacent nodes i other than j under the condition that the traffic amount flowing from the node n to the link connected to the adjacent node j of j∈A′n is equal to or less than a predetermined target value B _{n, j} The transfer probability r _ij (n) to the node j for the packet and the transfer probability R _j (n) of the packet having the node n as the originating node to the node j are set to the same value.

  The expected value of the number of packets sd that arrive from the node i to the adjacent node n at the k-th hop after leaving the node s.

とすると、ｋ＝1に対しては、

Then, for k = 1,

が成立する。また２≦ｋ≦Ｍ_sdの各ｋに対して、以下の漸近式が成立する。

Is established. Further, the following asymptotic formula is established for each k of 2 ≦ k ≦ M _sd .

上記の式（２）の漸近式をｋ＝２から順に解くことで、１≦ｋ≦M_sdの各ｋに対して

For each k of 1 ≦ k ≦ M _sd by solving the asymptotic formula of the above equation (2) in order from k = 2.

を得ることができる。

Can be obtained.

  In addition, the number of packets sd that node i arrives at node n per unit time

とすると、

Then,

より得られる。但し、ｖ_sdはノードｓを発ノード、ノードｄを着ノードとするトラヒック量（単位時間あたりに発生するパケット数）である。さらにノードｉからノードｎに単位時間あたりに到着するパケット数をｗ_inとする。

More obtained. However, v _sd is the traffic volume (number of packets generated per unit time) with the node s as the originating node and the node d as the destination node. The number of packets that arrive per unit time and w _in the node n from further node i.

今、任意のｊ∈A'_nに対して、ｉ∈A _n、ｉ≠ｊの全てのｉに対するr_ij(n)と、R _j (n)を同一の値ｒに設定することを考える。ノードｎを発ノードとするパケットの単位時間当たりの発生数をＶ_nとすると、リンク（ｎ，ｊ）に流れるトラヒック量（単位時間あたりのパケット数）は、

Now, for any jεA ′ _n , consider setting r _ij (n) and R _j (n) for all i for iεA _n and i ≠ j to the same value r. Assuming that the number of occurrences of packets originating from the node n per unit time is V _n , the amount of traffic (number of packets per unit time) flowing through the link (n, j) is

となり、これが目標値Ｂ_nj以下となるようｒを設定すればよい。またｒは明らかにｒ∈１を満たすため、ｒを次式で設定する。

Therefore, r may be set so that it becomes equal to or less than the target value B _nj . Since r clearly satisfies rε1, r is set by the following equation.

[最大最小到達率法]
ノードsを発ノード、ノードdを着ノードとするパケットは、経由ノードにおける確率的な転送を反復する結果、目的ノードdに複数のパケットが到着する可能性もあれば、全く到着しない可能性もある。一つ以上のパケットがノードｄに到着した場合に、このパケット転送が成功したものとみなせるため、パケットsdの到達確率y^(sd)をその確率と定義すると、

[Maximum minimum arrival rate method]
A packet with node s as the originating node and node d as the destination node may repeat a stochastic transfer at the transit node, resulting in multiple packets arriving at the destination node d or not at all. is there. When one or more packets arrive at the node d, this packet transfer can be regarded as successful. Therefore, when the arrival probability y ^(sd) of the packet sd is defined as the probability,

となる。均等割当法を用いた場合、特定のsdペア間においてパケット到達確率ｙ^(sd)が低くなる可能性がある。パケット到達確率が低いと、パケットが正しく目的ノードに届くまでに、何度も再送を繰り返す必要があり、データ転送の完了までに要する時間が長くなり、実用上、問題となる。そこで、均等割当法を用いてノードnにおける隣接ノードjに対するパケット転送確率R_j (n)、r_ij (n)を設定した状態において、さらにR_j (n)，ｒ_ij (n)を調整することで、NW上の最小パケット到達確率の向上を図る。

It becomes. When the equal allocation method is used, there is a possibility that the packet arrival probability y ^(sd) is lowered between specific sd pairs. If the packet arrival probability is low, it is necessary to repeat retransmission many times before the packet reaches the target node correctly, and it takes a long time to complete the data transfer, which is a problem in practice. Therefore, packet transmission probability for neighboring nodes j at the node n by using the equally Assignment R _j (n), in a state of setting the r _ij (n), further adjust the _{R j (n), r ij} (n) As a result, the minimum packet arrival probability on the NW is improved.

Hereinafter, a method of adjusting the packet transfer probability r _ij (n), iεA ′ _n , R _j (n) to the output port to the adjacent node j in the node n will be described. When the nodes i and j are adjacent to the node n, the arrival probability of the packet sd when the packet sd arriving from the node i to the node n is transferred to the node j.

とする。同様に、ノードnにノードiから到着したパケットsdをノードｊに転送しない場合の、パケットsdの到達確率を

And Similarly, when the packet sd arriving at the node n from the node i is not transferred to the node j, the arrival probability of the packet sd is

とする。

And

を、ノードｉからノードnに到達したパケットをノードｊに転送する確率がr_ij(n)のときの、パケットsdの到達確率とすると、

Is the arrival probability of the packet sd when the probability that the packet that has reached the node n from the node i is transferred to the node j is r _ij (n),

は、ノードnを経由しないでノードdに到達する場合の貢献成分

Is the contribution component when reaching node d without going through node n

と、ノードnを経由してノードdに到達する場合の貢献成分

And the contribution component when reaching node d via node n

との和となるため、

To be the sum of

で与えられる。同様にノードnを発ノードとするパケットに対しても、ノードnにおいて隣接ノードｊに転送した場合と転送しない場合の到達確率を各々、

Given in. Similarly, for packets originating from the node n, the arrival probabilities when the node n is transferred to the adjacent node j and when it is not transferred are respectively

とすると、到達確率

Then, the arrival probability

は、

Is

となる。

It becomes.

When the packet is transferred again to the node i to which the packet has been transferred, a loopback occurs, so r _ii (n) = 0 is set.

の各々において、これらの値が最小となるsdペアを、各々

Sd pairs that minimize these values in each

とするとき、a_k，b_k，r_kを以下のように定義する。

Then, a _k , b _k and r _k are defined as follows.

また、0≦ｋ≦A'_n，k≠ｊの各ｋに対して、ｙ_kをｙ_k＝ｂ_kｒ_kで定義する。

Further, y _k is _defined as y _k = b _k r _k for each k where 0 ≦ k ≦ A ′ _n and k ≠ j.

In _k satisfying b _k > 0 and r _k <1, _k having the smallest y _k is defined as k ₁ . R _k > 0,
Among k satisfying k ≠ k ₁ , _k having the largest y _k is defined as k ₂ . Decrease the transfer amount of the packet from the input link k ₂ (which originates from the node n when k ₂ = 0) to the output link j, and instead from the input link k ₁ (when k ₁ = 0 By increasing the transfer amount of packets to the output link j (which originates from the node n), the traffic amount of the output link j is maintained below B _nj ,

となるよう次式のように、更新前の確率

As shown in the following equation, the probability before update

と、更新後の確率

And updated probability

を更新する。但し、w_0n=V_nとする。

Update. However, w _0n = V _{n is} assumed.

但し、更新後の

However, after the update

が更新前の

Before update

を下回る場合には、更新は行わず、次の候補をｋ₁に設定して同様に更新を試みる処理を、更新対象が見つかるまで反復する。

If it is less than, the update is not performed, and the process of setting the next candidate to k ₁ and attempting the update in the same manner is repeated until an update target is found.

[Performance evaluation]
For the evaluation, 27 NWs having a node number N of 40 or less are used from commercial ISP backbone NWs whose topology is disclosed on the Web page of CAIDA (Cooperative Association for Internet Data Analysis). The total traffic demand amount of the entire NW (the total for all N nodes of the number of packet transfer requests generated in each node per unit time) V, the AC traffic demand amount is the product of the population ratio of the nodes at both ends Proportional distribution. All initial packet transfer probabilities are set to 0.5. For each sd pair, the via hop count upper limit M _sd is set to β times the shortest hop length H _sd between sd using the real parameter β as follows.

また、ノードij間に設置されたリンクを流れるトラヒック量の目標上限値β_ijを、全てのリンクにおいて同一の値αVに設定する。但し、αは実数パラメータである。

Further, the target upper limit value β _ij of the traffic amount flowing through the link installed between the nodes ij is set to the same value αV for all the links. Where α is a real parameter.

FIG. 3 shows the result of plotting the minimum packet arrival probability y _min in each of 27 NWs using the equal allocation method against α. However, the results when β is set to 1.1 and 1.5 are shown, respectively, and the graph is divided into nine NWs for easy viewing. The process of updating the transfer rate in a randomly selected node was repeated 10N times. Most NWs require, for example, a link capacity of about 2 to 10 times the total demand of the entire NW to achieve y _min = 0.9. In addition, since the full mesh NW (6, 12) has many routes with a hop length of M _sd or less, y _min = 0.9 can be achieved even if the link capacity is about 0.1 to 0.5 times the total demand. The small-scale NW (15, 17) can achieve y _min = 0.9 when α is about 0.9 because the link load is kept low. As β increases, the diversity of route candidates increases and the packet arrival probability improves. On the other hand, the link load increases, so the packet transfer probability must be kept low, and the packet arrival probability decreases. is there. Since the influence of the latter is larger, y _min decreases as β increases.

Further, in FIG. 4, a minimum packet arrival probability y _min in each NW in the case of using the maximum and minimum arrival rate method, is plotted against the value obtained by dividing α by y _min in the case of using the uniform allocation method. In the region of α <1, the minimum packet arrival probability y _min is improved by several times, but the improvement effect is small in the region of α> 1. This means that the minimum packet arrival probability y _min can be improved by adjusting the transfer probability using the maximum / minimum arrival rate method as α is small (link capacity is small) and the condition is severe. Further, the smaller β is, the greater the improvement effect by the maximum / minimum arrival rate method is.

In FIG. 5, the minimum packet arrival probability y _min in each NW when the maximum / minimum arrival rate method is used is plotted against α. In many NWs, a link capacity of about 2 to 10 times the total demand of the entire NW is required to achieve y _min = 0.9. However, the improvement effect of the minimum packet arrival probability y _min is large in the region of 0.1 <α <1, and is about 0.3 to 0.8 in many NWs.

The average arrival probability is defined as the arrival probability y ^sd of each sd pair multiplied by the traffic demand amount ratio of the sd pair. In FIG. 6, the average arrival probability in each NW when the maximum and minimum arrival rate method is used is plotted against α. In many NWs, the average arrival probability is close to 1 if there is a link capacity corresponding to the traffic demand of the entire NW. In addition, the average arrival probability slightly decreases as β increases.

  As described above, in the present invention, the output port to which the packet is to be transferred is determined based on the target address. Therefore, as the NW scale (number of target addresses) and the amount of traffic flowing through the NW increase, packet routing processing is required. The problem of the prior art that the load increases can be solved.

  In the present invention, the operation of each component of the node shown in FIG. 1 is constructed as a program, installed in a computer used as each node device that performs packet transfer, or executed via a network. Can be distributed.

  The present invention is not limited to the above-described embodiments, and various modifications and applications can be made within the scope of the claims.

101 Adjacent node information exchange unit 102 Topology information storage unit 103 Shortest hop route calculation unit 104 Allowable hop upper limit storage unit 105 Transfer probability calculation unit 106 Transfer probability storage unit 107 Packet transfer processing unit

Claims (7)

A packet transfer system in a network (NW) in which a plurality of nodes having a plurality of input ports and output ports are connected in an arbitrary shape,
When packet transfer from any node (originating node) to any node (destination node),
The transit node is
A packet transfer system comprising transfer processing means for transferring packets by probabilistically selecting a link of a next node to be transferred based on a preset logic without using an address of a destination node . The transfer processing means includes
By exchanging information about adjacent nodes with adjacent nodes, the connection structure (hereinafter referred to as “topology”) of the entire network is grasped, and the allowable hop length upper limit value determined from the shortest hop distance between arbitrary nodes Shortest hop route calculation means for calculating and storing the allowable hop upper limit storage means,
A packet transfer means for transferring packets;
Have
The packet transfer means includes
For a packet having itself as a source node and an arbitrary node as a destination node, an allowable hop length upper limit value of the allowable hop upper limit storage means is set as a residual allowable hop length for the set of itself and the destination node, Means for inserting the ID of the node into the packet header and sending it out;
When a packet arrives from an arbitrary input port, confirmation means for confirming whether it matches the destination node ID entered in the header;
If it is determined by the checking means that the packet arrives, the packet that arrived is not transferred. If the packet does not match, the residual allowable hop length is decreased by 1, and if the residual allowable hop length is 2 or more, The packet forwarding system according to claim 1, further comprising means for forwarding to another output port excluding an adjacent node to which the packet has been forwarded. The transfer processing means includes
Transfer probability storage means for storing transfer probability r _ij (n) and R _j (n) of a predetermined packet;
The packet transfer means includes
Using the r _ij (n) and the R _j (n), the node n forwards the arrival packet from the adjacent node i to the adjacent node j with the probability r _ij (n), and also issues the node n. 3. The packet forwarding system according to claim 2, further comprising means for forwarding a packet to be a node from the node n to the adjacent node j with a probability R _j (n). Under the condition that the amount of traffic flowing through the link connecting from the node n to the adjacent node j is equal to or less than a predetermined target value B _{nj, the} received packets from all the adjacent nodes i other than the adjacent node j The transfer probability r _ij (n) to the adjacent node j and the transfer probability R _j (n) of the packet originating from the node n to the adjacent node j are set to the same value at random time intervals. 4. The packet transfer system according to claim 3, further comprising transfer probability calculation means for obtaining and storing the transfer probability storage means. The transfer probability calculation means includes:

Is a packet having the node s as the originating node and the node d as the destination node (hereinafter referred to as “packet sd”), and is a probability of reaching one or more of the nodes d. When the arrival packet sd from is transferred to the adjacent node j and when it is not transferred

The difference of

And then

S *, d * is a pair of a node s and a node d that minimizes s, and similarly, i = 0 is defined for a packet having the node n as a source node, and r _ij (n)> 0 Among the neighboring nodes i that satisfy,

Let k ₂ be the adjacent node i with the largest value r _{2 ij} (n) <1,

Among adjacent nodes i satisfying

When k ₁ is the smallest adjacent node i, the amount of traffic from the adjacent node k ₂ to the adjacent node j (in the case of k ₂ , the node n is the originating node) is reduced, and vice versa. The traffic flowing on the link from the node n to the node j is increased by increasing the transfer amount of the traffic from the adjacent node k ₁ (the node n is the source node in the case of k ₁ =) to the adjacent node j. While maintaining the amount at the target value B _nj ,

To be

(Or R _j (n)) and

The packet transfer system according to claim 4, further comprising means for updating (or R _j (n)) at random time in the node n. A packet transfer method in a network (NW) in which a plurality of nodes having a plurality of input ports and output ports are connected in an arbitrary shape,
In a node device located between adjacent nodes and transferring packets from an arbitrary node (originating node) to an arbitrary node (destination node),
The shortest hop route calculation means keeps track of the connection structure (hereinafter referred to as “topology”) of the entire NW by repeatedly exchanging information about adjacent nodes with adjacent nodes, and the shortest hop distance between any nodes Calculating an allowable hop length upper limit value determined from the shortest hop route calculating step for storing in an allowable hop upper limit storage means;
The probability r _ij (n) of the transfer probability calculation means for transferring a packet arriving from each input port i to each output port j, and the probability R _j (n for transferring a packet originating from node n to each output port j ) And storing the transfer probability in the transfer probability storage means,
A packet transfer step in which packet transfer means writes the allowable hop length upper limit value stored in the allowable hop upper limit storage means in a packet header, and transfers the packet to each output port according to the probability of the transfer probability storage means; ,
Consists of
In the transfer probability calculation step,
Under the condition that the amount of traffic flowing through the link connecting from the node n to the adjacent node j is equal to or less than a predetermined target value B _{nj, the} received packets from all the adjacent nodes i other than the adjacent node j The transfer probability r _ij (n) to the adjacent node j and the transfer probability R _j (n) of the packet originating from the node n to the adjacent node j are set to the same value at random time intervals. Obtained and stored in the transfer probability storage means,
In the packet transfer step,
Using the r _ij (n) and the R _j (n), the node n forwards the arrival packet from the adjacent node i to the adjacent node j with the probability r _ij (n), and also issues the node n. A packet transfer method characterized by transferring a packet to be a node from a node n to an adjacent node j with a probability R _j (n). In a packet transfer system in a network (NW) in which a plurality of nodes having a plurality of input ports and output ports are connected in an arbitrary shape, located between adjacent nodes, from an arbitrary node (originating node) to an arbitrary node (destination node) A node device that forwards packets to
By exchanging information about adjacent nodes with adjacent nodes, the connection structure (hereinafter referred to as “topology”) of the entire network is grasped, and the allowable hop length upper limit value determined from the shortest hop distance between arbitrary nodes Shortest hop route calculation means for calculating and storing the allowable hop upper limit storage means,
Calculating a probability r _ij (n) of transferring a packet arriving from each input port i to each output port j and a probability R _j (n) of transferring a packet originating from the node n to each output port j; Transfer probability calculation means for storing in the transfer probability storage means;
A packet transfer means for entering the allowable hop length upper limit value stored in the allowable hop upper limit storage means in a packet header, and transferring the packet to each output port according to the probability of the transfer probability storage means;
Have
The transfer probability calculation means includes:
Under the condition that the amount of traffic flowing through the link connecting from the node n to the adjacent node j is equal to or less than a predetermined target value B _{nj, the} received packets from all the adjacent nodes i other than the adjacent node j The transfer probability r _ij (n) to the adjacent node j and the transfer probability R _j (n) of the packet originating from the node n to the adjacent node j are set to the same value at random time intervals. Means for obtaining and storing in the transfer probability storage means,
The packet transfer means includes
Using the r _ij (n) and the R _j (n), the node n forwards the arrival packet from the adjacent node i to the adjacent node j with the probability r _ij (n), and also issues the node n. A node device comprising: means for transferring a node packet with probability R _j (n) from node n to adjacent node j.

Images