Method for dynamic communication path selection and switching for full inter-networking

Abstract

The application discloses a dynamic communication path selection and switching method for a full interconnection network, specifically comprising the following steps: renumbering all nodes in the full interconnection network; planning a communication path between any two nodes in the full interconnection network; after all the communication paths are sorted out, each communication path is tried in turn, and the message sending time corresponding to each path is recorded; and the time of acquiring the message transmission of each path is taken as the basis to select the shortest message transmission path from the historical record of message transmission as the message transmission path in the future. The application is based on the improved e greedy algorithm, dynamically adjusts the communication path of concurrent tasks, frequently selects and switches the communication path between communication nodes, avoids the subnet area that may have a conflict in the interconnection network, and further reduces the communication delay and the processing time of concurrent tasks in the network.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of Internet communication, and particularly relates to a dynamic communication path selection and switching method for a full interconnection network. BACKGROUND

[0002] A full interconnection network is a network structure extended from a hypercube network structure, usually composed of medium-scale PCs, and is often used to support the concurrency of interactive scientific simulation programs. In the full interconnection network, each node belongs to different subnets at the same time, and the subnets are connected to each other to form a cubic structure network. This network structure is different from the general hypercube network. Each node in the full interconnection network belongs to three-dimensional subnets according to the X-Y-Z axis. The full interconnection network, as an internally interconnected network structure, also supports three-dimensional neighborhood and three-dimensional reduction communication modes. Such an interconnection network makes it possible to use different types of tasks to be executed at the same time through its redundant subnets and appropriate communication scheduling.

[0003] Specifically, Figure 1 A structural diagram of the full interconnection network is shown. In the figure, 64 nodes are arranged in a regular cube. According to the position on each axis, each node of the full interconnection network can be marked as a three-dimensional coordinate (N x , N y , N z , ) (0≤N x , N y , N z ≤3). In addition, the full interconnection network is composed of three groups of switches (SW-Z, SW-Y, SW-SX), and each switch is composed of several independent small-scale ready-made gigabit Ethernet switches (abbreviated as GbESW). The nodes marked (i, *, *) are connected to SW-Zi, the nodes marked (*, j, *) are connected to SW-Yj, and the nodes marked (*, *, k) are connected to SW-Xk.

[0004] The full interconnection network with full connection topology of each subnet can support the concurrent execution of multiple tasks and fully release the computing power of the full interconnection network. However, when a large number of tasks are running on the full interconnection network at the same time, there may be multiple tasks passing through the same subnet for message passing. Based on this, these tasks will be affected by the network bandwidth, resulting in network congestion and conflicts, prolonging the message sending and waiting time, and thus prolonging the running time of these tasks. In order to avoid this situation, it is necessary to reasonably plan the message passing path of each task in the interconnection network, avoid congestion, and try to eliminate message retransmission and waiting time. SUMMARY

[0005] The application aims to provide a dynamic communication path selection and switching method for a full interconnection network, which selects and switches the communication path between communication nodes to avoid the conflict sub-network area in the interconnection network and reduce the communication delay.

[0006] The technical solution of the application is a dynamic communication path selection and switching method for a full interconnection network, which is implemented according to the following steps:

[0007] Step 1, renumber all nodes in the full interconnection network;

[0008] Step 2, plan the communication path between any two nodes in the full interconnection network;

[0009] Step 3, after all the communication paths are sorted out, each communication path is tried in turn, and the message sending time corresponding to each path is recorded; the message sending time of each communication is recorded by the time stamp function in SOCKET programming to form a historical record;

[0010] Step 4, the time of transmitting the message of each path is taken as the basis to select the message transmission path with the shortest time from the historical record of message transmission as the message transmission path in the future.

[0011] The application also has the following characteristics:

[0012] In step 1, the following is specifically provided:

[0013] The full interconnection network is composed of nodes numbered 1-64 in order, the numbers 1-64 are renumbered, and each number is converted into a three-bit 4 binary number in the order of X-Y-Z axis; that is, for any node numbered n, it can be converted into n=4 0 ×x+4×y+4 2 ×z form, then the node n can be defined in the form of Node(x,y,z) in the new numbering rule form, after renumbering, the position of the node n in the XYZ coordinate system can be determined; that is, for the Node(x,y,z) corresponding to the node n, (x,y) indicates the position of the node n in the sub-network SWZ z ; (x,z) represents that the node n is in the xth row and zth column in the sub-network SWY y ; (y,z) indicates the position of the node n in the sub-network SWX x .

[0014] In step 2, the following is specifically provided:

[0015] Step 2.1, calculate the relative difference of the nodes needing communication in different coordinate axes;

[0016] For any node A(x s ,ys ,z s ) and B(x e ,y e ,z e If node A needs to send a message to node B, then calculate the relative distance difference between node A and B on different coordinate axes; as shown in equation (1);

[0017]

[0018] Where Δx, Δy, and Δz are the distance differences between nodes A and B on the X-axis, Y-axis, and Z-axis, respectively;

[0019] Step 2.2: Based on the distance difference, determine whether nodes A and B are in the same subnet;

[0020] If two of Δx, Δy, and Δz have a value of 0, it means that nodes A and B exist in two different subnets of the fully interconnected network. Proceed to step 2.3.1.

[0021] If any of Δx, Δy, or Δz has a value of 0, it means that nodes A and B exist in at least the same subnet of the fully interconnected network, and proceed to step 2.3.2;

[0022] If Δx, Δy, and Δz are all non-zero, it means that nodes A and B belong to different subnets in the fully interconnected network, and proceed to step 2.3.3;

[0023] Step 2.3: Based on the subnet determination results, list the possible communication paths between nodes A and B for different scenarios; specifically:

[0024] Step 2.3.1: If it is determined that nodes A and B exist in two different subnets of the fully interconnected network, then the possible communication paths between nodes A and B are as follows:

[0025] ①If This indicates that nodes A and B both exist in the subnet. In the middle, nodes A and B can be reached through... With their help, the message was exchanged between them through a single message relay.

[0026] ②If This indicates that nodes A and B both exist in the subnet. In the middle, nodes A and B can be reached through... With their help, the message was exchanged between them through a single message relay.

[0027] ③If This indicates that nodes A and B both exist in the subnet. In the middle, nodes A and B can be reached through... With their help, the message was exchanged between them through a single message relay.

[0028] Step 2.3.2: If nodes A and B exist in at least the same subnet of the fully interconnected network, then the possible communication paths between nodes A and B are as follows:

[0029] ① If only Δx = 0, it means that nodes A and B both exist in subnet SWXn. Nodes A and B can be connected via subnet SWXn (n = x). e After one message exchange, the message exchange between them is completed;

[0030] ② If only Δy = 0, it means that nodes A and B both exist in subnet SWYn. Nodes A and B can be connected via subnet SWYn (n = y e After one message exchange, the message exchange between them is completed;

[0031] ③ If only Δz = 0, it means that nodes A and B both exist in subnet SWZn. Nodes A and B can use subnet SWZn (n = z) to connect. e After one message exchange, the message exchange between them is completed;

[0032] Step 2.3.3: If nodes A and B belong to different subnets of the fully interconnected network, then there are a total of 6 shortest message passing paths between nodes A and B:

[0033] ① The message is first relayed through subnet X, and then delivered through subnet Y. That is, the message transmission path requires the assistance of relay node A′(x). s ,y e ,z e To achieve this, firstly A(x) s ,y s ,z s )→A′(x s ,y e ,z e ), through subnet SWXn (n = x s ) relay message, then A′(x s ,y e ,z e )→B(x e ,y e ,z e ), through subnet SWYn (n = y e ) Deliver the message;

[0034] ② First, the message is relayed through subnet X, and then delivered through subnet Z. That is, the message transmission path requires the assistance of relay node A′(x). s ,y e ,z eTo achieve this, firstly A(x) s ,y s ,z s )→A′(x s ,y e ,z e ), through subnet SWXn (n = x s ) relay message, then A′(x s ,y e ,z e )→B(x e ,y e ,z e ), through subnet SWZn (n = z e ) Deliver the message;

[0035] ③ The message is first relayed through subnet Y, and then delivered through subnet X. That is, the message transmission path requires the assistance of relay node A′(x). e ,y s ,z e To achieve this, firstly A(x) s ,y s ,z s )→A′(x e ,y s ,z e ), through subnet SWYn (n = y s ) relay message, then A′(x e ,y s ,z e )→B(x e ,y e ,z e ), through subnet SWXn (n = x e ) Deliver the message;

[0036] ④ First, the message is relayed through subnet Y, and then delivered through subnet Z. That is, the message transmission path requires the assistance of relay node A′(x). e ,y s ,z e To achieve this, firstly A(x) s ,y s ,z s )→A′(x e ,y s ,z e ), through subnet SWYn (n = y s ) relay message, then A′(x e ,y s ,z e )→B(x e ,y e ,z e ), through subnet SWZn (n = ze ) Deliver the message;

[0037] ⑤ First, the message is relayed through the Z subnet, and then delivered through the X subnet. That is, the message transmission path requires the assistance of relay node A′(x). e ,y e ,z s To achieve this, firstly A(x) s ,y s ,z s )→A′(x e ,y e ,z s ), through subnet SWZn (n = z s ) relay message, then A′(x e ,y e ,z s )→B(x e ,y e ,z e ), through subnet SWXn (n = x e ) Deliver the message;

[0038] ⑥ First, the message is relayed through subnet Z, and then delivered through subnet Y. That is, the message transmission path requires the assistance of relay node A′(x). e ,y e ,z s To achieve this, firstly A(x) s ,y s ,z s )→A′(x e ,y e ,z s ), through subnet SWZn (n = z s ) relay message, then A′(x e ,y e ,z s )→B(x e ,y e ,z e ), through subnet SWYn (n = y e ) Deliver the message;

[0039] Step 2.4: Based on the above analysis, all communication paths based on the relative positions of nodes A and B are obtained.

[0040] In step 4, specifically: based on the improved greedy e algorithm, solve the EE problem and select the path with the shortest message transmission time as the communication path between nodes A and B;

[0041] The greed level e is dynamically adjusted in each round of decision-making so that in each round of decision-making, exploration can be carried out with a probability of e(k), and the path with the shortest known message transmission time can be selected from the historical records with a probability of 1-e(k) as the current message transmission path; the path selection strategy is continuously adjusted so that the message transmission between nodes is always kept in a short state until the communication between nodes A and B is declared over.

[0042] The formula for calculating e(k) is shown in equation (2):

[0043] e(k) = K ρ [1-γ{e(kd)-e(kd-1)} 2 ]Δr (2);

[0044] In the formula, e(k) represents the e value of the lazy decision-making module when it makes a path selection in the kth instance, that is, with probability e(k) in the kth instance, K p γ represents the degree of stimulation of the current pathfinding module by the historical optimal route, and γ is the negative feedback adjustment of the pathfinding module to the historical optimal module; d is a hyperparameter, representing the d-th decision, and 0 < d < k.

[0045] The calculation of Δr is shown in formula (3):

[0046] Δr=r(k)-r(kd) (3);

[0047] In the formula, r represents the selection preference of the lazy decision-making module, and its calculation formula is: C(k) represents the number of times the current optimal path is selected at time k, and T(k) represents the total number of candidate paths at time k.

[0048] The beneficial effects of this invention are that, based on the real-time network status of the entire interconnected network and an improved greedy algorithm, the invention dynamically adjusts the communication paths of concurrent tasks. By frequently selecting and switching the communication paths between communication nodes, it avoids subnet areas in the interconnected network that may conflict, thereby reducing communication latency and the processing time of concurrent tasks in the network. Attached Figure Description

[0049] Figure 1 This is a schematic diagram of the structure of a fully interconnected network;

[0050] Figure 2 This is a diagram showing the positions of the nodes in the coordinate system after renumbering. Detailed Implementation

[0051] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0052] This invention provides a dynamic communication path selection and switching method for fully interconnected networks, implemented according to the following steps:

[0053] Step 1: Renumber all nodes in the fully interconnected network;

[0054] The fully interconnected network is composed of nodes numbered 1-64 arranged sequentially. These nodes are then renumbered according to the XYZ axes, and each number is converted into a three-digit 4-ary number. That is, any node numbered n can be converted into n=4... 0 ×x+4×y+4 2 If the form is ×z, then node n can be defined in the new numbering rule as Node(x,y,z), such as Figure 2 As shown, after renumbering, the position of node n in the XYZ coordinate system can be uniquely determined; at the same time, any pairwise combination of the three coordinate components also indicates the position of the node in a certain subnet; that is, for Node(x,y,z) corresponding to node n, (x,y) indicates that node n is in subnet SWZ. z The position within the subnet; (x, z) represents the position of node n within the subnet SWY. y The x-th row and z-th column of the subnet SWX; (y, z) refers to node n in the subnet SWX. x The location inside;

[0055] Step 2, plan the communication path between any two nodes in the fully interconnected network, specifically as follows:

[0056] Step 2.1: Calculate the relative differences between the nodes that need to communicate on different coordinate axes;

[0057] For any node A(x) s ,y s ,z s ) and B(x e ,y e ,z e For example, if node A needs to send a message to node B, it is necessary to calculate the relative distance difference between node A and B on different coordinate axes; the formula for calculating the distance difference is shown in equation (1).

[0058]

[0059] Where Δx, Δy, and Δz are the distance differences between nodes A and B on the X-axis, Y-axis, and Z-axis, respectively;

[0060] Step 2.2: Based on the distance difference, determine whether nodes A and B are in the same subnet;

[0061] If two of Δx, Δy, and Δz have a value of 0, it means that nodes A and B exist in two different subnets of the fully interconnected network. Proceed to step 2.3.1.

[0062] If any of Δx, Δy, or Δz has a value of 0, it means that nodes A and B exist in at least the same subnet of the fully interconnected network, and proceed to step 2.3.2;

[0063] If Δx, Δy, and Δz are all non-zero, it means that nodes A and B belong to different subnets in the fully interconnected network, and proceed to step 2.3.3;

[0064] Step 2.3: Based on the subnet determination results, list the possible communication paths (or message passing paths) between nodes A and B, categorized by case; specifically:

[0065] Step 2.3.1: If it is determined that nodes A and B exist in two different subnets of the fully interconnected network, then the possible communication paths between nodes A and B are as follows:

[0066] [1] If This indicates that nodes A and B both exist in the subnet. In the middle, nodes A and B can be reached through... With their help, the message was exchanged between them through a single message relay.

[0067] [2] If This indicates that nodes A and B both exist in the subnet. In the middle, nodes A and B can be reached through... With their help, the message was exchanged between them through a single message relay.

[0068] [3] If This indicates that nodes A and B both exist in the subnet. In the middle, nodes A and B can be reached through... With their help, the message was exchanged between them through a single message relay.

[0069] Step 2.3.2: If nodes A and B exist in at least the same subnet of the fully interconnected network, then the possible communication paths between nodes A and B are as follows:

[0070] [1] If only Δx = 0, it means that nodes A and B exist simultaneously in subnet SWXn. Nodes A and B can use subnet SWXn (n = x) to connect. e After one message exchange, the message exchange between them is completed;

[0071] [2] If only Δy = 0, it means that nodes A and B exist simultaneously in subnet SWYn. Nodes A and B can use subnet SWYn (n = y) to... e After one message exchange, the message exchange between them is completed;

[0072] [3] If only Δz = 0, it means that nodes A and B exist simultaneously in subnet SWZn. Nodes A and B can use subnet SWZn (n = z) to connect. e After one message exchange, the message exchange between them is completed;

[0073] Step 2.3.3: If nodes A and B belong to different subnets of the fully interconnected network, then there are 6 possible shortest message passing paths between nodes A and B:

[0074] [1] The message is first relayed through subnet X, and then delivered through subnet Y. In this case, the message transmission path between A and B needs to rely on relay node A′(x s ,y e ,z e The specific path is as follows:

[0075] First A(x) s ,y s ,z s )→A′(x s ,y e ,z e ), via subnet SWXn (n=xs) relay message, then A′ (x s ,y e ,z e )→B(x e ,y e ,z e The message is delivered via the subnet SWYn (n = ye).

[0076] [2] First, the message is relayed through the X subnet, and then delivered through the Z subnet. In this case, the message transmission path between A and B needs to rely on the relay node A′(x s ,y e ,z e The specific path is as follows:

[0077] First A(x) s ,y s ,z s )→A′(x s ,y e ,z e ), through subnet SWXn (n = x s ) relay message, then A′(x s ,ye ,z e )→B(x e ,y e ,z e ), through subnet SWZn (n = z e ) Deliver the message.

[0078] [3] The message is first relayed through the Y subnet and then delivered through the X subnet. In this case, the message transmission path between A and B needs to rely on the relay node A′(x e ,y s ,z e The specific path is as follows:

[0079] First A(x) s ,y s ,z s )→A′(x e ,y s ,z e ), through subnet SWYn (n = y s ) relay message, then A′(x e ,y s ,z e )→B(x e ,y e ,z e ), through subnet SWXn (n = x e ) Deliver the message.

[0080] [4] The message is first relayed through the Y subnet and then delivered through the Z subnet. In this case, the message transmission path between A and B needs to rely on the relay node A′(x e ,y s ,z e The specific path is as follows:

[0081] First A(x) s ,y s ,z s )→A′(x e ,y s ,z e ), through subnet SWYn (n = y s ) relay message, then A′(x e ,y s ,z e )→B(x e ,y e ,z e ), through subnet SWZn (n = z e ) Deliver the message.

[0082] [5] The message is first relayed through the Z subnet and then delivered through the X subnet. In this case, the message transmission path between A and B requires the help of relay node A′(x e ,y e ,z s The specific path is as follows:

[0083] First A(x) s ,y s ,z s )→A′(x e ,y e ,z s ), through subnet SWZn (n = z s ) relay message, then A′(x e ,y e ,z s )→B(x e ,y e ,z e ), through subnet SWXn (n = x e ) Deliver the message.

[0084] [6] The message is first relayed through the Z subnet, and then delivered through the Y subnet. In this case, the message transmission path between A and B needs to rely on the relay node A′(x e ,y e ,z s The specific path is as follows:

[0085] First A(x) s ,y s ,z s )→A′(x e ,y e ,z s ), through subnet SWZn (n = z s ) relay message, then A′(x e ,y e ,z s )→B(x e ,y e ,z e ), through subnet SWYn (n = y e ) Deliver the message.

[0086] Step 2.4: Based on the above analysis, all communication paths based on the relative positions of nodes A and B are obtained;

[0087] Step 3: After compiling all communication paths, try each communication path in turn and record the message sending time corresponding to each path. The specific method is to use the timestamp function in SOCKET programming to record the message sending time of each communication, form a historical record, and provide a reference for selecting the path with the shortest message transmission time.

[0088] Step 4: Based on the time of message transmission for each path, select the message transmission path with the shortest time from the message transmission history as the future message transmission path.

[0089] The mathematical method used is an improved greedy algorithm, which abstracts the path selection problem into an EE (Exploration and Exploitation) problem. In the EE problem, exploration refers to choosing a new message transmission path to shorten message propagation latency; while exploitation refers to using the path with the shortest known message transmission time to maximize the benefit and ensure that the message transmission latency between nodes A and B is minimized under the current network conditions.

[0090] The traditional e-greedy method is an improvement on the greedy algorithm. It trades off exploration and exploitation based on a probability, with the value of e remaining constant. Applying this method to a lazy decision-making module can lead to problems. If e is too small, the module cannot quickly find the shortest message-passing path, resulting in slow convergence. If e is too large, while it ensures fast convergence and finds the shortest path within a short time window, it still has a high probability of exploring new paths after finding the shortest one, rather than fully utilizing the already found path. This can also make the lazy decision-making module unstable and prone to making suboptimal decisions. Therefore, the method used in this invention differs from the traditional e-greedy algorithm. By dynamically adjusting the value of e, the module can achieve a more robust selection strategy while ensuring fast convergence. Specifically, when the lazy decision-making module starts working, the initial value of e is set relatively large to quickly find the shortest message-passing path. As the number of decisions increases, more historical data is collected. The e-value is adjusted, gradually decreasing from a high level to a low level, so that the decision-making module can better utilize the shortest message transmission path that has been "discovered" to continue transmitting messages. This decision-making process continues until the network experiences severe fluctuations and the currently known optimal decision changes. At this point, the e-value will continue to change, shifting the strategy of the lazy decision-making module from "conservative" to "aggressive" in order to quickly re-determine the shortest message transmission path.

[0091] Based on the improved e-greedy algorithm, solve the EE problem and select the path with the shortest message transmission time as the communication path between A and B. Specifically: First, improve the traditional e-greedy method to make the greed degree e dynamically adjustable. Specifically, use formula (2) to dynamically adjust the value of e at each round of decision-making, so that in each round of decision-making, a path can be randomly selected from the candidate paths with the probability of e(k), and the path with the shortest known message transmission time in the historical record can be selected as the message transmission path of this time with the probability of 1 - e(k).

[0092] e(k) = K ρ [1 - γ{e(k - d) - e(k - d - 1)} 2 Δr (2);

[0093] In formula (2), e(k) represents the e value when the lazy decision-making module makes a path selection at the k-th time. That is, at the k-th time, a path is randomly selected from the candidate paths with the probability of e(k) as the message transmission path of this time, and the path with the shortest known message transmission time in the historical record is selected as the message transmission path of this time with the probability of 1 - e(k).

[0094] K p represents the stimulation degree of the historical optimal line to the current path search module. γ is the negative feedback adjustment of the path search module to the historical optimal module. d is a hyperparameter representing the d-th decision, and 0 < d < k. Through these parameters, the stability of the policy decision can be ensured, and it can be ensured that the lazy decision-making module will not make overly drastic parameter adjustments when facing severe fluctuations in historical data;

[0095] r represents the selection preference of the lazy decision-making module, and its calculation formula is where C(k) is the number of times the current optimal path is selected at time k, and T(k) represents the total number of candidate paths at time k. The calculation of Δr is shown in formula (3).

[0096] Δr = r(k) - r(k - d) (3);

[0097] On the basis of clarifying the improved e-greedy algorithm, start using this algorithm to select the communication path between nodes A and B. Initially, the e value is set relatively large to quickly "explore" the message transmission path with the shortest time. As the number of decision-making times increases, the historical data collected by the dynamic monitoring module also increases. Then use formula (3) to adjust the e value, so that the e value gradually drops from a high level to a low level, so that the decision-making module can better use the message transmission path with the shortest time that has been "explored" to continue transmitting messages. This decision-making process continues until the network fluctuates violently. If the current known optimal decision has changed, then the e value will continue to change next, and the strategy of the lazy decision-making module will change from "conservative" to "aggressive" to quickly re-determine the message transmission path with the shortest time. The specific algorithm is as follows:

[0098] Algorithm 1: Lazy Decision Module Based on Improved E-Greedy Algorithm

[0099] Input: Candidate message passing path array N[t], inhibition factor d, stimulation parameter K p

[0100] Output: Decision record H[n] and corresponding time T[n]

[0101] Begin

[0102] Initialize the historical record array H[n], the historical time consumption array T[n], and the greedy probability e.

[0103] Choose any path N[t]. i Transmit messages and record the time t.

[0104] for(i=0,i<=n,i++)

[0105] N i Include H[i], and include the corresponding time t in T[i].

[0106] Calculate the inert decision module selection preference r(i) = C(i) / T(i) in the i-th round.

[0107] Calculate the i-th round according to formula (3)

[0108] Calculate e in the i-th round according to formula (2)

[0109] r = a random number between 0 and 1

[0110] If r>e then

[0111] Randomly select a path from N[t] as the new N. i The corresponding time t is included in T[i].

[0112] else select the shortest path in H[n] as the new N. i The corresponding time t is included in T[i].

[0113] return H[n], T[n]

[0114] End

[0115] This process continues, constantly updating the message sending time for each communication and continuously adjusting e(k). Simultaneously, the new e(k) is combined with Algorithm 1 to continuously adjust the network path selection strategy, ensuring that message passing between nodes remains consistently short. This continues until communication between nodes A and B is declared complete.

Claims

1. A dynamic communication path selection and switching method for fully interconnected networks, characterized in that, The specific steps are as follows: Step 1: Renumber all nodes in the fully interconnected network; Step 2: Plan the communication path between any two nodes in the fully interconnected network; Step 3: After compiling all communication paths, try each communication path in turn and record the message sending time corresponding to each path; The timestamp function in SOCKET programming is used to record the message sending time of each communication, forming a historical record. Step 4: Based on the message transmission time of each path, select the message transmission path with the shortest transmission time from the message transmission history as the subsequent message transmission path. Specifically: Based on improvement e A greedy algorithm is used to solve the EE problem, selecting the path with the shortest message transmission time as the communication path between nodes A and B. Degree of greed e The value is dynamically adjusted in each round of decision-making, so that in each round of decision-making, it can be... e ( k Explore with a probability of 1- e ( k The probability is used to select the path with the shortest known message transmission time from the historical records as the current message transmission path; the path selection strategy is continuously adjusted to keep the message transmission between nodes in a short state until the communication between nodes A and B is declared over.

2. The dynamic communication path selection and switching method for fully interconnected networks according to claim 1, characterized in that, In step 1, specifically: The fully interconnected network is composed of nodes numbered 1-64 arranged sequentially. These nodes are renumbered according to the XYZ axes, and each number is converted into a three-digit 4-ary number. That is, for any node numbered n, it is converted into... In the form of the new numbering rule, node n is defined as... After renumbering the nodes in the XYZ coordinate system, the position of node n can be determined; that is, for node n, the position of node n can be determined. In other words, Indicates node n in the subnet The location inside; This indicates that node n is in the subnet. The x-th row and z-th column of the array; Indicates node n in the subnet The location inside.

3. The dynamic communication path selection and switching method for fully interconnected networks according to claim 2, characterized in that, Step 2 specifically involves: Step 2.1: Calculate the relative differences between the nodes that need to communicate on different coordinate axes; For any node and If node A needs to send a message to node B, then calculate the relative distance difference between node A and B on different coordinate axes, as shown in equation (1). （1）； in, , , These represent the distance differences between nodes A and B along the X, Y, and Z axes, respectively. Step 2.2: Based on the distance difference, determine whether nodes A and B are in the same subnet; like , , If two values ​​are 0, it means that nodes A and B exist in two different subnets of the fully interconnected network. Proceed to step 2.3.

1. like , , If one of the values ​​is 0, it means that nodes A and B exist in at least the same subnet of the fully interconnected network, and proceed to step 2.3.2; like , , If neither is 0, it means that nodes A and B belong to different subnets in the fully interconnected network, and proceed to step 2.3.3; Step 2.3: Based on the subnet determination results, list the possible communication paths between nodes A and B for different scenarios; specifically: Step 2.3.1: If it is determined that nodes A and B exist in two different subnets of the fully interconnected network, then the possible communication paths between nodes A and B are as follows: like This indicates that nodes A and B both exist in the subnet. In this case, nodes A and B can be connected through the subnet. Through one message exchange, the message between them is completed; like This indicates that nodes A and B both exist in the subnet. In this case, nodes A and B can be connected through the subnet. Through one message exchange, the message between them is completed; like This indicates that nodes A and B both exist in the subnet. In this case, nodes A and B can be connected through the subnet. Through one message exchange, the message between them is completed; Step 2.3.2: If nodes A and B exist in at least the same subnet of the fully interconnected network, then the possible communication paths between nodes A and B are as follows: If only This indicates that nodes A and B both exist in the subnet. In the network, nodes A and B can use subnets. After one message exchange, the message exchange between them is completed; If only This indicates that nodes A and B both exist in the subnet. In the network, nodes A and B can use subnets. After one message exchange, the message exchange between them is completed; If only This indicates that nodes A and B both exist in the subnet. In the network, nodes A and B can use subnets. After one message exchange, the message exchange between them is completed; Step 2.3.3: If nodes A and B belong to different subnets of the fully interconnected network, then there are a total of 6 shortest message passing paths between nodes A and B: The message is first relayed through subnet X, and then delivered via subnet Y. In other words, the message delivery path requires the assistance of relay nodes. To achieve this, firstly via subnet Relay message, again via subnet Deliver the message; The message is first relayed through subnet X, and then delivered through subnet Z. In other words, the message delivery path requires the assistance of relay nodes. To achieve this, firstly via subnet Relay message, again via subnet Deliver the message; The message is first relayed through subnet Y, and then delivered through subnet X. In other words, the message delivery path requires the assistance of relay nodes. To achieve this, firstly via subnet Relay message, again via subnet Deliver the message; The message is first relayed through subnet Y, and then delivered through subnet Z. In other words, the message delivery path requires the assistance of relay nodes. To achieve this, firstly via subnet Relay message, again via subnet Deliver the message; The message is first relayed through the Z subnet, and then delivered through the X subnet; that is, the message transmission path requires the use of relay nodes. To achieve this, firstly via subnet Relay message, again via subnet Deliver the message; The message is first relayed through subnet Z, and then delivered through subnet Y. In other words, the message delivery path requires the assistance of relay nodes. To achieve this, firstly via subnet Relay message, again via subnet Deliver the message; Step 2.4: Based on the above analysis, all communication paths based on the relative positions of nodes A and B are obtained.

4. The dynamic communication path selection and switching method for fully interconnected networks according to claim 3, characterized in that, e ( k The calculation formula is shown in equation (2): （2）； In the formula, e ( k The inert decision-making module is represented in the first... k When making the next path selection e Value, that is, the value at the 1st k Next e ( k The probability of ), K p This indicates the degree to which the historical best route stimulates the current pathfinding module. This is a negative feedback adjustment of the pathfinding module to the historical best module; d For hyperparameters, it represents the first... d Secondary decision, and 0 <d<k ; The calculation is shown in formula (3): （3）； In the formula, r The formula for representing the selection preference of the lazy decision-making module is as follows: , C ( k )for k At any given time, the number of times the current optimal path has been selected. T ( k )express k The total number of candidate paths at any given time.

Images