CN112039786B

CN112039786B - Torus network-based broadcasting method

Info

Publication number: CN112039786B
Application number: CN201910482573.5A
Authority: CN
Inventors: 向东; 潘群杨
Original assignee: Tsinghua University
Current assignee: Tsinghua University
Priority date: 2019-06-04
Filing date: 2019-06-04
Publication date: 2021-11-19
Anticipated expiration: 2039-06-04
Also published as: CN112039786A

Abstract

The embodiment of the invention provides a Torus network-based broadcasting method, which comprises the following steps: the Torus network is divided into a plurality of hypercube sub-networks, and for any hypercube sub-network, one node is selected from the hypercube sub-network to serve as an initial receiving point of the hypercube sub-network; sending a broadcast message to the initial receiving point of each hypercube sub-network; for any of the initial receiving points, broadcasting the broadcast message to all nodes in the hypercube sub-network to which the node belongs based on the node. The embodiment of the invention realizes the parallel broadcast of the broadcast message in each hypercube sub-network, thereby greatly reducing the broadcast delay.

Description

Torus network-based broadcasting method

Technical Field

The invention belongs to the technical field of computer networks, and particularly relates to a Torus network-based broadcasting method.

Background

High-performance computers have been widely researched and developed in recent years due to the increasing demand for computing power in various fields. Torus networks are widely used in high performance interconnects, particularly in high dimensional network applications. However, as the network scale is continuously increased, the reliability problem becomes more severe, and it becomes necessary and very challenging to design a high performance interconnection network with fault tolerance.

The transmission of data in a network includes not only the transmission of individual data but also the migration and control operations of global data, which are collectively referred to as aggregate communication. Aggregated communication services mainly include four basic types, which are point-to-point communication, one-to-all communication, many-to-one communication, and many-to-many communication, respectively. Wherein a pair of full communications includes broadcast and dissemination. Broadcast means that the same message is sent from a sender to all receivers in the network. With the widespread use of collective communications in multiple computer systems, efficient broadcasting of data has become a key performance indicator. The broadcast may be implemented by hardware and software, where the hardware has two categories, tree-based algorithms and path-based algorithms.

The software implementation of broadcasting is not as efficient as the hardware implementation, but the advantages of low cost, compatibility with unicast and the like are well developed. The most classical of these is the dimension-by-dimension broadcast algorithm using wildcard tags. The tags of the algorithm include the current dimension broadcast tag and a tag matrix of one other dimension broadcast. The algorithm firstly broadcasts in the current dimension, when the node receives the data packet, the label needing broadcasting in the next dimension is found from the label matrix, and the rest label matrix is reserved. And broadcasting in a recursive mode until all current dimension broadcasting labels are all zero and all label matrixes are empty. The algorithm is simple and convenient, but the number of hops is long, so that the broadcasting delay is high.

Disclosure of Invention

To overcome the problem of high latency of the existing broadcasting methods described above or to at least partially solve the problem, embodiments of the present invention provide a Torus network-based broadcasting method.

According to a first aspect of the embodiments of the present invention, there is provided a Torus network-based broadcasting method, including:

the Torus network is divided into a plurality of hypercube sub-networks, and for any hypercube sub-network, one node is selected from the hypercube sub-network to serve as an initial receiving point of the hypercube sub-network;

sending a broadcast message to the initial receiving point of each hypercube sub-network;

for any of the initial receiving points, broadcasting the broadcast message to all nodes in the hypercube subnetwork to which the initial receiving point belongs based on the initial receiving point.

According to a second aspect of the embodiments of the present invention, there is also provided an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, the processor calling the program instructions to be able to execute the Torus network-based broadcasting method provided in any of the various possible implementations of the first aspect.

According to a third aspect of embodiments of the present invention, there is also provided a non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the Torus network based broadcasting method provided by any of the various possible implementations of the first aspect.

The embodiment of the invention provides a Torus network-based broadcasting method, which comprises the steps of firstly dividing a Torus network into a plurality of hypercube sub-networks, then distributing broadcast messages to a node of each hypercube sub-network, and finally enabling an initial receiving point of the obtained messages to be responsible for message broadcasting of the whole hypercube sub-network, so that the broadcast messages are broadcasted in each hypercube sub-network in parallel, and the broadcasting delay is greatly reduced.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and those skilled in the art can also obtain other drawings according to the drawings without creative efforts.

FIG. 1 is a schematic overall flow chart of a Torus network-based broadcasting method according to an embodiment of the present invention;

FIG. 2 is a schematic diagram of Torus network broadcasting in a Torus network-based broadcasting method according to an embodiment of the present invention;

FIG. 3 is a schematic diagram illustrating the segmentation of the hypercube network of the Torus network in the Torus network-based broadcasting method according to the embodiment of the present invention;

fig. 4 is a schematic multicast diagram of a shortest distance chain in a Torus network-based broadcasting method according to an embodiment of the present invention;

FIG. 5 is a schematic diagram of the broadcasting of the hypercube sub-network in the Torus network-based broadcasting method according to the embodiment of the present invention;

FIG. 6 is a schematic diagram illustrating comparison of average delay of simulation effect of five-dimensional Tours network in the Torus network-based broadcasting method according to the embodiment of the present invention;

fig. 7 is a schematic diagram illustrating a comparison of receivable flows of simulation effects of a five-dimensional Tours network in a Torus network-based broadcasting method according to an embodiment of the present invention;

FIG. 8 is a schematic diagram illustrating comparison of average delay of simulation effect of six-dimensional Tours network in the Torus network-based broadcasting method according to the embodiment of the present invention;

FIG. 9 is a schematic diagram illustrating a comparison of receivable flows of simulation effects of a six-dimensional Tours network in a Torus network-based broadcasting method according to an embodiment of the present invention;

FIG. 10 is a schematic diagram illustrating comparison of average delay of simulation effect of an eight-dimensional Tours network in a Torus network-based broadcasting method according to an embodiment of the present invention;

fig. 11 is a schematic diagram illustrating a comparison of receivable flows of the simulation effect of the eight-dimensional Tours network in the Torus network-based broadcasting method according to the embodiment of the present invention;

FIG. 12 is a schematic diagram illustrating an influence of the number of failures on the mean delay of the simulation effect of the Tours network in the Torus network-based broadcasting method according to the embodiment of the present invention;

fig. 13 is a schematic diagram illustrating an influence of the number of failures on receivable traffic of the Tours network simulation effect in the Torus network-based broadcasting method according to the embodiment of the present invention;

fig. 14 is a schematic view of an overall structure of an electronic device according to an embodiment of the present invention.

Detailed Description

In an embodiment of the present invention, a Torus network based broadcasting method is provided, and fig. 1 is a schematic overall flow chart of the Torus network based broadcasting method provided in the embodiment of the present invention, where the method includes: s101, a Torus network is divided into a plurality of hypercube sub-networks, and for any hypercube sub-network, one node is selected from the hypercube sub-network to serve as an initial receiving point of the hypercube sub-network;

wherein, Torus network is a multidimensional surrounding grid network formed by interconnecting nodes with fixed sizes. The nodes in the Torus network are information interaction units, such as routers. A hypercube network refers to a network in which the number of nodes in each dimension is 2. The Torus network in high dimension is divided into a plurality of hypercube networks with the same dimension as the Torus network. One node from each hypercube network is selected as the initial receiving point. The initial receiving point is responsible for broadcasting the broadcast message in the hypercube subnetwork to which the initial receiving point belongs.

S102, sending a broadcast message to the initial receiving point of each hypercube sub-network;

wherein, the broadcast message is a message to be broadcast. The broadcast message is first distributed to the initial receiving point of each hypercube subnetwork, and the embodiment is not limited to the distribution mode, for example, the distribution mode may be a multicast mode, and the multicast mode may use a shortest distance chain multicast algorithm.

S103, for any initial receiving point, broadcasting the broadcast message to all nodes in the hypercube sub-network to which the initial receiving point belongs based on the initial receiving point.

For any initial receiving point, after receiving the broadcast message, the initial receiving point broadcasts the broadcast message to all nodes in the hypercube sub-network to which the initial receiving point belongs.

Figure 2 is a schematic diagram of a 4 x 4 three-dimensional Torus network broadcast. First, the Torus network was partitioned into 8 hypercube structures as shown by the dashed boxes in the figure. The broadcast message is firstly transmitted to the source node No. 0, and the broadcast message is transmitted to a node of each hypercube through shortest distance chain multicast. First to the shortest distance initial receiving point No. 2.

Nodes

0 and 2 then send broadcast messages to corresponding initial receiving points 32 and 34, respectively. And finally, the nodes No. 0, No. 2, No. 32 and No. 34 transmit the broadcast messages to the nodes No. 8, No. 10, No. 40 and No. 42, and finally, the nodes which receive the broadcast messages in each hypercube sub-network broadcast according to the broadcast labels (1,1, 1).

In the embodiment, the Torus network is firstly divided into a plurality of hypercube sub-networks, then the broadcast message is distributed to one node of each hypercube sub-network, and finally the node which obtains the message is taken as an initial receiving point to be responsible for message broadcast of the whole hypercube sub-network, so that the broadcast message is broadcast in each hypercube sub-network in parallel, and the broadcast delay is greatly reduced.

On the basis of the above embodiment, the step of dividing the Torus network into a plurality of hypercube sub-networks in this embodiment specifically includes: for any dimension in the Torus network, if the difference between the coordinate of any node in the Torus network in the dimension and the coordinate of an initial source node in the Torus network in the dimension is an integral multiple of 2, taking the node as an initial receiving point; wherein the initial receiving point is any node in the Torus network; if the length of the dimension is an integral multiple of 2, setting the direction label of the hypercube network to which the node belongs on the dimension as 1; if the length of the dimension is not an integral multiple of the 2, judging whether the node is adjacent to the initial source node; if the node is not adjacent to the initial source node, setting the direction label of the hypercube network to which the node belongs to 1 in the dimension; if the node is adjacent to the initial source node, setting the direction label of the hypercube network to which the node belongs to 0 in the dimension; and representing the hypercube network to which the node belongs by using the direction labels of the node and the hypercube network to which the node belongs in all dimensions.

Specifically, when segmenting the Torus network, the dimension of the segmented hypercube network and the Torus network are as large as possibleAre the same, and try to make all dimensions in any hypercube sub-network the same in length, i.e. the number of nodes in each dimension is the same. For example, the dimensions of the partitioned hypercube network and the dimensions of the Torus network are three-dimensional, and the number of nodes in each dimension of any hypercube network is 4. Using the ith hypercube sub-network as the initial receiving point D in the hypercube sub-network_iAnd the orientation label L of the hypercube network in each dimension_i(H₀,H₁,…,H_i,…,H_n-1) I.e. the ith hypercube subnetwork is denoted N_i(D_i，L_i(H₀,H₁,…,H_i,…,H_n-1) Where n is the dimension of the Torus network. Suppose each node in any hypercube sub-network is formed by moving the source node in that hypercube sub-network in the positive direction of each dimension, i.e. the default direction label of the hypercube sub-network is L (1,1, …,1), and the set of all hypercube sub-networks is N.

If the number of nodes in each dimension of the Torus network is even, the Torus network can be divided into a plurality of hypercube sub-networks with the same dimension as the Torus network. For example, a 5-dimensional Torus network of 6 x 6 can be segmented into 3 x 3 super cubic sub-networks of 243 total dimensions. But a hypercube network below the dimensionality of the Torus network would appear if the number of nodes in any dimension of the Torus network is odd. Figure 3 is a schematic of the segmentation of a 5 x 5 two-dimensional Torus network. It can be seen that the number of nodes is odd for both dimensions. The network cannot be completely partitioned into two-dimensional hypercube subnetworks. These include two-dimensional sub-hypercube networks (0,1,5,6), (2,3,7,8), (10,11,15,16) and (12,13,17, 18). One-dimensional hypercube networks (4,9), (14,19), (20,21) and (22, 23). There is also a single node 24. The representation of the hypercube sub-network is represented using the initial receiving point and directional label combination. The direction 1 in the direction label is a positive direction in a certain dimension, the direction-1 is a negative direction, and 0 is not included. If the initial receiving point in the hypercube sub-network (0,1,5,6) is selected to be 0, the directional label of the hypercube sub-network is (1,1), and if the initial receiving point is selected to be 5, the directional label is (1, -1). Also, the hypercube network (20,21) selects the direction label to be (1,0) when the initial receiving point is 20.

The steps for segmenting the Torus network are as follows:

d1, sequentially executing the following operations for each dimension from i to n, wherein the operations are as follows:

1) if the number of nodes in the ith dimension is an even number, j is 1 to k, and k is the total number of nodes in the ith dimension, the following operations are carried out on the nodes in the ith dimension:

a) if the coordinate of the node in the ith dimension is different from the coordinate of the initial source node in the ith dimension by an integer multiple of 2, adding the node and the default direction label to H.

2) If the number of nodes in the ith dimension is odd, from j to 1 to k, the following operations are performed on the nodes in the ith dimension:

a) if the coordinate of the node in the ith dimension is different from the coordinate of the initial source node in the ith dimension by an integer multiple of 2, and the node is not adjacent to the initial source node, then the node and the default direction label are added to H.

b) If the difference between the coordinate of the node in the ith dimension and the coordinate of the initial source node in the ith dimension is an integer multiple of 2, and the node is adjacent to the initial source node, the direction label in the dimension in the default direction labels is set to be 0, which represents that the dimension does not need to be broadcast. This node and the modified default direction label are added to H.

D2: through the operation of the step D1, the initial receiving points and direction labels of all the hypercube sub-networks are obtained.

On the basis of the foregoing embodiment, in this embodiment, the step of sending a broadcast message to the initial receiving point of each hypercube subnetwork specifically includes: sending a broadcast message to one of the initial receiving points of all the hypercube sub-networks, taking the initial receiving point as a target initial receiving point, and taking other initial receiving points except the target initial receiving point in the initial receiving points of all the hypercube sub-networks as terminal nodes; generating a shortest distance chain according to the weighted jump step number between the target initial receiving point and each terminal node and the weighted jump step number between all the terminal nodes; and according to the shortest distance chain, sending the broadcast message to initial receiving points of all the hypercube sub-networks by a multicast method based on a multicast tree.

Specifically, one initial receiving point is selected from the initial receiving points of all the hypercube subnetworks as a target initial receiving point, the target initial receiving point firstly receives the broadcast message, and then the broadcast message is multicast to the initial receiving points of other hypercube subnetworks. Taking the initial receiving points of other hypercube sub-networks as the final nodes, and forming a chain C by the target initial receiving point and all the final nodes: { D₀，D₁，…，D_mIn which D is₀For the target initial reception point, D₁，…，D_mM +1 is the total number of hypercube nets as a terminal node. And generating a shortest distance chain C' according to the weighted hop step number between the target initial receiving point and each terminal node and the weighted hop step number between all the terminal nodes. Firstly, placing a target initial receiving point at the head of a shortest distance chain; and for the target initial receiving point or any terminal node placed in the shortest distance chain, selecting the terminal node with the minimum weighted hop count from all terminal nodes not placed in the shortest distance chain, and inserting the selected terminal node behind the target initial receiving point or the terminal node and in close proximity to the target initial receiving point or the terminal node until all terminal nodes are placed in the shortest distance chain.

The steps for generating the shortest distance chain are as follows:

and D1, initializing. Target initial receiving point D₀Placed in the set seed of the shortest distance nodes. Removing the target initial reception point D from the chain C₀Adding a target initial receiving point D in the shortest distance chain C₀。

D2, calculating a distance chain

E1, initializing end flag bit Finish ═ 0;

e2, if the flag bit is 0, performing the following operations:

1) initializing the Tempseed to null;

2) from i to 0 to the number of nodes in the seed set, the following operations are performed in sequence:

a) finding the point Cmin with the minimum distance seed [ i ] weighting step number in the chain C;

b) deleting Cmin from chain C;

c) adding Cmin to the Tempseed;

d) the node seed [ i ] is found in C' and the node Cmin is inserted after it.

e) If the chain C is empty, the end flag Finish is set to 1.

3) Adding nodes in the Tempseed into the seed, namely seed + Tempseed;

d3, the shortest distance chain C' can be obtained through D2, and the algorithm is ended.

Multicast based on multicast tree is used for the formed shortest distance chain C'. Assume that multicast includes a total of m-1 nodes. m is the total number of initial reception points. D is used for all initial receiving points according to the sequence of each initial receiving point in the shortest distance chain C₀,d₁,…d_m-2,d_m-1Is shown in which d is₀Is a target source point, d₁,…d_m-2,d_m-1Is the end node. At first only d₀Receiving a broadcast message, the first step is to send the message to an intermediate node d_m/2，d_m/2The broadcast message is duplicated. After d₀And d_m/2Respectively sending broadcast messages to two other nodes d_m/4And d_3m/2. Repeating the above steps until all the initial receiving points obtain the broadcast message, and seeing that if there are m-1 initial receiving points in multicast, it needs to be

And (5) carrying out the steps.

The target initial receiving point in this embodiment needs to multicast a broadcast message to the initial receiving point of each hypercube subnetwork. The shortest distance chain multicast is used, and the algorithm belongs to a greedy algorithm, so that each route in the multicast only needs minimum steps. Fig. 4 is a multicast diagram of a shortest distance chain. The wrap around connection is not drawn here. Wherein, the node 0 is a target source point, and the black dot is a terminal node of the multicast. Firstly, a point 100 node closest to a node 0 is found in the terminal nodes, 1 jump is needed to send out the broadcast message, and then the node 0 and the node 100 send out the broadcast message to the terminal nodes 26 and 106 closest to each other. The last nodes No. 0, 100, 26 and 106 are sent to the nearest nodes No. 2, 110, 37 and 87, respectively.

On the basis of the foregoing embodiment, in this embodiment, for any initial receiving point, the step of broadcasting the broadcast message to all nodes in the hypercube subnetwork to which the initial receiving point belongs based on the initial receiving point specifically includes: setting a broadcast label for the initial receiving point; if the initial receiving point needs to send a message in any dimension, the broadcast tag of the initial receiving point in the dimension is set to be 1, and if the initial receiving point does not need to send a message in any dimension, the broadcast tag of the initial receiving point in the dimension is set to be 0; acquiring all dimensions of the broadcast label of the initial receiving point as 1 according to the broadcast label; for any dimension of which the broadcast label of the initial receiving point is 1, if the channel of the dimension is idle, acquiring the cache number of the channel of the dimension; and sending the broadcast message along the dimension corresponding to the channel with the largest cache number based on the initial receiving point, and modifying the broadcast label of the initial receiving point on the dimension to be 0 until all 1 in the broadcast label are set to be 0.

Specifically, the initial point of reception in each hypercube subnetwork at which the broadcast message is obtained is broadcast using the broadcast tag of that hypercube subnetwork. First, a broadcast tag, namely B (B), is set for the initial receiving point in each hypercube subnetwork₀,B₁,...,B_i,…,B_n-1) N denotes the dimensionality of the Torus network, the ith label value B in B_iA value of 1 indicates that the ith dimension needs to send a broadcast message, and a value of 0 indicates that no broadcast message needs to be sent. The algorithm recursively sends broadcast messages per broadcast tag until all values in each broadcast tag become 0. The specific algorithm is described as follows:

s1, let S be the initial receiving point of any hypercube subnetwork, S be the initial broadcast tag (1,1 … 1,1), and the buffer in each broadcast tag channel direction is marked as C (0, 0 … 0, 0).

S2, from i to n, for one time:

if B [ i ] is equal to 1 and the channel of the ith dimension is idle, assigning the cache number of the ith dimension to C [ i ];

and S3, finding the dimension corresponding to the channel with the maximum buffer number in the C, sending the broadcast message along the dimension, and modifying the broadcast label value of the dimension to 0.

On the basis of the above embodiment, the nodes in the Torus network in this embodiment include a normal node, a failed node, and a failed connection node; correspondingly, for any one of the hypercube sub-networks, the step of selecting a node from the hypercube sub-network as an initial receiving point of the hypercube sub-network specifically comprises: for any one hypercube sub-network, selecting a normal node from the hypercube sub-network as an initial receiving point of the hypercube sub-network; correspondingly, the step of segmenting the Torus network into a plurality of hypercube sub-networks specifically comprises: for any dimension in the Torus network, if the difference between the coordinate of any node in the Torus network in the dimension and the coordinate of the initial source node in the dimension is an integral multiple of 2, and the node is a normal node, taking the node as an initial receiving point; wherein the initial source node is any node in the Torus network; for any dimension in the Torus network, if the difference between the coordinate of any node in the Torus network in the dimension and the coordinate of the initial source node in the dimension is an integral multiple of 2, and the node is a failed node, acquiring a normal node which is directly adjacent to the node in any dimension, taking the adjacent normal node as an initial receiving point, and setting a direction label of a hypercube network to which the node belongs in the dimension corresponding to the adjacent normal node as-1; wherein the coordinate of the adjacent normal node in any dimension is larger than the coordinate of the node in the dimension; if the length of the dimension is an integral multiple of 2 and the node is a normal node, setting the direction label of the hypercube sub-network to which the node belongs on the dimension as 1; if the length of the dimension is not an integral multiple of 2, judging whether the node is adjacent to the initial source node; if the node is adjacent to the initial source node, setting the direction label of the hypercube network to which the node belongs to 0 in the dimension; and representing the hypercube network to which the initial receiving point belongs by using the direction labels of the node and the hypercube network to which the node belongs in all dimensions.

Specifically, the present embodiment is a fault-tolerant broadcast algorithm that handles failures. The normal nodes are nodes capable of routing normally, the failure nodes are nodes incapable of routing normally, and the nodes on the failure connection are nodes with disconnected connection line segments. In selecting the initial receiving point of each hypercube sub-network, the default initial receiving point is selected first as in the fault-free broadcast algorithm. When the point is found to be a fault node or a failure connection node near a node by broadcasting, acquiring a normal node adjacent to the node in the positive direction of any dimension, taking the adjacent normal node as an initial receiving point, sending a message to the node, and setting a direction label of a hypercube network to which the node belongs in the dimension corresponding to the adjacent normal node as-1;

in this embodiment, for any of the initial receiving points, the step of broadcasting the broadcast message to all nodes in the hypercube subnetwork to which the initial receiving point belongs based on the initial receiving point specifically includes: for any node in each hypercube sub-network, determining the safety information of the node according to the condition that the adjacent node of the node is an invalid node or a normal node; and for any initial receiving point, broadcasting the broadcast message to all nodes in the hypercube sub-network to which the initial receiving point belongs according to the security information of all nodes in the hypercube sub-network to which the initial receiving point belongs.

Specifically, the present embodiment is a fault-tolerant broadcast method, which uses the security information of the Torus network to guide the fault-tolerant broadcast of broadcast messages in the hypercube sub-network. Each node in the Torus network may consist of 2ⁿCovering all adjacent nodes by n-dimensional sub-cubes, storing the security information of the sub-network of the hypercube to which all adjacent nodes of each node belong, and using 2n x 2ⁿIs represented by a matrix of (a). Root of herbaceous plantAnd determining the safety information of the node according to the safety information corresponding to any node. And broadcasting the broadcast message to all nodes in each hypercube sub-network according to the safety information of all nodes in each hypercube sub-network, thereby realizing the fault-tolerant broadcast of the Torus network.

On the basis of the foregoing embodiment, in this embodiment, for any node in each hypercube sub-network, the step of determining the security information of the node according to whether the neighboring node of the node is an invalid node or a normal node specifically includes: for any node in each hypercube sub-network, if the single connection from the node to the adjacent node of the node in the hypercube sub-network to which the node belongs is invalid, the node is taken as a level 1 unsafe node; if at least 2 adjacent nodes of the node are level 1 unsafe nodes or failure nodes, taking the node as a level 2 unsafe node; if at least 3 adjacent nodes of the node are level 1 unsafe nodes, failure nodes or level 2 unsafe nodes, taking the node as the level 2 unsafe node; and if the node is not the level 1 unsafe node or the level 2 unsafe node, the node is taken as a local safe node.

Specifically, for example, the first preset number is 2, the second preset number is 3, and the level 2 insecure nodes can be classified into ordinary insecure nodes and strong insecure nodes. Wherein ordinary insecure means that the node has a secure neighbor, and strong insecure means that all neighbors of the node are insecure nodes.

On the basis of the foregoing embodiment, in this embodiment, for any initial receiving point, according to the security information of all nodes in the hypercube sub-network to which the initial receiving point belongs, the step of broadcasting the broadcast message to all nodes in the hypercube sub-network to which the initial receiving point belongs specifically includes: if the number of the failed neighbor nodes of the initial receiving point is not more than 1, executing the following steps S1-S2: s1, for any dimension of which the broadcast label of the initial receiving point is 1, if the initial receiving point is a local security node, sending the broadcast message along the dimension based on the initial receiving point, and modifying the broadcast label of the initial receiving point on the dimension to 0; s2, if the initial receiving point is a level 2 unsafe node and the initial receiving point has at most 1 adjacent node as a failure node, then based on the initial receiving point, the broadcast message is sent along the dimension, and the broadcast label of the initial receiving point on the dimension is modified to 0; if the number of the failed neighbor nodes of the initial receiving point is more than 1, when each step of the steps S1-S2 is executed, before modifying the broadcast label of the initial receiving point in the dimension, it is determined whether the dimension is the last dimension in which a message needs to be sent, and if not, the broadcast label of the initial receiving point in the dimension is modified to 0; if yes, the broadcast tag of the initial receiving point in the dimension is kept to be 1.

Specifically, the two most important principles when guiding broadcasting using the security information of the Torus network are: principle 1, avoiding sending broadcast messages to a certain node with two failed adjacent nodes as much as possible; in principle 2, if the initial receiving point has more than or equal to 2 failed neighbor nodes, the broadcast tag is not reset when sending to the last dimension that can be sent. The specific steps are described as follows:

s1, if any initial receiving point is S and the label is (1,1 … 1,1), if the number of the failed adjacent nodes of S is not more than 1, the operation is carried out according to S2-S4. Otherwise, operation is performed according to S5

S2，For i＝1 to n

a) If B [ i ] is 1 and s is a local security node then (B) is performed;

b) b [ i ] is set to 0 and the broadcast message is sent along the i dimension.

S3，For i＝1 to n

a) If B [ i ] is 1 and s is a level 2 very insecure node then (B) is performed;

b) if s has at most 1 failed neighbor node, performing (c), wherein the third preset number is 1;

c) b [ i ] is set to 0 and the message is sent along the i dimension.

S4，For i＝1 to n

a) If B [ i ] is 1, (B) is performed;

b) b [ i ] is set to 0 and the message is sent along the i dimension.

S5, steps S2-S4 are carried out, but each time whether the dimension is the last dimension which can send the message is checked, if not, B [ i ] is set to be 0; if so, then B [ i ] remains 1, which can continue to send broadcast messages to the failed nodes of the dimension.

FIG. 5 is an example of a four-dimensional hypercube subnetwork broadcast. Two circles represent the initial reception points, the black dots represent the failed nodes, and one circle represents the normal nodes. The initial reception point is (1101) and the broadcast tag is [1111 ]. First, the initial receiving point sends a broadcast message to the security node (0101), and the initial receiving point and the broadcast tag of the point are updated to [0111 ]. Then node (0101) is responsible for the broadcast of the left 3-dimensional hypercube and the initial receiving point is responsible for the broadcast of the right three-dimensional hypercube. The left side node (0101) would then send the broadcast tag [0110] to node (0100), the broadcast tag [0100] to node (0111), and the broadcast tag [0000] to node (0001). Recursion is done until the broadcast tags for all nodes become [0000] to complete the broadcast of the left 3-dimensional hypercube. After the initial receiving point becomes [0111], the last node that can be broadcasted is (1111), since the initial receiving point has 2 failed neighbor stages, namely (1100) and (1001). So a message is sent to (1111) but the broadcast tag at this point is not updated, still [0111 ]. Then the node (1111) sends the broadcast message to (1011), the node (1011) sends the broadcast message to (1010), and the node (1010) sends the broadcast message to (1000), thereby completing the broadcast of the whole right three-dimensional hypercube.

The broadcasting algorithm provided by the embodiment of the invention is subjected to simulation test on a microchip-level simulator. In simulations, embodiments of the present invention used VCT switching technology for Torus networks. In a simulated Torus network, each unicast routing packet is set to include 10 flits. Simulations were performed in three different high dimensional torus networks, including: a five-dimensional network of 6 × 6 × 6 × 6, a six-dimensional network of 2 × 2 × 3 × 6 × 8 × 8, and an eight-dimensional network of 3 × 3 × 3 × 3 × 5 × 5 × 4. The uniform transmission mode was chosen in the simulations of three different networks. The source node is randomly generated in all nodes with the same probability

The performance of the following three broadcast algorithms were compared:

a) the basic routing algorithm adopts a non-fault-tolerant classical bubble flow control routing algorithm. The broadcast algorithm employs the widdcard's dimension-by-dimension broadcast algorithm. Here denoted by Wildcard.

b) The basic routing algorithm adopts a non-fault-tolerant flow control routing algorithm. The broadcast algorithm adopts the broadcast algorithm of the embodiment of the invention. Here denoted FC.

b) The basic routing algorithm adopts a fault-tolerant flow control routing algorithm. The broadcast algorithm adopts the broadcast algorithm of the embodiment of the invention. Here denoted FTFC.

Fig. 6 and 7 are simulation results in a 6 × 6 × 6 × 6 five-dimensional ours network. In the simulation, 25 failed nodes and 25 failed connections are randomly added in the multicast of the FTFC routing algorithm. The indicators compared include average delay and acceptable flow. Where the average delay is expressed in cycles. Acceptable traffic is represented by packets received in every 1000 cycles. It can be seen from fig. 6 and 7 that the broadcast algorithms FC and FTFC using embodiments of the present invention are far superior to the Wildcard broadcast algorithm in both average delay and acceptable traffic. But the FC of non-fault tolerant broadcasts is slightly better than the FTFC of fault tolerant broadcasts.

Fig. 8 and 9 are simulation results in a 2 × 2 × 3 × 6 × 8 × 8 six-dimensional ours network. In the simulation, 25 failed nodes and 25 failed connections are randomly added in the multicast adopting the FTFC routing algorithm. It can be seen from the figure that the broadcast algorithms FC and FTFC using embodiments of the present invention are much lower in average delay than the Wildcard broadcast algorithm. But also later in terms of acceptable flow to saturation. In the comparison of FC of non-fault tolerant broadcasts versus FTFC of fault tolerant broadcasts, a non-faulty network is also superior to a faulty fault tolerant network.

Fig. 10 and 11 are simulation results in a 3 × 3 × 3 × 3 × 5 × 5 × 4 eight-dimensional ours network. In the simulation, 100 failed nodes and 100 failed connections are randomly added in the multicast adopting the FTFC routing algorithm. It can be seen from the figure that the broadcast algorithms FC and FTFC using embodiments of the present invention are much lower in average delay than the Wildcard broadcast algorithm. And saturation is reached later in terms of acceptable flow. In the comparison of the FC of non-fault tolerant broadcasts to the FTFC of fault tolerant broadcasts, the performance of the non-faulty network and the faulty fault-tolerant network are very close.

Fig. 12 and 13 are performance comparisons for injecting different numbers of failures in a 3 x 5 x 4 network. The three algorithms were compared under a homogeneous model. The injection rate was fixed at 12.000294packets/1000 cycles. It can be seen from fig. 12 that FTFC and FC are very close and much better than the Wildcard algorithm in terms of average delay with a number of failures less than or equal to 600. However, if the number of failures is greater than 600, the FTFC becomes sensitive to the number of failures. The delay increases rapidly and the wildcard algorithm starts to outperform the FTFC algorithm when the number of failures reaches 1000. With the number of failures equal to or less than 600 as shown in fig. 13 at acceptable flow rates, FTFC and FC are very close and far superior to the Wildcard algorithm. However, if the number of failures is greater than 600, the acceptable flow rate of the FTFC rapidly decreases. This is because as the number of failures increases, the saturation point of the FTFC comes earlier, resulting in a drop in the received traffic.

The embodiment provides an electronic device, and fig. 14 is a schematic view of an overall structure of the electronic device according to the embodiment of the present invention, where the electronic device includes: at least one processor 141, at least one memory 142, and a bus 143; wherein the content of the first and second substances,

processor 141 and memory 142 communicate with each other via bus 143;

the memory 142 stores program instructions executable by the processor 141, and the processor calls the program instructions to perform the methods provided by the above method embodiments, for example, the method includes: the Torus network is divided into a plurality of hypercube sub-networks, and for any hypercube sub-network, one node is selected from the hypercube sub-network to serve as an initial receiving point of the hypercube sub-network; sending a broadcast message to the initial receiving point of each hypercube sub-network; for any of the initial receiving points, broadcasting the broadcast message to all nodes in the hypercube subnetwork to which the initial receiving point belongs based on the initial receiving point.

The present embodiments provide a non-transitory computer-readable storage medium storing computer instructions that cause a computer to perform the methods provided by the above method embodiments, for example, including: the Torus network is divided into a plurality of hypercube sub-networks, and for any hypercube sub-network, one node is selected from the hypercube sub-network to serve as an initial receiving point of the hypercube sub-network; sending a broadcast message to the initial receiving point of each hypercube sub-network; for any of the initial receiving points, broadcasting the broadcast message to all nodes in the hypercube subnetwork to which the initial receiving point belongs based on the initial receiving point.

Those of ordinary skill in the art will understand that: all or part of the steps for implementing the method embodiments may be implemented by hardware related to program instructions, and the program may be stored in a computer readable storage medium, and when executed, the program performs the steps including the method embodiments; and the aforementioned storage medium includes: various media that can store program codes, such as ROM, RAM, magnetic or optical disks.

The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.

Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware. With this understanding in mind, the above-described technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments.

Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.

Claims

1. A Torus network based broadcast method, comprising:

for any of the initial receiving points, broadcasting the broadcast message to all nodes in the hypercube sub-network to which the node belongs based on the node.

2. The Torus network based broadcasting method of claim 1, wherein the step of segmenting the Torus network into a plurality of hypercube subnetworks specifically comprises:

for any dimension in the Torus network, if the difference between the coordinate of any node in the Torus network in the dimension and the coordinate of an initial source node in the Torus network in the dimension is an integral multiple of 2, taking the node as an initial receiving point; wherein the initial source node is any node in the Torus network;

if the length of the dimension is an integral multiple of 2, setting the direction label of the hypercube network to which the node belongs on the dimension to be 1;

if the length of the dimension is not an integral multiple of 2, judging whether the node is adjacent to the initial source node;

if the node is not adjacent to the initial source node, setting the direction label of the hypercube network to which the node belongs to 1 in the dimension;

if the node is adjacent to the initial source node, setting the direction label of the hypercube network to which the node belongs to 0 in the dimension;

and representing the hypercube network to which the node belongs by using the direction labels of the node and the hypercube network to which the node belongs in all dimensions.

3. The Torus network-based broadcast method of claim 1, wherein the step of sending a broadcast message to the initial receiving point of each hypercube subnetwork comprises:

sending a broadcast message to one initial receiving point in all the initial receiving points of the hypercube sub-network, taking the initial receiving point as a target initial receiving point, and taking other initial receiving points except the target initial receiving point in all the initial receiving points of the hypercube sub-network as terminal nodes;

generating a shortest distance chain according to the weighted jump step number between the target initial receiving point and each terminal node and the weighted jump step number between all the terminal nodes;

and according to the shortest distance chain, sending the broadcast message to initial receiving points of all the hypercube sub-networks by a multicast method based on a multicast tree.

4. The Torus network based broadcasting method of claim 1, wherein for any of the initial receiving points, the step of broadcasting the broadcast message to all nodes in the hypercube subnetwork to which the initial receiving point belongs based on the initial receiving point specifically comprises:

setting a broadcast label for the initial receiving point; if the initial receiving point needs to send a message in any dimension, the broadcast tag of the initial receiving point in the dimension is set to be 1, and if the initial receiving point does not need to send a message in any dimension, the broadcast tag of the initial receiving point in the dimension is set to be 0;

acquiring all dimensions of the broadcast label of the initial receiving point as 1 according to the broadcast label;

for any dimension of which the broadcast label of the initial receiving point is 1, if the channel of the dimension is idle, acquiring the cache number of the channel of the dimension;

and sending the broadcast message along the dimension corresponding to the channel with the largest cache number based on the initial receiving point, and modifying the broadcast label of the initial receiving point on the dimension to be 0 until all 1 in the broadcast label are set to be 0.

5. The Torus network based broadcasting method of claim 1, wherein the nodes in the Torus network include regular nodes, failed nodes and nodes on failed connections; the node on the failure connection is a node with a disconnected connecting line segment, and the failure node is a node which cannot be normally routed;

correspondingly, the step of sending the broadcast message to the initial receiving point of each hypercube subnetwork specifically comprises:

forming a shortest distance chain according to the initial receiving points of the selected hypercube sub-networks, and then sending the broadcast message to the initial receiving points of all the hypercube sub-networks according to a multicast method based on a multicast tree; if the initial receiving point of any hypercube sub-network is a failure node or a node in failure connection, when the broadcast message reaches the vicinity of the initial receiving point, acquiring a normal node adjacent to the initial receiving point in the positive direction of any dimension, taking the adjacent normal node as the initial receiving point, sending the message to the adjacent normal node, and setting a direction label of the original hypercube sub-network to which the initial receiving point belongs in the dimension corresponding to the adjacent normal node as-1; the direction labels and the initial receiving points are used for representing the hypercube sub-network to which the initial receiving points belong as a whole, the direction 1 in the direction labels is a positive direction in any dimension, the direction-1 is a negative direction, and the direction 0 is not included.

6. The Torus network based broadcasting method of claim 5, wherein for any of the initial receiving points, the step of broadcasting the broadcast message to all nodes in the hypercube subnetwork to which the initial receiving point belongs based on the initial receiving point specifically comprises:

for any node in each hypercube sub-network, determining the safety information of the node according to the condition that the adjacent node of the node is a failure node or a normal node;

and for any initial receiving point, broadcasting the broadcast message to all nodes in the hypercube sub-network to which the initial receiving point belongs according to the security information of all nodes in the hypercube sub-network to which the initial receiving point belongs.

7. The Torus network-based broadcasting method of claim 6, wherein the step of determining the security information of any node in each hypercube sub-network for the failed node or the normal node based on its neighboring nodes specifically comprises:

for any node in each hypercube sub-network, if the single connection from the node to the adjacent node of the node in the hypercube sub-network to which the node belongs is invalid, the node is taken as a level 1 unsafe node;

if at least 2 adjacent nodes of the node are level 1 unsafe nodes or failure nodes, taking the node as a level 2 unsafe node;

if at least 3 adjacent nodes of the node are level 1 unsafe nodes, failure nodes or level 2 unsafe nodes, taking the node as the level 2 unsafe node;

and if the node is not the level 1 unsafe node or the level 2 unsafe node, the node is taken as a local safe node.

8. The Torus network based broadcasting method of claim 7, wherein the step of broadcasting the broadcast message to all nodes in the hypercube subnetwork to which the initial receiving point belongs according to the security information of all nodes in the hypercube subnetwork to which the initial receiving point belongs specifically comprises:

if the number of failed nodes adjacent to the initial receiving point is not more than 1, executing the following steps S1-S2:

s1, for any dimension of which the broadcast label of the initial receiving point is 1, if the initial receiving point is a local security node, sending the broadcast message along the dimension based on the initial receiving point, and modifying the broadcast label of the initial receiving point on the dimension to 0;

s2, if the initial receiving point is a level 2 unsafe node and the initial receiving point has at most 1 adjacent node as a failure node, then based on the initial receiving point, the broadcast message is sent along the dimension, and the broadcast label of the initial receiving point on the dimension is modified to 0;

if the number of the failed nodes adjacent to the initial receiving point is more than 1, when each of the steps S1-S2 is executed, before modifying the broadcast tag of the initial receiving point in the dimension, it is determined whether the dimension is the last dimension in which a message needs to be sent, and if not, the broadcast tag of the initial receiving point in the dimension is modified to 0; if yes, the broadcast tag of the initial receiving point in the dimension is kept to be 1.

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor when executing the program performs the steps of the Torus network based broadcast method of any of claims 1 to 8.

10. A non-transitory computer readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, performs the steps of the Torus network based broadcasting method of any of claims 1 to 8.