Adaptive dynamic routing method and system for NoC (Network on Chip)

Abstract

The invention discloses a self-adaptive dynamic routing method and system for an NoC (Network on Chip), and relates to the technical field of routing management. Comprising the steps that the local congestion degree of each routing node is collected through a communication link, the global congestion degree is generated through a weighted average mechanism, each routing node integrates delay counters in the east direction, the west direction, the south direction and the north direction according to a local router, the delay counters are used for monitoring the time interval when adjacent data packets leave a buffer area in real time, and the global congestion degree is generated; taking a time interval of leaving a buffer area of adjacent data packets as output delay, spreading the output delay to neighbor routing nodes, calculating local congestion degree and global congestion degree of the routing nodes, generating all possible shortest paths and a candidate direction set corresponding to output directions according to addresses of a source routing node and a destination routing node, and sending the candidate direction set to the destination routing node; screening an optimal path through the local congestion degree and the global congestion degree in sequence; and when a plurality of equivalent paths exist, the routing direction is randomly selected to balance the network load.

Description

TECHNICAL FIELD

[0001] The application discloses an adaptive dynamic routing method and system of a NoC on-chip network and relates to the technical field of routing management. BACKGROUND

[0002] With the continuous progress of semiconductor technology, the number of processor cores integrated in a system on chip (SoC) is increasing, and the traditional bus interconnection architecture cannot meet the requirements of multi-core systems for communication bandwidth and delay. As a new on-chip interconnection architecture, a network on chip (NoC) provides a scalable communication solution for multi-core systems by introducing packet switching and routing mechanisms. In the NoC architecture, the performance of the routing algorithm directly affects the delay, throughput and reliability of the system. Existing NoC routing algorithms are mainly divided into two categories: deterministic routing and adaptive routing. Although the XY routing algorithm is simple to implement, it cannot dynamically adjust the routing path according to the network conditions, which can easily lead to network congestion. Although the Odd-Even routing algorithm can adjust the routing path according to the local congestion conditions, it lacks the ability to perceive the global network state, making it difficult to make optimal routing decisions. SUMMARY

[0003] The application provides a NoC on-chip network adaptive dynamic routing method and system, based on a local and global perception dynamic routing algorithm and communication architecture, to improve the communication efficiency and congestion management capability of a multi-core processor system.

[0004] The specific scheme provided by the application is as follows:

[0005] The application provides a NoC on-chip network adaptive dynamic routing method, comprising:

[0006] The local congestion degree of each routing node is collected through the communication link, and the global congestion degree is generated through a weighted average mechanism, wherein each routing node calculates the local congestion degree and the global congestion degree according to the delay counters integrated in the east, west, south and north directions of the local router, uses the delay counters to monitor the time interval of the adjacent data packets leaving the buffer in real time, takes the time interval of the adjacent data packets leaving the buffer as the output delay, and propagates to the neighbor routing node, and at the same time, calculates the local congestion degree and the global congestion degree,

[0007] According to the source routing node and the destination routing node address, a candidate direction set of all possible shortest paths and corresponding output directions is generated, and the optimal path is filtered through the local congestion degree and the global congestion degree in sequence; when there are multiple equivalent paths, a routing direction is randomly selected to balance the network load.

[0008] Further, the local congestion degree of the routing node and the global congestion degree in the adaptive dynamic routing method of the NoC on-chip network include:

[0009] According to the output delay, the average delay is calculated by a sliding window, the local congestion degree of the routing node is output, and the global congestion degree is calculated according to the local congestion degree and the neighbor routing node congestion degree.

[0010] Further, the adaptive dynamic routing method of the NoC on-chip network utilizes the formula:

[0011]

[0012] The global congestion degree is calculated, wherein C local (i,j) is the local congestion degree of the routing node (i,j), C nighbor (k) is the neighbor routing node congestion degree, N(i,j) is the neighbor routing node set, and the directions are east, west, south and north respectively, C final (i,j) is the global congestion degree related to the routing node, and a is a weight factor.

[0013] Further, the adaptive dynamic routing method of the NoC on-chip network includes:

[0014] The local congestion degree is preferentially screened: the local congestion degree C local (d) of each direction is calculated, and a direction subset with the minimum congestion degree is selected, and the formula is represented as:

[0015] D min_local ={d|C local (d)=minC local (D))}

[0016] If |D min_local | = 1, the direction is directly selected, otherwise the global congestion degree is screened,

[0017] When the global congestion degree is screened, the global congestion degree C min_local (d) of each direction in D global is calculated, a direction subset D min_global with the minimum global congestion degree is selected, if |D min_global | = 1, the direction is selected, otherwise when there are multiple equivalent paths, a routing direction is randomly selected to balance the load.

[0018] The application also provides an adaptive dynamic routing system of the NoC on-chip network, which includes a weighted fusion module and a routing decision module,

[0019] The weighting fusion module collects local congestion degrees of each routing node through a communication link, and generates a global congestion degree through a weighted average mechanism, wherein each routing node integrates delay counters of east, west, south and north four directions according to a local router, the weighting fusion module utilizes the delay counters to monitor time intervals of adjacent data packets leaving a buffer in real time, takes the time intervals of the adjacent data packets leaving the buffer as output delays, and propagates to neighbor routing nodes, while calculating the local congestion degree of the routing node and the global congestion degree,

[0020] The routing decision module generates all possible shortest paths and a candidate direction set of corresponding output directions according to addresses of a source routing node and a destination routing node, and filters optimal paths through the local congestion degree and the global congestion degree in sequence; when there are multiple equivalent paths, a routing direction is randomly selected to balance network load.

[0021] Further, the weighting fusion module in the adaptive dynamic routing system of the NoC on-chip network calculates the local congestion degree of the routing node and the global congestion degree, and includes:

[0022] According to the output delay, an average delay is calculated through a sliding window, the local congestion degree of the routing node is output, and the global congestion degree is calculated according to the local congestion degree and the congestion degree of the neighbor routing node.

[0023] Further, the weighting fusion module in the adaptive dynamic routing system of the NoC on-chip network utilizes a formula:

[0024]

[0025] to calculate the global congestion degree, wherein C local (i,j) is the local congestion degree of the routing node (i,j), C nighbor (k) is the congestion degree of the neighbor routing node, N(i,j) is a neighbor routing node set, and directions are east, west, south and north respectively, C final (i,j) is the global congestion degree related to the routing node, and a is a weight factor.

[0026] Further, the routing decision module in the adaptive dynamic routing system of the NoC on-chip network filters the optimal paths through the local congestion degree and the global congestion degree in sequence, and includes:

[0027] The local congestion degree C local (d) of each direction is calculated preferentially, a direction subset with the minimum congestion degree is selected, and a formula is represented as:

[0028] D min_local ={d|C local (d)=minC local (D))}

[0029] If |D min_local |=1, then directly select this direction, otherwise perform global congestion screening.

[0030] When filtering global congestion, calculate D min_local The global congestion C in each direction global (d) Select the direction subset D with the minimum global congestion min_global , if |D min_global |=1, select this direction; otherwise, when there are multiple equal-cost paths, a random routing direction is selected to balance the load.

[0031] The benefits of the present invention are:

[0032] By setting up communication links at each routing node, aggregating global congestion information, and combining a two-tiered criterion—local congestion priority and global congestion optimization—path selection is achieved. When the local congestion levels of multiple available paths are similar, a final decision is made based on global congestion information. This paper proposes a weighted congestion propagation mechanism, where the current node's congestion level is calculated by weighting the local congestion and congestion information propagated by neighboring nodes. This significantly improves network throughput and reduces communication latency, while achieving low overhead and strong scalability, making it particularly suitable for large-scale multi-core systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 Schematic diagram of the application of the system of the present invention.

[0034] Figure 2 This is a schematic diagram of the routing decision process. DETAILED DESCRIPTION

[0035] The present invention will be further described below with reference to the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it. However, the embodiments are not intended to limit the present invention.

[0036] Example 1

[0037] The present invention provides an adaptive dynamic routing method for a NoC network on chip, comprising:

[0038] The local congestion of each routing node is collected through the communication link, and the global congestion is generated through a weighted average mechanism. Each routing node integrates delay counters in the east, west, south, and north directions according to the local router. The delay counters are used to monitor the time interval between adjacent data packets leaving the buffer in real time. The time interval between adjacent data packets leaving the buffer is used as the output delay and propagated to neighboring routing nodes. At the same time, the local congestion of the routing node and the global congestion are calculated.

[0039] The local congestion degree of the routing node and the global congestion degree are calculated, including:

[0040] The local congestion degree of the routing node is outputted by calculating the average delay through a sliding window according to the output delay, and the global congestion degree is calculated according to the local congestion degree and the congestion degree of the neighbor routing node.

[0041] The formula can be further used:

[0042]

[0043] The global congestion degree is calculated, wherein C local (i,j) is the local congestion degree of the routing node (i,j), C nighbor (k) is the congestion degree of the neighbor routing node, N(i,j) is the neighbor routing node set, and the directions are east, west, south and north respectively, C final (i,j) is the global congestion degree related to the routing node, and α is a weight factor.

[0044] For example Figure 1 In the congestion awareness unit CAU, delay counters in the east, west, south and north directions can be integrated to calculate the output delay and the congestion degree in real time. Meanwhile, the congestion degree information is transmitted to the east and west neighbor nodes through the on-chip bus to realize the horizontal and vertical diffusion within the same row and column, i.e. through the X / Y dimension hierarchical propagation.

[0045] According to the addresses of the source routing node and the destination routing node, a candidate direction set D={d1,d2,d3,…,d n N} of all possible shortest paths and corresponding output directions is generated, and the optimal path is filtered through the local congestion degree and the global congestion degree in sequence; when there are multiple equivalent paths, the routing direction is randomly selected to balance the network load.

[0046] The optimal path is filtered through the local congestion degree and the global congestion degree in sequence, which can include:

[0047] The local congestion degree is preferentially filtered: the local congestion degree C local (d) of each direction is calculated, and a direction subset with the minimum congestion degree is selected, which is expressed by the formula:

[0048] D min_local ={d|C local (d)=minC local (D))}

[0049] If |D min_local | = 1, the direction is directly selected, otherwise the global congestion degree is filtered,

[0050] When the global congestion degree is filtered, the global congestion degree C min_local of each direction in Dglobal (d), selecting a direction subset D with the minimum global congestion min_global , if |D min_global | = 1, selecting this direction; otherwise, when there are multiple equivalent paths, randomly selecting a routing direction to balance the load.

[0051] Embodiment 2

[0052] The application also provides an adaptive dynamic routing system of a NoC on-chip network, comprising a weighted fusion module and a routing decision module,

[0053] The weighted fusion module collects the local congestion of each routing node through a communication link, and generates a global congestion through a weighted average mechanism, wherein each routing node calculates the delay counter of the integrated east, west, south and north four directions according to the local router, the weighted fusion module uses the delay counter to monitor the time interval of the adjacent data packet leaving the buffer in real time, takes the time interval of the adjacent data packet leaving the buffer as the output delay, and transmits it to the neighbor routing node, while calculating the local congestion of the routing node and the global congestion.

[0054] The routing decision module generates all possible shortest paths and the candidate direction set of the corresponding output direction according to the source routing node and the destination routing node address, and filters the optimal path through the local congestion and the global congestion in turn; when there are multiple equivalent paths, a routing direction is randomly selected to balance the network load.

[0055] The information interaction and execution process between the modules in the above system are based on the same concept as the method embodiments of the application, and the specific content can be referred to the description in the method embodiments of the application, which will not be repeated here.

[0056] Similarly, the system of the application aggregates the global congestion information by setting a communication link at each routing node, and realizes path selection by combining the double-layer criteria of local congestion priority and global congestion optimization. When the local congestion degrees of multiple selectable paths are similar, the final decision is made based on the global congestion information. The application proposes a weighted congestion propagation mechanism, the congestion degree of the current node is calculated by weighted calculation of the local congestion and the congestion information propagated by the adjacent node, which can significantly improve the network throughput and reduce the communication delay, while realizing low overhead and strong scalability, and is particularly suitable for large-scale multi-core systems.

[0057] It should be noted that not all steps and modules in the above processes and system structures are necessary, and some steps or modules can be omitted according to actual needs. The execution order of each step is not fixed and can be adjusted as needed. The system structure described in each of the above embodiments can be a physical structure or a logical structure, that is, some modules can be implemented by the same physical entity, or some modules can be implemented by multiple physical entities, or can be implemented by some components in multiple independent devices together.

[0058] The above-described embodiments are merely preferred embodiments of the present application and are not intended to limit the protection scope of the present application. Any equivalent replacement or transformation made by those skilled in the art based on the present application shall fall within the protection scope of the present application. The protection scope of the present application is defined by the claims.

Claims

1. An adaptive dynamic routing method for NoC on-chip network, characterized by include: The local congestion of each routing node is collected through the communication link, and the global congestion is generated through a weighted average mechanism. Each routing node integrates delay counters in the east, west, south, and north directions according to the local router. The delay counters are used to monitor the time interval between adjacent data packets leaving the buffer in real time. The time interval between adjacent data packets leaving the buffer is used as the output delay and propagated to neighboring routing nodes. At the same time, the local congestion of the routing node and the global congestion are calculated. Based on the source and destination routing node addresses, all possible shortest paths and a set of candidate directions for the corresponding output directions are generated. The optimal path is selected by local congestion and global congestion in turn. When there are multiple equal-cost paths, the routing direction is randomly selected to balance the network load.

2. The adaptive dynamic routing method of NoC network on chip according to claim 1, characterized in that The calculation of the local congestion degree of the routing node and the global congestion degree includes: According to the output delay, the average delay is calculated through the sliding window, the local congestion of the output routing node is output, and the global congestion is calculated based on the local congestion and the congestion of the neighboring routing nodes.

3. The adaptive dynamic routing method of NoC network on chip according to claim 2 is characterized by Using the formula: Calculate the global congestion, where C local (i, j) is the local congestion of routing node (i, j), C nighbor (k) is the congestion degree of the neighbor routing node, N(i,j) is the set of neighbor routing nodes, and the directions are east, west, south, and north respectively. final (i, j) is the global congestion degree associated with the routing node, and α is the weight factor.

4. The adaptive dynamic routing method of NoC network on chip according to claim 1, characterized in that The step of selecting the optimal path by sequentially selecting the local congestion and the global congestion includes: Prioritize local congestion screening: Calculate the local congestion C in each direction local (d) Select the direction subset with the minimum congestion, which is expressed as: D min_local ={d|C local (d)=minC local (D))} If |D min_local |=1, then directly select this direction, otherwise perform global congestion screening. When filtering global congestion, calculate D min_local The global congestion C in each direction global (d) Select the direction subset D with the minimum global congestion min_global , if |D min_global |=1, select this direction; otherwise, when there are multiple equal-cost paths, a random routing direction is selected to balance the load.

5. An adaptive dynamic routing system for NoC network-on-chip, characterized by Including weighted fusion module and routing decision module, The weighted fusion module collects the local congestion of each routing node through the communication link and generates the global congestion through a weighted average mechanism. Each routing node integrates the delay counters in the east, west, south and north directions according to the local router. The weighted fusion module uses the delay counters to monitor the time interval between adjacent data packets leaving the buffer in real time, takes the time interval between adjacent data packets leaving the buffer as the output delay, and propagates it to neighboring routing nodes. At the same time, it calculates the local congestion of the routing node and the global congestion. The routing decision module generates all possible shortest paths and a set of candidate directions for the corresponding output directions based on the addresses of the source and destination routing nodes, and selects the optimal path based on the local and global congestion levels. When there are multiple equal-cost paths, the routing direction is randomly selected to balance the network load.

6. The adaptive dynamic routing system of NoC network on chip according to claim 5, characterized in that The weighted fusion module calculates the local congestion of the routing node and the global congestion, including: According to the output delay, the average delay is calculated through the sliding window, the local congestion of the output routing node is output, and the global congestion is calculated based on the local congestion and the congestion of the neighboring routing nodes.

7. The adaptive dynamic routing system of NoC network on chip according to claim 6, characterized in that The weighted fusion module uses the formula: Calculate the global congestion, where C local (i, j) is the local congestion of routing node (i, j), C nighbor (k) is the congestion degree of the neighbor routing node, N(i,j) is the set of neighbor routing nodes, and the directions are east, west, south, and north respectively. final (i, j) is the global congestion degree associated with the routing node, and α is the weight factor.

8. The adaptive dynamic routing system of NoC network on chip according to claim 5, characterized in that The routing decision module selects the optimal path according to the local congestion and the global congestion in turn, including: Prioritize local congestion screening: Calculate the local congestion C in each direction local (d) Select the direction subset with the minimum congestion, which is expressed as: D min_local ={d|C local (d)=minC local (D))} If |D min_local |=1, then directly select this direction, otherwise perform global congestion screening. When filtering global congestion, calculate D min_local The global congestion C in each direction global (d) Select the direction subset D with the minimum global congestion min_global , if |D min_global |=1, select this direction; otherwise, when there are multiple equal-cost paths, a random routing direction is selected to balance the load.

Images