Multi-core bandwidth management system and control method thereof

Abstract

The invention provides a multi-core bandwidth management system and a control method thereof. The multi-core bandwidth management system comprises a full-link monitoring module; an intelligent prediction model; a hierarchical management and control module; the execution feedback module is respectively connected with the full-link monitoring module and the layered management and control module, and the execution feedback module issues a bandwidth allocation result to a hardware controller, feeds back an execution effect in real time, collects performance indexes of each core after bandwidth allocation in real time, and sends the performance indexes to the hardware controller; whether a comparison deviation exists or not is determined through comparison with a preset scene allowable threshold value, and then a bandwidth adjustment allocation management and control strategy corresponds to the deviation degree of the comparison deviation; through the arrangement of the full-link monitoring module, full-link cooperative monitoring can be realized, multi-dimensional bandwidth data such as cache, interconnection and memory are integrated, a complete bandwidth feature vector is formed on the basis, and the problem of one-sidedness of traditional monitoring of single dimension is solved.

Description

Technical Field

[0001] This invention relates to the technical field of multi-core processor bandwidth management technology, and specifically to a multi-core bandwidth management system and its control method. Background Technology

[0002] With the widespread application of multi-core processors in various fields, the competition for shared resources such as memory bandwidth and interconnect bandwidth is becoming increasingly prominent. 1) Existing static bandwidth allocation methods cannot adapt to dynamic load changes, resulting in low resource utilization; 2) Dynamic allocation methods rely heavily on real-time feedback adjustments, which can lead to performance fluctuations due to the time lag between load fluctuations and subsequent adjustments. 3) At the same time, traditional monitoring technologies often focus on a single dimension (such as memory bandwidth) and lack coordinated monitoring of the entire bandwidth of cache, interconnect, and memory, making it difficult to accurately locate bottlenecks; 4) In addition, different application scenarios (such as cloud computing requiring fairness and avionics requiring determinism) have significantly different bandwidth requirements, and existing solutions are difficult to balance compatibility across multiple scenarios.

[0003] Therefore, there is an urgent need for a bandwidth management solution that integrates intelligent prediction, hierarchical control, and multi-scenario adaptation to address the shortcomings of existing technologies. Summary of the Invention

[0004] To address the deficiencies and shortcomings of the existing technology, this invention provides a multi-core bandwidth management system and its control method.

[0005] The specific solution provided by this invention is as follows: A multi-core bandwidth management system, characterized in that: the system comprises: The end-to-end monitoring module collects multi-dimensional bandwidth data of each core in real time and converts the multi-dimensional bandwidth data corresponding to each core into a bandwidth feature vector corresponding to each core. An intelligent prediction model, which is connected to the end-to-end monitoring module, is used to predict the bandwidth requirements of each core based on bandwidth feature vectors. The hierarchical management module is connected to an intelligent prediction model to load a preset management strategy based on the current application scenario of the system; and allocates bandwidth to each core according to the preset scenario strategy based on the core bandwidth demand predicted by the intelligent prediction model; and when multiple cores request shared bandwidth resources at the same time, arbitrates the sharing requests of each core according to a preset priority. The execution feedback module is connected to the end-to-end monitoring module and the hierarchical management and control module. The execution feedback module sends the bandwidth allocation result to the hardware controller and provides real-time feedback on the execution effect. It collects the performance indicators of each core after bandwidth allocation in real time, and determines whether there is a comparison deviation by comparing them with the preset scenario allowable threshold. Then, it adjusts the bandwidth allocation strategy according to the degree of the comparison deviation.

[0006] As a further preferred embodiment of the present invention, the end-to-end monitoring module includes an integrated hardware performance monitoring unit, a cache monitoring unit, an interconnect monitoring unit, and a memory monitoring unit; wherein... The integrated hardware performance monitoring unit collects the cache hit count, cache miss count, memory read / write bytes, and interconnect bus transmission latency of each core in real time. The cache monitoring unit calculates the cache bandwidth utilization of each core in real time. The cache bandwidth utilization is calculated as: total number of bytes accessed in the cache / total cache capacity × time. The interconnection monitoring unit tracks the number of interconnection bandwidth preemptions between each core in real time. The memory monitoring unit records the bandwidth load of the memory controller in real time.

[0007] As a further preferred embodiment of the present invention, the end-to-end monitoring module is further provided with a normalization processing unit, which normalizes the collected multi-dimensional bandwidth data to form a bandwidth feature vector.

[0008] As a further preferred embodiment of the present invention, the hierarchical management and control module is provided with a scene adaptation unit, which loads a preset management and control strategy according to the current application scenario of the system: When the current application scenario of the system is a cloud computing fairness scenario, set bandwidth fairness weights for each core, and set the bandwidth allocation deviation between any two cores to be lower than the preset deviation value; When the current application scenario of the system is a deterministic avionics scenario, a bandwidth reservation threshold is set for the real-time core. When the current application scenario of the system is a high-performance computing scenario, a bandwidth aggregation threshold is set for each core. When the bandwidth utilization of each core is lower than the bandwidth aggregation threshold, its preset proportion of bandwidth is reclaimed to the shared bandwidth pool.

[0009] As a further preferred embodiment of the present invention, the hierarchical management and control module is provided with a resource allocation unit, which allocates bandwidth according to a preset scenario strategy based on the predicted core bandwidth demand: For real-time cores: allocate reserved bandwidth and predicted incremental bandwidth, and ensure that the predicted incremental bandwidth is higher than the predicted incremental demand. For ordinary cores: a weighted fair allocation method is used to allocate bandwidth ratios to them, wherein the bandwidth allocation ratio = ordinary core weight / total weight, and the ordinary core weight can be dynamically adjusted according to task priority; For idle cores: reclaim a preset proportion of their bandwidth to the shared bandwidth pool for temporary use by high-demand cores.

[0010] As a further preferred embodiment of the present invention, the hierarchical management module is provided with a conflict arbitration unit. When multiple cores simultaneously request shared bandwidth pool resources, the conflict arbitration unit arbitrates the simultaneous requests of multiple cores according to a preset priority. 1) The conflict arbitration unit calculates the demand urgency of each core, where demand urgency = predicted bandwidth demand / currently allocated bandwidth; 2) Sort the urgency of each core requirement from highest to lowest; 3) Shared bandwidth is allocated according to the order of priority; 4) When the urgency of each core requirement is the same, arbitration is performed according to the preset scenario priority. The preset scenario priority satisfies the following: real-time core priority is higher than ordinary core priority, and ordinary core priority is higher than idle core priority.

[0011] As a further preferred embodiment of the present invention, the performance indicators of each core after bandwidth allocation collected in real time by the execution feedback module include at least task latency and bandwidth throughput.

[0012] As a further preferred embodiment of the present invention, when the execution feedback module determines whether there is a comparison deviation by comparing with a preset scene allowable threshold: When the performance indicators of each core after bandwidth allocation collected in real time by the execution feedback module exceed the preset scenario allowable threshold, it is determined that there is a comparison deviation; When the performance indicators of each core after bandwidth allocation collected in real time by the execution feedback module do not exceed the preset scenario allowable threshold, it is determined that there is no comparison deviation.

[0013] As a further preferred embodiment of the present invention, when the execution feedback module adjusts the bandwidth allocation and control strategy according to the degree of deviation of the comparison deviation, if it is determined that the current core does have a comparison deviation, it allocates an additional preset proportion of bandwidth from the shared bandwidth pool to the core; and the higher the degree of deviation of the comparison deviation, the higher the preset proportion of bandwidth allocated from the shared bandwidth pool; at the same time, the deviation data is used as training samples to be sent into the intelligent prediction model to update and optimize the intelligent prediction model.

[0014] Furthermore, the present invention also provides a control method for a multi-core bandwidth management system, characterized by comprising the following steps: S100: The end-to-end monitoring module collects multi-dimensional bandwidth data of each core in real time and converts the multi-dimensional bandwidth data corresponding to each core into a bandwidth feature vector corresponding to each core. S200: The intelligent prediction module predicts the bandwidth requirements of each core based on the bandwidth feature vector. S300: The hierarchical management module loads preset management policies based on the current application scenario of the system; and allocates bandwidth to each core according to the preset scenario policy based on the core bandwidth demand predicted by the intelligent prediction model; and when multiple cores request shared bandwidth resources at the same time, it arbitrates the sharing requests of each core according to the preset priority. S400: The execution feedback module sends the bandwidth allocation result to the hardware controller and provides real-time feedback on the execution effect. It collects the performance indicators of each core after bandwidth allocation in real time, and determines whether there is a comparison deviation by comparing them with the preset scenario allowable threshold. Then, it adjusts the bandwidth allocation control strategy according to the degree of the comparison deviation.

[0015] Compared with existing technologies, the technical effects that this invention can achieve include: 1) This invention provides a multi-core bandwidth management system and its control method. By setting up a full-link monitoring module, full-link collaborative monitoring can be achieved, integrating multi-dimensional bandwidth data such as cache, interconnect, and memory, and forming a complete bandwidth feature vector on this basis, thus solving the one-sided problem of traditional monitoring with a single dimension.

[0016] 2) This invention provides a multi-core bandwidth management system and its control method. By setting up a hierarchical management and control module, it can load preset management and control strategies according to the current application scenario of the system, taking into account multiple scenarios such as cloud computing, aviation systems, and high-performance computing, solving the problem of poor compatibility of existing solutions, eliminating the need to redevelop solutions, and reducing adaptation costs.

[0017] 3) This invention provides a multi-core bandwidth management system and its control method. By incorporating a hierarchical management module and using an intelligent prediction model to anticipate core bandwidth demands, it allocates corresponding bandwidth to cores with different loads according to preset scenario strategies. This allows for real-time adjustment of bandwidth allocation based on dynamic load changes, effectively improving resource utilization. It avoids the time lag of prior art where adjustments are made only after initial fluctuations.

[0018] 4) This invention provides a multi-core bandwidth management system and its control method. By setting an execution feedback module, feedback optimization closed-loop control is achieved. The execution effect is linked with the prediction model and bandwidth allocation strategy to realize continuous iteration of prediction, allocation, feedback and optimization, continuously improve system stability, reduce bandwidth allocation deviation rate and reduce system performance jitter. Attached Figure Description

[0019] Figure 1The diagram shown is a logical structure diagram of the system provided by the present invention.

[0020] Figure 2 The diagram shown is a flowchart of the steps of the method provided by the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] In the description of this invention, it should be noted that the terms "upper," "lower," "inner," "outer," "front end," "rear end," "both ends," "one end," and "the other end," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0023] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installed," "equipped with," "connected," etc., should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be a connection within two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0024] [First Embodiment] like Figure 1 As shown, the first embodiment of the present invention provides a multi-core bandwidth management system, where n represents the number of processor cores and n is a natural number ≥ 2. The system includes: The end-to-end monitoring module collects multi-dimensional bandwidth data of each core in real time and transforms the multi-dimensional bandwidth data corresponding to each core into a bandwidth feature vector corresponding to each core. In this embodiment, the end-to-end monitoring module includes an integrated hardware performance monitoring unit, a cache monitoring unit, an interconnect monitoring unit, and a memory monitoring unit; wherein, The integrated hardware performance monitoring unit (PMU) collects data in real time on each core, including cache hit count, cache miss count (e.g., L2 cache miss count), memory read / write bytes, and interconnect bus transmission latency, to provide necessary data for predicting bandwidth requirements. The cache monitoring unit provides real-time statistics on the cache bandwidth utilization of each core. Cache bandwidth utilization = total number of cache accesses / total cache capacity × time. This is useful when the system is used in a high-performance computing scenario. It helps to compare the current core's bandwidth utilization with the bandwidth aggregation threshold based on the cache bandwidth utilization, and then determine whether it is necessary to reclaim a preset proportion of its bandwidth to the shared bandwidth pool. The interconnect monitoring unit tracks the number of interconnect bandwidth preemptions between cores in real time; for example, the number of request conflicts on the CCI bus (an interface technology that enables cache coherency between multiple processor cores); so that when multiple cores request shared bandwidth pool resources at the same time in the future, the conflict arbitration unit can provide an arbitration basis for the arbitration process of multiple cores' simultaneous requests according to preset priorities. At the same time, since there are also cache hits and cache misses, it also provides a predictive basis for the intelligent prediction model to predict the bandwidth demand of each core to a certain extent. The memory monitoring unit records the bandwidth load of the memory controller in real time, such as the length of the memory access request queue and the average response latency. Since these all affect the number of bytes read and written to memory, they can also provide a basis for the subsequent intelligent prediction model to predict the bandwidth requirements of each core.

[0025] In this embodiment, the end-to-end monitoring module also includes a normalization processing unit. This unit normalizes the collected multi-dimensional bandwidth data to eliminate dimensional differences and form a bandwidth feature vector, which facilitates the subsequent reception, calculation, statistics, and prediction processes of the intelligent prediction model. For example, the normalization process can normalize units such as times / ms, GB / s, and ns, which represent cache bandwidth utilization, interconnect bandwidth load, and memory bandwidth load, to the range [0, 1]. The result of normalization for the core unit with index 1 is [0.6, 0.8, 0.4], which represent cache bandwidth utilization, interconnect bandwidth load, and memory bandwidth load, respectively.

[0026] The system in this embodiment also includes an intelligent prediction model, which is connected to the end-to-end monitoring module to predict the bandwidth requirements of each core based on the bandwidth feature vector. The intelligent prediction model in this embodiment needs to consider at least the following factors when predicting the bandwidth requirements of each core: 1) Cache hit count H and cache miss count M, because the number of cache misses M will cause memory access, thus consuming memory bandwidth; 2) The number of bytes read and written to memory (B) directly reflects the use of memory bandwidth; 3) Since the predicted bandwidth demand BW is positively correlated with the number of cache misses M and the number of bytes read and written in memory B, while the number of cache hits H may reduce memory access, the predicted bandwidth demand BW is negatively correlated with the number of cache hits H.

[0027] That is, the bandwidth demand prediction model in this embodiment is: BW = aM + bB - cH; Where a, b, and c are model coefficients corresponding to the number of cache misses M, the number of bytes read / written in memory B, and the number of cache hits H, respectively, which need to be obtained by fitting actual data.

[0028] The system in this embodiment also includes a hierarchical management and control module, which is connected to an intelligent prediction model to load a preset management and control strategy according to the current application scenario of the system; and allocate bandwidth to each core according to the preset scenario strategy based on the core bandwidth requirements predicted by the intelligent prediction model; and arbitrate the sharing requests of each core according to a preset priority when multiple cores request shared bandwidth resources at the same time. In this embodiment, the hierarchical management module includes a scenario adaptation unit, which loads preset management strategies based on the current application scenario of the system. When the current application scenario of the system is a cloud computing fairness scenario, bandwidth fairness weights are set for each core to prioritize fairness, and the bandwidth allocation deviation between any two cores is set to be lower than a preset deviation value. For example, the preset deviation value can be set to 5%, that is, the bandwidth allocation deviation between any two cores is set to be lower than 5%. When the current application scenario of the system is a deterministic avionics scenario, a bandwidth reservation threshold is set for the real-time core (such as the lidar data processing core). For example, the bandwidth reservation threshold can be set to 60%, that is, at least 60% of the memory bandwidth is reserved for the real-time core to prioritize the low latency of each core. When the current application scenario of the system is a high-performance computing scenario, a bandwidth aggregation threshold is set for each core. When the bandwidth utilization of each core is lower than the bandwidth aggregation threshold, its preset proportion of bandwidth is recycled to the shared bandwidth pool to prioritize improving the utilization rate. For example, if the bandwidth aggregation threshold is set to 30% and the preset proportion of bandwidth is set to 20%, then when the bandwidth utilization of a certain core is lower than 30%, 20% of its bandwidth is recycled to the aggregation pool.

[0029] By setting up a hierarchical management and control module, preset management and control policies can be loaded according to the current application scenario of the system, taking into account multiple scenarios such as cloud computing, aviation systems, and high-performance computing. This solves the problem of poor compatibility of existing solutions, eliminates the need to redevelop solutions, and reduces adaptation costs.

[0030] Preferably, in this embodiment, the current application scenario of the system can be manually selected based on the actual usage scenario of the system. For example, a corresponding dialog box can be provided for the user to select based on the current application scenario; alternatively, the corresponding application scenario can be determined based on relevant parameters of the actual usage scenario. For instance, the current application scenario can be automatically determined as an avionics deterministic scenario based on whether there are urgent tasks in the system that must be completed within a certain time (to ensure that these urgent tasks always have sufficient memory bandwidth); or a combination of priority ordering and manual adjustment can be used based on relevant parameters of the actual usage scenario. For example, the avionics deterministic scenario can be automatically determined as having the highest priority based on whether there are urgent tasks in the system that must be completed within a certain time (to ensure that these urgent tasks always have sufficient memory bandwidth). In this case, a high-performance computing scenario or a cloud computing fairness scenario, which is not the highest priority, can be selected by manual adjustment to meet different user needs.

[0031] In this embodiment, the hierarchical management module includes a resource allocation unit, which allocates bandwidth according to a preset scenario strategy based on the predicted core bandwidth demand. For the real-time core: allocate reserved bandwidth and predicted incremental bandwidth, and ensure that the predicted incremental bandwidth is higher than the predicted incremental demand to meet the predicted bandwidth requirements; for example, if the reserved bandwidth is set to 60% and the predicted incremental demand is 5%, then the predicted incremental bandwidth is set to 10%. For the real-time core, 60% bandwidth is reserved, and if the predicted incremental demand is 5%, an additional 10% bandwidth is allocated to it; this is to allocate relatively high bandwidth to the real-time core to ensure its real-time efficiency. For ordinary cores: a weighted fair allocation method is used to allocate bandwidth proportions to them. The bandwidth allocation proportion = ordinary core weight / total weight, and the ordinary core weight can be dynamically adjusted according to task priority to meet different bandwidth requirements. For idle cores: reclaim a preset percentage of their bandwidth to the shared bandwidth pool for temporary use by high-demand cores. The preset percentage of bandwidth to be reclaimed can be set to, for example, 80%.

[0032] By incorporating a hierarchical management module and employing an intelligent prediction model to anticipate core bandwidth demands, bandwidth is allocated to different load cores according to preset scenario strategies. This allows for real-time adjustment of bandwidth allocation based on dynamic load changes, effectively improving resource utilization. It avoids the time lag of adjustments made after initial fluctuations, a problem common in existing technologies.

[0033] In this embodiment, the core type is determined as a real-time core, a normal core, or an idle core based on the urgency of the tasks running on the core. Specifically, the core on which the tasks running are of the highest urgency is determined as a real-time core, while the core on which the tasks running are of the lowest urgency is determined as an idle core, and the normal core is somewhere in between.

[0034] In this embodiment, the hierarchical management module includes a conflict arbitration unit. When multiple cores simultaneously request shared bandwidth pool resources, the conflict arbitration unit arbitrates the simultaneous requests from multiple cores according to a preset priority. The arbitration process includes: 1) The conflict arbitration unit calculates the demand urgency of each core. Demand urgency = predicted bandwidth demand / currently allocated bandwidth; 2) Sort the urgency of each core requirement from highest to lowest; 3) Shared bandwidth is allocated according to the order of priority; 4) When the urgency of each core requirement is the same, arbitration is carried out according to the preset scenario priority. The preset scenario priority satisfies the following: real-time core priority is higher than ordinary core priority, and ordinary core priority is higher than idle core priority.

[0035] When multiple cores request arbitration simultaneously, bandwidth can be allocated based on the urgency of the demand, prioritizing the most urgent bandwidth requirement. If the urgency is the same, bandwidth can be allocated according to core category to ensure that real-time cores have sufficient bandwidth to improve processing efficiency.

[0036] like Figure 1 As shown, the system also includes an execution feedback module, which is connected to the end-to-end monitoring module and the hierarchical management and control module. The execution feedback module sends the bandwidth allocation results to the hardware controller and provides real-time feedback on the execution effect. It collects the performance indicators of each core after bandwidth allocation in real time, and determines whether there is a comparison deviation by comparing them with the preset scenario allowable threshold. Then, it adjusts the bandwidth allocation strategy according to the degree of the comparison deviation.

[0037] In this embodiment, the performance indicators of each core after bandwidth allocation collected in real time by the execution feedback module include at least task latency and bandwidth throughput, so as to determine whether there is a comparison deviation by comparing the collected performance indicators of each core with the preset scenario allowable threshold.

[0038] When the execution feedback module determines whether there is a comparison deviation by comparing it with the preset scene allowable threshold: When the performance indicators of each core after bandwidth allocation collected in real time by the execution feedback module exceed the preset scenario allowable threshold, it is determined that there is a comparison deviation; When the performance indicators of each core after bandwidth allocation collected in real time by the execution feedback module do not exceed the preset scenario allowable threshold, it is determined that there is no comparison deviation.

[0039] For example, if the allowable threshold for the partition switching scenario delay time is set to 5µs, then if the real-time collected partition switching scenario delay time is greater than 5µs, then it is considered that there is a comparison deviation; otherwise, it is considered that there is no comparison deviation.

[0040] When the execution feedback module adjusts the bandwidth allocation and control strategy according to the degree of comparison deviation, if it is determined that the current core does indeed have a comparison deviation, it will allocate an additional preset proportion of bandwidth from the shared bandwidth pool to the core. For example, if the latency exceeds the allowable threshold of 20%, an additional 10% of the bandwidth will be allocated from the shared bandwidth pool to the core. The higher the degree of comparison deviation, the higher the preset proportion of additional bandwidth allocated from the shared bandwidth pool to meet the bandwidth requirements. At the same time, the deviation data is used as training samples to feed into the intelligent prediction model to update and optimize the intelligent prediction model, so as to further improve the prediction accuracy continuously and stably.

[0041] [Second Embodiment] like Figure 2 As shown, the second embodiment of the present invention also provides a control method for a multi-core bandwidth management system, comprising the following steps: S100: The end-to-end monitoring module collects multi-dimensional bandwidth data of each core in real time and converts the multi-dimensional bandwidth data corresponding to each core into a bandwidth feature vector corresponding to each core. S200: The intelligent prediction module predicts the bandwidth requirements of each core based on the bandwidth feature vector. S300: The hierarchical management module loads preset management policies based on the current application scenario of the system; and allocates bandwidth to each core according to the preset scenario policy based on the core bandwidth demand predicted by the intelligent prediction model; and when multiple cores request shared bandwidth resources at the same time, it arbitrates the sharing requests of each core according to the preset priority. S400: The execution feedback module sends the bandwidth allocation result to the hardware controller and provides real-time feedback on the execution effect. It collects the performance indicators of each core after bandwidth allocation in real time, and determines whether there is a comparison deviation by comparing them with the preset scenario allowable threshold. Then, it adjusts the bandwidth allocation strategy according to the degree of the comparison deviation.

[0042] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A multi-core bandwidth management system, characterized in that: The system includes: The end-to-end monitoring module collects multi-dimensional bandwidth data of each core in real time and converts the multi-dimensional bandwidth data corresponding to each core into a bandwidth feature vector corresponding to each core. An intelligent prediction model, which is connected to the end-to-end monitoring module, is used to predict the bandwidth requirements of each core based on bandwidth feature vectors. The hierarchical management module is connected to an intelligent prediction model to load a preset management strategy based on the current application scenario of the system; and allocates bandwidth to each core according to the preset scenario strategy based on the core bandwidth demand predicted by the intelligent prediction model; and when multiple cores request shared bandwidth resources at the same time, arbitrates the sharing requests of each core according to a preset priority. The execution feedback module is connected to the end-to-end monitoring module and the hierarchical management and control module. The execution feedback module sends the bandwidth allocation result to the hardware controller and provides real-time feedback on the execution effect. It collects the performance indicators of each core after bandwidth allocation in real time, and determines whether there is a comparison deviation by comparing them with the preset scenario allowable threshold. Then, it adjusts the bandwidth allocation strategy according to the degree of the comparison deviation.

2. The multi-core bandwidth management system according to claim 1, characterized in that: The end-to-end monitoring module includes an integrated hardware performance monitoring unit, a cache monitoring unit, an interconnect monitoring unit, and a memory monitoring unit; wherein... The integrated hardware performance monitoring unit collects the cache hit count, cache miss count, memory read / write bytes, and interconnect bus transmission latency of each core in real time. The cache monitoring unit calculates the cache bandwidth utilization of each core in real time. The cache bandwidth utilization is calculated as: total number of bytes accessed in the cache / total cache capacity × time. The interconnection monitoring unit tracks the number of interconnection bandwidth preemptions between each core in real time. The memory monitoring unit records the bandwidth load of the memory controller in real time.

3. The multi-core bandwidth management system according to claim 1, characterized in that: The end-to-end monitoring module also includes a normalization processing unit, which normalizes the collected multi-dimensional bandwidth data to form a bandwidth feature vector.

4. The multi-core bandwidth management system according to claim 1, characterized in that: The hierarchical management module includes a scenario adaptation unit, which loads preset management strategies based on the current application scenario of the system. When the current application scenario of the system is a cloud computing fairness scenario, set bandwidth fairness weights for each core, and set the bandwidth allocation deviation between any two cores to be lower than the preset deviation value; When the current application scenario of the system is a deterministic avionics scenario, a bandwidth reservation threshold is set for the real-time core. When the current application scenario of the system is a high-performance computing scenario, a bandwidth aggregation threshold is set for each core. When the bandwidth utilization of each core is lower than the bandwidth aggregation threshold, its preset proportion of bandwidth is reclaimed to the shared bandwidth pool.

5. A multi-core bandwidth management system according to claim 4, characterized in that: The hierarchical management module includes a resource allocation unit, which allocates bandwidth according to a preset scenario strategy based on the predicted core bandwidth demand. For real-time cores: allocate reserved bandwidth and predicted incremental bandwidth, and ensure that the predicted incremental bandwidth is higher than the predicted incremental demand. For ordinary cores: a weighted fair allocation method is used to allocate bandwidth ratios to them, wherein the bandwidth allocation ratio = ordinary core weight / total weight, and the ordinary core weight can be dynamically adjusted according to task priority; For idle cores: reclaim a preset proportion of their bandwidth to the shared bandwidth pool for temporary use by high-demand cores.

6. A multi-core bandwidth management system according to claim 5, characterized in that: The hierarchical management module includes a conflict arbitration unit. When multiple cores simultaneously request shared bandwidth pool resources, the conflict arbitration unit arbitrates the simultaneous requests from multiple cores according to a preset priority. 1) The conflict arbitration unit calculates the demand urgency of each core, where demand urgency = predicted bandwidth demand / currently allocated bandwidth; 2) Sort the urgency of each core requirement from highest to lowest; 3) Shared bandwidth is allocated according to the order of priority; 4) When the urgency of each core requirement is the same, arbitration is performed according to the preset scenario priority. The preset scenario priority satisfies the following: real-time core priority is higher than ordinary core priority, and ordinary core priority is higher than idle core priority.

7. A multi-core bandwidth management system according to claim 1, characterized in that: The performance metrics of each core after bandwidth allocation, which are collected in real time by the execution feedback module, include at least task latency and bandwidth throughput.

8. A multi-core bandwidth management system according to claim 7, characterized in that: When the execution feedback module determines whether there is a comparison deviation by comparing it with a preset scene allowable threshold: When the performance indicators of each core after bandwidth allocation collected in real time by the execution feedback module exceed the preset scenario allowable threshold, it is determined that there is a comparison deviation; When the performance indicators of each core after bandwidth allocation collected in real time by the execution feedback module do not exceed the preset scenario allowable threshold, it is determined that there is no comparison deviation.

9. A multi-core bandwidth management system according to claim 8, characterized in that: When the execution feedback module adjusts the bandwidth allocation and control strategy according to the degree of deviation of the comparison deviation, if it is determined that the current core does have a comparison deviation, it will allocate an additional preset proportion of bandwidth from the shared bandwidth pool to the core; and the higher the degree of deviation of the comparison deviation, the higher the preset proportion of bandwidth allocated from the shared bandwidth pool; at the same time, the deviation data is used as training samples to be sent into the intelligent prediction model to update and optimize the intelligent prediction model.

10. A control method for a multi-core bandwidth management system according to any one of claims 1-9, characterized in that: Includes the following steps: S100: The end-to-end monitoring module collects multi-dimensional bandwidth data of each core in real time and converts the multi-dimensional bandwidth data corresponding to each core into a bandwidth feature vector corresponding to each core. S200: The intelligent prediction module predicts the bandwidth requirements of each core based on the bandwidth feature vector. S300: The hierarchical management module loads preset management policies based on the current application scenario of the system; In addition, based on the core bandwidth demand predicted by the intelligent prediction model, bandwidth is allocated to each core according to the preset scenario strategy: and when multiple cores request shared bandwidth resources at the same time, the sharing requests of each core are arbitrated according to the preset priority. S400: The execution feedback module sends the bandwidth allocation result to the hardware controller and provides real-time feedback on the execution effect. It collects the performance indicators of each core after bandwidth allocation in real time, and determines whether there is a comparison deviation by comparing them with the preset scenario allowable threshold. Then, it adjusts the bandwidth allocation control strategy according to the degree of the comparison deviation.

Images