Performance and power consumption collaborative optimization method for PCle switching board
By constructing dynamic load feature vectors and real-time power consumption monitoring, the load distribution and power consumption management of the PCIe switch board are optimized, solving the problems of unreasonable link load and non-dynamic power consumption, and achieving efficient and energy-saving data transmission.
Patent Information
- Application Number
- CN202510879057.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-27
- Publication Date
- 2025-10-03
AI Technical Summary
The existing technology has problems in the performance optimization of PCIe switch boards, such as unreasonable link load distribution and non-dynamic power consumption management, which leads to increased data transmission delays, resource waste and unnecessary increase in energy consumption.
By constructing a dynamic load feature vector, using a dynamic load balancing algorithm to calculate the load distribution weight, combining the business traffic model and link training model, monitoring power consumption parameters in real time, adjusting the data transmission rate and link status, and optimizing data transmission scheduling and power consumption management.
It achieves efficient operation of the PCle switch board in different application scenarios, avoids link load imbalance, reduces power consumption, improves data transmission efficiency and overall performance, and has self-optimization capabilities.
Smart Images
Figure CN120750859A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of PCIe switch board power consumption optimization, and in particular to a performance and power consumption collaborative optimization method for PCIe switch boards. Background Art
[0002] With the explosive growth in data processing demands, PCIe switch boards, as key components for high-speed data transmission, face a pressing challenge in optimizing both performance and power consumption. In modern data centers and high-performance computing systems, PCIe switch boards are crucial for connecting multiple devices and enabling high-speed data exchange. However, traditional PCIe switch board optimization methods often focus solely on either improving performance or reducing power consumption, making it difficult to achieve a balance between these two goals in complex and ever-changing business scenarios.
[0003] Existing technologies for optimizing the performance of PCIe switch boards have significant shortcomings. First, during data transmission, link load distribution is irrational and lacks a dynamic adjustment mechanism. Loads on different links vary significantly, with some links operating at high loads, leading to increased data transmission latency and decreased throughput, while other links remain idle or underloaded, resulting in wasted resources and severely impacting the overall performance of the PCIe switch board.
[0004] Existing methods for power management also have drawbacks. Most solutions employ fixed power control strategies that fail to dynamically adjust based on actual service load and link status. When the system load is low, the PCIe switch module maintains a high-power operation mode, resulting in unnecessary energy consumption. However, in high-load scenarios, power consumption cannot be properly adjusted to maintain performance. This makes it difficult to find the optimal balance between power consumption and performance, limiting the efficient operation of the PCIe switch module in different application scenarios. Summary of the Invention
[0005] In order to overcome the shortcomings and deficiencies of the prior art, the present invention provides a performance and power consumption collaborative optimization method for a PCIe switch board.
[0006] The technical solution adopted by the present invention is a method for collaboratively optimizing the performance and power consumption of a PCIe switch board, which includes: Step S1: Based on the current working status and historical load data of the PCIe switch board, a dynamic load feature vector is constructed, and the feature vector is processed using a preset dynamic load balancing algorithm to obtain the load distribution weight of each link; Step S2: constructing a link training initial parameter set based on the electrical characteristic parameters of each port of the PCle switch board and the load distribution weight; Step S3: Based on the link training and negotiation model, the link of the PCIe switch board is trained according to the link training initial parameter set. During the training process, the signal quality parameters of the link are dynamically monitored and the link training parameters are adjusted according to the signal quality parameters until the link training is completed. Step S4: After link training is completed, a data transmission scheduling plan is constructed based on the service traffic model of the PCIe switch board and the load distribution weight of each link; Step S5: performing data transmission on the PCIe switch board according to the data transmission scheduling scheme, while simultaneously monitoring the power consumption parameters of the PCIe switch board in real time. When the power consumption parameters exceed a preset power consumption threshold, the power consumption of the PCIe switch board is reduced by adjusting the data transmission rate and link status. Step S6: Optimize and update the parameters of the dynamic load balancing algorithm and the link training and negotiation model according to the performance indicators and power consumption parameters of the PCle switch board during data transmission.
[0007] Furthermore, the dynamic load balancing algorithm uses the following model formula to calculate the load distribution weight of each link: ; in, For the The load distribution weight of the links, For the The historical load average of the links, For the The current number of connections of the link, is the total number of links, The weight adjustment coefficient is used to adjust the impact of the historical load average and the current number of connections on the load distribution weight.
[0008] Furthermore, in the link training and negotiation model, the following model formula is used to adjust the link training parameters: ; in, is the adjusted link training parameter, is the link training parameter before adjustment, is the adjustment factor, is the signal quality parameter of the current link, is the preset signal quality target parameter.
[0009] Furthermore, the data transmission scheduling scheme is constructed by using the following model formula to determine the transmission order of data on each link: ; in, For the The priority of a data block in the link transmission order, For the The business urgency of each data block, For the The data size of each data block, is the total number of data blocks, It is the priority adjustment coefficient, which is used to adjust the impact of business urgency and data size on the transmission sequence priority.
[0010] Furthermore, when the power consumption parameter exceeds a preset power consumption threshold, the data transmission rate is adjusted using the following model formula: ; in, is the adjusted data transmission rate, is the data transmission rate before adjustment, is the rate adjustment coefficient, is the power consumption parameter of the current PCle switch board, is the preset power consumption threshold, The maximum allowable power consumption of the PCIe switch board.
[0011] Furthermore, when optimizing and updating the parameters of the dynamic load balancing algorithm and the link training and negotiation model, the following model formula is used: ; in, To optimize the updated model parameters, To optimize the model parameters before updating, is the update coefficient, is the target power consumption, is the current power consumption, is the target performance indicator, is the current performance indicator.
[0012] Furthermore, the step S3, training the link based on the link training and negotiation model, includes the following sub-steps: Step S31: Sending a training signal to the link of the PCIe switch board according to the link training initial parameter set to initialize the physical layer connection state of the link; Step S32: receiving the signal quality parameter fed back by the link, and judging whether the link meets the signal quality requirement for data transmission according to a preset signal quality evaluation standard; Step S33: If the requirements are not met, adjust the link training parameters according to the link training and negotiation model, and send the training signal again until the link meets the signal quality requirements for data transmission.
[0013] Furthermore, the step S4 of constructing a data transmission scheduling scheme includes the following sub-steps: Step S41: Analyze the service traffic model of the PCIe switch board to determine the data traffic characteristics and service priorities of different service types; Step S42: combining the load distribution weight of each link and the service priority, classifying the data according to service type and priority; Step S43: Allocate transmission links for different categories of data according to the classification results and the bandwidth and delay parameters of the links, and determine the transmission order of the data on each link.
[0014] Furthermore, the step S5 of real-time monitoring and adjusting the power consumption parameters of the PCIe switch board includes the following sub-steps: Step S51: collecting power consumption parameters of the PCle switch board in real time through the power consumption monitoring module on the PCle switch board; Step S52: Compare the collected power consumption parameters with a preset power consumption threshold to determine whether the power consumption exceeds the threshold; Step S53: If the power consumption exceeds the standard, first try to reduce the data transmission rate of non-critical services and observe the changes in power consumption. If the power consumption still exceeds the standard, further adjust the link status and switch some links to low power mode.
[0015] A performance and power consumption collaborative optimization method for PCIe switch boards is implemented through the following units: A load feature construction unit is used to construct a dynamic load feature vector based on the current working status and historical load data of the PCIe switch board; a load weight calculation unit, connected to the load feature construction unit, for processing the feature vector using a preset dynamic load balancing algorithm to obtain a load distribution weight for each link; A link training parameter construction unit, connected to the load weight calculation unit, for constructing a link training initial parameter set based on the electrical characteristic parameters of each port of the PCle switch board and the load distribution weight; A link training execution unit, connected to the link training parameter construction unit, for training the link of the PCIe switch board based on the link training and negotiation model and the link training initial parameter set; A data transmission scheduling construction unit, connected to the link training execution unit, is used to construct a data transmission scheduling plan based on the service traffic model of the PCle switch board and the load distribution weight of each link after the link training is completed; The power consumption monitoring and adjustment unit is connected to the data transmission scheduling construction unit and is used to monitor the power consumption parameters of the PCle switch board in real time when data is transmitted on the PCle switch board according to the data transmission scheduling scheme, and when the power consumption parameters exceed the preset power consumption threshold, reduce the power consumption of the PCle switch board by adjusting the data transmission rate and link status.
[0016] Beneficial effects: The present invention proposes a method for collaborative optimization of performance and power consumption for PCle switch boards. The method constructs a dynamic load feature vector based on the current working status and historical load data of the PCle switch board, uses a dynamic load balancing algorithm to obtain the load distribution weight of each link, and constructs a data transmission scheduling scheme in combination with the service traffic model, so that data can be reasonably distributed along the transmission path based on factors such as link load and service priority, avoiding situations where some links are highly loaded and congested and some links are idle, making full use of link resources, and effectively improving data transmission efficiency and the overall performance of the PCle switch board. In terms of power consumption management, by real-time monitoring of the power consumption parameters of the PCle switch board, when the preset threshold is exceeded, the data transmission rate and link status are intelligently adjusted, non-critical services are slowed down or the link is switched to a low-power mode, thus changing the drawbacks of the traditional fixed power consumption control strategy. At the same time, the link training and negotiation model constructs an initial parameter set based on the electrical characteristics of the port and the load weight, and dynamically adjusts it according to the signal quality during training to ensure that the link operates in the optimal state, reduce unnecessary power consumption waste, and achieve collaborative optimization of performance and power consumption. In addition, the algorithm and model parameters can be optimized and updated based on the performance indicators and power consumption parameters in actual operation, so that the optimization solution can adapt to different application scenarios and continuously improve the performance of the PCle switch board in performance and power consumption management. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] Figure 1 is a flow chart of the method steps of the present invention; Figure 2 This is a diagram of the unit composition of the method implementation of the present invention. DETAILED DESCRIPTION
[0018] It should be noted that, unless there is a conflict, the embodiments in this application and the features described in the embodiments can be combined with each other. The application is further described in detail below with reference to the accompanying drawings and specific embodiments.
[0019] like Figure 1 As shown in FIG, a performance and power consumption collaborative optimization method for a PCIe switch board includes: Step S1: Based on the current working status and historical load data of the PCIe switch board, a dynamic load feature vector is constructed, and the feature vector is processed using a preset dynamic load balancing algorithm to obtain the load distribution weight of each link; Specifically, step S1 constructs a dynamic load feature vector based on the current operating status and historical load data of the PCIe switch board. This feature vector is then processed using a preset dynamic load balancing algorithm to obtain load distribution weights for each link. This step collects operating status data such as real-time traffic, bandwidth utilization, and transmission delay for each link on the PCIe switch board, and combines this with historical load statistics to generate vector data that comprehensively reflects the current load distribution characteristics. The dynamic load balancing algorithm, based on this feature vector, comprehensively considers the link's real-time performance and historical load conditions to calculate the load distribution weights for each link in its current state, providing a basis for subsequent data transmission scheduling.
[0020] In terms of implementation, a comprehensive data collection mechanism must first be established to ensure accurate acquisition of the operating status and historical load data for each link on the PCIe switch board. Feature extraction techniques are then used to process the collected data to construct a representative dynamic load feature vector. Finally, this feature vector is input into a pre-defined dynamic load balancing algorithm. Through calculation and analysis, the algorithm determines the load distribution weight for each link, achieving dynamic optimization of the PCIe switch board load distribution.
[0021] Step S2: constructing a link training initial parameter set based on the electrical characteristic parameters of each port of the PCle switch board and the load distribution weight; Specifically, step S2 constructs a set of initial link training parameters based on the electrical characteristic parameters of each port on the PCIe switch board and the load distribution weights. In this step, the electrical characteristic parameters include indicators such as the port's signal strength, noise margin, and transmission rate, which directly impact the link's transmission quality and stability. Combined with the load distribution weights obtained in step S1, and taking into account the load conditions of each link and the electrical characteristics of the ports, a set of initial link training parameters appropriate for the current system state is constructed, providing the foundation for subsequent link training.
[0022] This step requires precise measurement and analysis of the electrical characteristics of each port on the PCIe switch board to obtain accurate parameter values. Furthermore, the training parameters for each link are appropriately configured based on the load distribution weights, ensuring that highly loaded links receive more optimized training parameters and improve transmission performance. This approach creates an initial set of link training parameters that better adapts to the actual operating conditions of the PCIe switch board, laying a solid foundation for subsequent link training and data transmission.
[0023] Step S3: Based on the link training and negotiation model, the link of the PCIe switch board is trained according to the link training initial parameter set. During the training process, the signal quality parameters of the link are dynamically monitored and the link training parameters are adjusted according to the signal quality parameters until the link training is completed. Specifically, step S3 trains the PCIe switch board link based on the link training and negotiation model and the initial link training parameter set. During training, the link's signal quality parameters are dynamically monitored and adjusted accordingly until link training is complete. In this step, the link training and negotiation model initializes link training using the initial parameter set according to the pre-set training process. During training, link signal quality parameters, such as bit error rate and signal-to-noise ratio, are monitored in real time. These parameters are analyzed and evaluated to determine the link's training effectiveness and transmission quality.
[0024] When monitored signal quality parameters do not meet requirements, the link training and negotiation model dynamically adjusts the link training parameters based on pre-set adjustment strategies. For example, if the bit error rate is too high, the model will appropriately reduce the transmission rate or increase the signal strength to improve link transmission stability. Through continuous monitoring and adjustment, link training is completed until the link signal quality parameters reach the preset standards, ensuring that the link can transmit data under optimal conditions.
[0025] Step S4: After link training is completed, a data transmission scheduling plan is constructed based on the service traffic model of the PCIe switch board and the load distribution weight of each link; Specifically, after link training is complete, step S4 constructs a data transmission scheduling plan based on the service traffic model of the PCIe switch board and the load distribution weights of each link. The service traffic model describes the data traffic characteristics of different service types on the PCIe switch board, including traffic volume, time distribution, priority, and other information. Combined with the load distribution weights obtained in step S1, and taking into account the load capacity of each link and service traffic requirements, a reasonable data transmission scheduling plan is formulated to ensure efficient and orderly data transmission between links.
[0026] When developing a data transmission scheduling solution, the service traffic on the PCIe switch board must be analyzed in detail to determine the traffic characteristics and priorities of different service types. Then, based on the load distribution weights of each link, the data of different service types is appropriately allocated to each link, and the data transmission order and timing on each link are determined. This approach allows the constructed data transmission scheduling solution to fully utilize the resources of each link, improve data transmission efficiency, and meet the quality of service requirements of different services.
[0027] Step S5: performing data transmission on the PCIe switch board according to the data transmission scheduling scheme, while simultaneously monitoring the power consumption parameters of the PCIe switch board in real time. When the power consumption parameters exceed a preset power consumption threshold, the power consumption of the PCIe switch board is reduced by adjusting the data transmission rate and link status. Specifically, step S5 performs data transmission on the PCIe switch board according to the data transmission scheduling scheme while simultaneously monitoring the power consumption parameters of the PCIe switch board in real time. When the power consumption parameters exceed a preset power consumption threshold, the data transmission rate and link status are adjusted to reduce the power consumption of the PCIe switch board. Power consumption is a key consideration during data transmission. Excessive power consumption not only increases energy consumption but also affects the stability and lifespan of the PCIe switch board.
[0028] To effectively control power consumption, this step deploys a power consumption monitoring module on the PCIe switch board. This module collects power consumption parameters in real time and compares them with preset power consumption thresholds. If the monitored power consumption parameters exceed the thresholds, the system first attempts to reduce the data transmission rate for non-critical services to minimize unnecessary energy consumption. If power consumption still falls short of the target after reducing the transmission rate, further adjustments are made to the link state, such as switching some links to low-power mode or temporarily shutting down idle links, to further reduce power consumption. This dynamic power consumption control mechanism effectively reduces PCIe switch board power consumption while ensuring data transmission performance.
[0029] Step S6: Optimize and update the parameters of the dynamic load balancing algorithm and the link training and negotiation model according to the performance indicators and power consumption parameters of the PCle switch board during data transmission.
[0030] Specifically, step S6 optimizes and updates the parameters of the dynamic load balancing algorithm and the link training and negotiation models based on the performance indicators and power consumption parameters of the PCIe switch board during data transmission. During actual operation of the PCIe switch board, system status and service requirements are constantly changing, and the original algorithm and model parameters may no longer be suitable for the current situation. By collecting performance indicators such as throughput and latency, as well as power consumption parameters during data transmission, and analyzing and evaluating this data, we can understand the operational effectiveness of the current algorithm and model.
[0031] Based on the analysis results, the parameters of the dynamic load balancing algorithm and link training and negotiation models are optimized and updated to better adapt to the actual operating conditions of the PCIe switch board. For example, if a link is found to be overloaded, resulting in performance degradation, the algorithm parameters can be adjusted to increase the load distribution weight of that link and improve its resource utilization. This continuous optimization and update process continuously improves the performance and power management capabilities of the PCIe switch board, ensuring efficient and stable system operation.
[0032] Preferably, the dynamic load balancing algorithm uses the following model formula to calculate the load distribution weight of each link: ; in, For the The load distribution weight of the links, For the The historical load average of the links, For the The current number of connections of the link, is the total number of links, The weight adjustment coefficient is used to adjust the impact of the historical load average and the current number of connections on the load distribution weight.
[0033] Specifically, the link load distribution weight is calculated by combining the historical load average and the current number of connections. The historical load average reflects the long-term stability of the link's working state, while the current number of connections reflects the real-time load pressure. The weight adjustment coefficient can adjust the degree of influence of the two on load distribution according to the actual application scenario. During implementation, a historical load database is first established. Traffic data of each link is collected and averaged according to a preset time window. The current number of connections is also counted in real time. The weighted sum of the two is then normalized to obtain the load distribution weight for each link. Finally, the weight is input into the scheduler to dynamically adjust the data traffic distribution to achieve load balancing and avoid performance degradation caused by local link overload.
[0034] Preferably, in the link training and negotiation model, the following model formula is used to adjust the link training parameters: ; in, is the adjusted link training parameter, is the link training parameter before adjustment, is the adjustment factor, is the signal quality parameter of the current link, is the preset signal quality target parameter.
[0035] Specifically, during link training, training parameters are dynamically adjusted by comparing current signal quality parameters with preset target values. Adjustment factors control the magnitude of parameter adjustments to balance convergence speed and stability. During implementation, initial training parameters are configured and a training sequence is sent. The receiver measures signal quality parameters such as bit error rate and signal-to-noise ratio. The measured values are compared with the target values. If the deviation exceeds the tolerance, the training parameters are modified according to the adjustment strategy, such as adjusting the transmitter pre-emphasis or the receiver equalization coefficient. The training process is repeated until signal quality meets the target, ensuring optimal transmission at the link physical layer and reducing bit error rate and retransmission rate.
[0036] Preferably, the data transmission scheduling scheme is constructed by using the following model formula to determine the transmission order of data on each link: ; in, For the The priority of a data block in the link transmission order, For the The business urgency of each data block, For the The data size of each data block, is the total number of data blocks, It is the priority adjustment coefficient, which is used to adjust the impact of business urgency and data size on the transmission sequence priority.
[0037] Specifically, when constructing a data transmission scheduling plan, transmission priority is determined based on both service urgency and data volume. A priority adjustment coefficient dynamically adjusts the weight of these two factors based on service type. During implementation, the service traffic model is analyzed, services are classified and quantified by urgency, and data block sizes are also measured. A priority score is calculated for each data block, with higher scores receiving higher transmission priority. Taking into account link load status, high-priority data blocks are allocated to low-load, high-performance links, while low-priority data blocks are allocated to the remaining links, achieving efficient resource utilization and ensuring quality of service.
[0038] Preferably, when the power consumption parameter exceeds a preset power consumption threshold, the data transmission rate is adjusted using the following model formula: ; in, is the adjusted data transmission rate, is the data transmission rate before adjustment, is the rate adjustment coefficient, is the power consumption parameter of the current PCle switch board, is the preset power consumption threshold, The maximum allowable power consumption of the PCIe switch board.
[0039] Specifically, in terms of power consumption control, when power consumption exceeds the limit, the transmission rate is adjusted to reduce power consumption. The rate adjustment coefficient is dynamically calculated based on the difference between the current power consumption and the threshold. During implementation, the power consumption monitoring module collects system power consumption data in real time and compares it with the preset threshold. If the limit is exceeded, the speed reduction mechanism is activated, first reducing the rate of non-critical services, and then gradually adjusting it according to the preset step size. After each adjustment, the power consumption is re-evaluated. If it still exceeds the limit, the speed reduction continues until the power consumption meets the limit or reaches the minimum rate limit. If the speed reduction fails to meet the requirements, the link sleep strategy is triggered, shutting down some non-critical links to achieve refined power consumption control.
[0040] Preferably, when optimizing and updating the parameters of the dynamic load balancing algorithm and the link training and negotiation model, the following model formula is used: ; in, To optimize the updated model parameters, To optimize the model parameters before updating, is the update coefficient, is the target power consumption, is the current power consumption, is the target performance indicator, is the current performance indicator.
[0041] Specifically, by comparing the target with the current power consumption and performance indicators, the algorithm and model parameters are optimized and updated. The update coefficients control the parameter adjustment step size to avoid oscillation. During implementation, system operation data is continuously collected to calculate the deviation between the current performance indicators and the target values. Based on the direction and magnitude of the deviation, parameters such as the weight coefficients in the load balancing algorithm and the adjustment factors in the link training model are adjusted according to the optimization strategy. Gradient descent is used to gradually approach the optimal parameter configuration. After each update, system performance is evaluated. If it improves, further adjustments are made; otherwise, the parameters are rolled back, forming a closed-loop optimization mechanism.
[0042] Preferably, the step S3, training the link based on the link training and negotiation model, includes the following sub-steps: Step S31: Sending a training signal to the link of the PCIe switch board according to the link training initial parameter set to initialize the physical layer connection state of the link; Step S32: receiving the signal quality parameter fed back by the link, and judging whether the link meets the signal quality requirement for data transmission according to a preset signal quality evaluation standard; Step S33: If the requirements are not met, adjust the link training parameters according to the link training and negotiation model, and send the training signal again until the link meets the signal quality requirements for data transmission.
[0043] Specifically, in step S3, the link training process begins by sending an initial training signal to establish a physical layer connection, including operations such as clock synchronization and signal strength calibration. Feedback signal quality parameters are then received and evaluated against pre-set standards to determine whether metrics such as bit error rate and eye opening meet the standards. If not, training parameters such as signal level and equalization settings are adjusted, and the training sequence is sent again, forming an iterative optimization process. During implementation, multiple rounds of training are performed according to the pre-set training protocol, and the optimal parameter combination is stored after each round until all links meet transmission requirements, ensuring stable and reliable physical layer connections and providing a foundation for the operation of higher-layer protocols.
[0044] Preferably, the step S4 of constructing a data transmission scheduling scheme includes the following sub-steps: Step S41: Analyze the service traffic model of the PCIe switch board to determine the data traffic characteristics and service priorities of different service types; Step S42: combining the load distribution weight of each link and the service priority, classifying the data according to service type and priority; Step S43: Allocate transmission links for different categories of data according to the classification results and the bandwidth and delay parameters of the links, and determine the transmission order of the data on each link.
[0045] Specifically, in step S4, the scheduling scheme construction process first analyzes service traffic characteristics, identifying different types of services, such as real-time services and batch data services, and determining their priorities and traffic patterns. Then, based on link load weights, data is classified by service type and priority, forming different queues. Finally, based on performance parameters such as link bandwidth and latency, transmission links are allocated to each queue and the transmission order is determined, with high-priority queues receiving preferentially high-quality link resources. During implementation, a multi-level scheduling algorithm is employed, first classifying services, then performing link mapping, and finally generating a scheduling table to ensure orderly and efficient data transmission.
[0046] Preferably, the step S5 of real-time monitoring and adjusting the power consumption parameters of the PCIe switch board includes the following sub-steps: Step S51: collecting power consumption parameters of the PCle switch board in real time through the power consumption monitoring module on the PCle switch board; Step S52: Compare the collected power consumption parameters with a preset power consumption threshold to determine whether the power consumption exceeds the threshold; Step S53: If the power consumption exceeds the standard, first try to reduce the data transmission rate of non-critical services and observe the changes in power consumption. If the power consumption still exceeds the standard, further adjust the link status and switch some links to low power mode.
[0047] Specifically, in step S5, the power consumption adjustment process collects system power consumption data in real time through the power consumption monitoring module, and compares it with the preset threshold to determine whether it exceeds the standard; if it exceeds the standard, the hierarchical adjustment strategy is activated, first reducing the non-critical business rate and observing the power consumption changes; if it still does not meet the standard, some low-utilization links are further closed or switched to low-power mode; after each adjustment, an observation window is set to evaluate the adjustment effect. If the power consumption drops, the current configuration is maintained, otherwise the adjustment is continued until the power consumption is controlled within the threshold range, achieving a dynamic balance between performance and power consumption.
[0048] like Figure 2 As shown in FIG, a performance and power consumption collaborative optimization method for a PCIe switch board is implemented through the following different units, including: A load feature construction unit is used to construct a dynamic load feature vector based on the current working status and historical load data of the PCIe switch board; a load weight calculation unit, connected to the load feature construction unit, for processing the feature vector using a preset dynamic load balancing algorithm to obtain a load distribution weight for each link; A link training parameter construction unit, connected to the load weight calculation unit, for constructing a link training initial parameter set based on the electrical characteristic parameters of each port of the PCle switch board and the load distribution weight; A link training execution unit, connected to the link training parameter construction unit, for training the link of the PCIe switch board based on the link training and negotiation model and the link training initial parameter set; A data transmission scheduling construction unit, connected to the link training execution unit, is used to construct a data transmission scheduling plan based on the service traffic model of the PCle switch board and the load distribution weight of each link after the link training is completed; The power consumption monitoring and adjustment unit is connected to the data transmission scheduling construction unit and is used to monitor the power consumption parameters of the PCle switch board in real time when data is transmitted on the PCle switch board according to the data transmission scheduling scheme, and when the power consumption parameters exceed the preset power consumption threshold, reduce the power consumption of the PCle switch board by adjusting the data transmission rate and link status.
[0049] This method for co-optimizing performance and power consumption for PCIe switch modules addresses the issue of irrational link load distribution by abandoning traditional static allocation. Instead, it captures the current operating status and historical load data of the PCIe switch module in real time, dynamically constructs a load signature vector, and then applies a load balancing algorithm to scientifically calculate the load distribution weights for each link. Furthermore, it combines service traffic models with link performance parameters to construct a data transmission scheduling scheme, ensuring on-demand data distribution to the appropriate link. Whether it's high-priority services or high-volume data transmission, it finds the optimal transmission path, effectively avoiding load imbalance between links and significantly improving data transmission efficiency and overall PCIe switch module performance.
[0050] To address the shortcomings of traditional fixed power consumption control strategies, this optimization method establishes a dynamic power consumption management system. Leveraging the power consumption monitoring module on the PCIe switch board, power consumption data is collected in real time and compared against preset thresholds. If power consumption exceeds the threshold, the data transmission rate for non-critical services is prioritized and reduced. If this still fails to meet the requirements, some links are switched to low-power mode. During the link training phase, an initial parameter set is constructed based on the port electrical characteristics and load weights. During training, these parameters are adjusted in real time based on signal quality to ensure that the link operates efficiently and effectively, achieving a dynamic balance between performance and power consumption.
[0051] Furthermore, this optimization method is self-evolving. By continuously monitoring the performance and power consumption of the PCIe switch board during data transmission, it optimizes and updates the parameters of the dynamic load balancing algorithm and link training and negotiation model. This allows the optimization solution to adapt to changing application scenarios and business needs. Over time, as operational data accumulates, it continuously improves the performance and power management of the PCIe switch board, maintaining efficient and energy-efficient operation.
[0052] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "disposed," "installed," "connected," "connected," and "fixed" should be understood in a broad sense. For example, they may refer to fixed connections, detachable connections, or integral connections; they may refer to mechanical connections or electrical connections; they may refer to direct connections or indirect connections through an intermediate medium; and they may refer to internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.
[0053] While embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.
Claims
1. A performance and power consumption collaborative optimization method for a PCIe switch board, characterized in that: The method includes: Step S1: Based on the current working status and historical load data of the PCIe switch board, a dynamic load feature vector is constructed, and the feature vector is processed using a preset dynamic load balancing algorithm to obtain the load distribution weight of each link; Step S2: constructing a link training initial parameter set based on the electrical characteristic parameters of each port of the PCle switch board and the load distribution weight; Step S3: Based on the link training and negotiation model, the link of the PCIe switch board is trained according to the link training initial parameter set. During the training process, the signal quality parameters of the link are dynamically monitored and the link training parameters are adjusted according to the signal quality parameters until the link training is completed. Step S4: After link training is completed, a data transmission scheduling plan is constructed based on the service traffic model of the PCIe switch board and the load distribution weight of each link; Step S5: performing data transmission on the PCIe switch board according to the data transmission scheduling scheme, while simultaneously monitoring the power consumption parameters of the PCIe switch board in real time. When the power consumption parameters exceed a preset power consumption threshold, the power consumption of the PCIe switch board is reduced by adjusting the data transmission rate and link status. Step S6: Optimize and update the parameters of the dynamic load balancing algorithm and the link training and negotiation model according to the performance indicators and power consumption parameters of the PCle switch board during data transmission.
2. The performance and power consumption collaborative optimization method for a PCIe switch board according to claim 1 is characterized in that: The dynamic load balancing algorithm uses the following model formula to calculate the load distribution weight of each link: ； in, For the The load distribution weight of the links, For the The historical load average of the links, For the The current number of connections of the link, is the total number of links, The weight adjustment coefficient is used to adjust the impact of the historical load average and the current number of connections on the load distribution weight.
3. The performance and power consumption collaborative optimization method for a PCIe switch board according to claim 1 is characterized in that: In the link training and negotiation model, the following model formula is used to adjust the link training parameters: ； in, is the adjusted link training parameter, is the link training parameter before adjustment, is the adjustment factor, is the signal quality parameter of the current link, is the preset signal quality target parameter.
4. The method for co-optimizing performance and power consumption of a PCIe switch board according to claim 1, wherein: The data transmission scheduling scheme is constructed by using the following model formula to determine the transmission order of data on each link: ； in, For the The priority of a data block in the link transmission order, For the The business urgency of each data block, For the The data size of each data block, is the total number of data blocks, It is the priority adjustment coefficient, which is used to adjust the impact of business urgency and data size on the transmission sequence priority.
5. The method for collaboratively optimizing performance and power consumption of a PCIe switch board according to claim 1, wherein: When the power consumption parameter exceeds the preset power consumption threshold, the data transmission rate is adjusted using the following model formula: ； in, is the adjusted data transmission rate, is the data transmission rate before adjustment, is the rate adjustment coefficient, is the power consumption parameter of the current PCle switch board, is the preset power consumption threshold, The maximum allowable power consumption of the PCIe switch board.
6. The method for co-optimizing performance and power consumption of a PCIe switch board according to claim 1, characterized in that: When optimizing and updating the parameters of the dynamic load balancing algorithm and the link training and negotiation models, the following model formula is used: ； in, To optimize the updated model parameters, To optimize the model parameters before updating, is the update coefficient, is the target power consumption, is the current power consumption, is the target performance indicator, is the current performance indicator.
7. The method for co-optimizing performance and power consumption of a PCIe switch board according to claim 1, wherein: Step S3, training the link based on the link training and negotiation model, includes the following sub-steps: Step S31: Sending a training signal to the link of the PCIe switch board according to the link training initial parameter set to initialize the physical layer connection state of the link; Step S32: receiving the signal quality parameter fed back by the link, and judging whether the link meets the signal quality requirement for data transmission according to a preset signal quality evaluation standard; Step S33: If the requirements are not met, adjust the link training parameters according to the link training and negotiation model, and send the training signal again until the link meets the signal quality requirements for data transmission.
8. The method for collaboratively optimizing performance and power consumption of a PCIe switch board according to claim 1, wherein: The step S4, constructing a data transmission scheduling scheme, includes the following sub-steps: Step S41: Analyze the service traffic model of the PCIe switch board to determine the data traffic characteristics and service priorities of different service types; Step S42: combining the load distribution weight of each link and the service priority, classifying the data according to service type and priority; Step S43: Allocate transmission links for different categories of data according to the classification results and the bandwidth and delay parameters of the links, and determine the transmission order of the data on each link.
9. The method for collaboratively optimizing performance and power consumption of a PCIe switch board according to claim 1, wherein: The step S5 of real-time monitoring and adjusting the power consumption parameters of the PCIe switch board includes the following sub-steps: Step S51: collecting power consumption parameters of the PCle switch board in real time through the power consumption monitoring module on the PCle switch board; Step S52: Compare the collected power consumption parameters with a preset power consumption threshold to determine whether the power consumption exceeds the threshold; Step S53: If the power consumption exceeds the standard, first try to reduce the data transmission rate of non-critical services and observe the changes in power consumption. If the power consumption still exceeds the standard, further adjust the link status and switch some links to low power mode.
10. The method for collaboratively optimizing performance and power consumption of a PCIe switch board according to any one of claims 1 to 9, characterized in that: This method is implemented through the following different units: include: A load feature construction unit is used to construct a dynamic load feature vector based on the current working status and historical load data of the PCIe switch board; a load weight calculation unit, connected to the load feature construction unit, for processing the feature vector using a preset dynamic load balancing algorithm to obtain a load distribution weight for each link; A link training parameter construction unit, connected to the load weight calculation unit, for constructing a link training initial parameter set based on the electrical characteristic parameters of each port of the PCle switch board and the load distribution weight; A link training execution unit, connected to the link training parameter construction unit, for training the link of the PCIe switch board based on the link training and negotiation model and the link training initial parameter set; A data transmission scheduling construction unit, connected to the link training execution unit, is used to construct a data transmission scheduling plan based on the service traffic model of the PCle switch board and the load distribution weight of each link after the link training is completed; The power consumption monitoring and adjustment unit is connected to the data transmission scheduling construction unit and is used to monitor the power consumption parameters of the PCle switch board in real time when data is transmitted on the PCle switch board according to the data transmission scheduling scheme, and when the power consumption parameters exceed the preset power consumption threshold, reduce the power consumption of the PCle switch board by adjusting the data transmission rate and link status.