A Performance Space Search Method, Device, Equipment and Medium for an Adjustable System

Abstract

An embodiment of the present invention discloses a method, device, equipment and medium for searching the performance space of an adjustable system. Initialize according to the value ranges of the parameters in the key parameter set to obtain the current parameter values and the current parameter change trends of the parameters; obtain the system output result of the adjustable system according to the current parameter values of the parameters; perform performance evaluation on the system output result corresponding to the current parameter values of the parameters according to a preset optimization strategy to obtain a performance evaluation result; determine the update weights of the current parameter change trends of the parameters according to the performance evaluation results, and update the current parameter change trends according to the update weights; update the current parameter values according to the updated current parameter change trends; iteratively return to the step of determining the system output result according to the parameter values, and determine the optimal performance subspace of the adjustable system according to the parameter values of the parameters when the search termination condition is satisfied. Adjusting the parameter change trends through the update weights can improve the space search efficiency and applicability.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular, to a method, device, equipment and medium for searching the performance space of an adjustable system. Background Art

[0002] With the development of artificial intelligence technology, the demand for computing power has increased sharply, and thus the complexity of the artificial intelligence processor architecture has become higher and higher. The artificial intelligence processor architecture is an adjustable system, and its performance is affected by various parameters, and the number of parameters affecting its performance increases exponentially.

[0003] In an adjustable system, the parameters do not individually affect the architecture performance, but interact with each other, resulting in a highly discretized characteristic of the optimal performance subspace. Therefore, it is increasingly difficult to meet the current computing performance requirements for the adjustable system by manually tuning the parameters in the prior art. Summary of the Invention

[0004] The present invention provides a method, device, equipment and medium for searching the performance space of an adjustable system to improve the search efficiency and applicability of the optimal performance subspace.

[0005] According to one aspect of the present invention, there is provided a method for searching the performance space of an adjustable system, the method comprising:

[0006] Performing parameter initialization according to the value ranges of the parameters in the key parameter set affecting the performance of the adjustable system to obtain the current parameter values of the parameters and the corresponding current parameter change trends;

[0007] Obtaining the system output result corresponding to the adjustable system according to the current parameter values of the parameters in the key parameter set;

[0008] Performing performance evaluation on the system output result corresponding to the current parameter value of each parameter in the key parameter set according to a preset optimization strategy to obtain a performance evaluation result;

[0009] Determining the update weights of the current parameter change trends of the parameters in the key parameter set according to the performance evaluation results, and updating the corresponding current parameter change trends according to the update weights;

[0010] Updating the current parameter values of the corresponding parameters in the key parameter set according to the updated current parameter change trends;

[0011] Returning to the step of obtaining the system output result corresponding to the adjustable system according to the current parameter values of the parameters in the key parameter set for iteration, and determining the optimal performance subspace of the adjustable system according to the parameter values of the parameters in the key parameter set when the search termination condition is satisfied.

[0012] According to another aspect of the present invention, there is provided a performance space search device for an adjustable system, the device comprising:

[0013] A parameter initialization module, configured to perform parameter initialization according to the value ranges of the parameters in the key parameter set affecting the performance of the adjustable system, so as to obtain the current parameter values of the parameters and the corresponding current parameter change trends;

[0014] A system output result determination module, configured to obtain the system output result corresponding to the adjustable system according to the current parameter values of the parameters in the key parameter set;

[0015] A performance evaluation module, configured to perform performance evaluation on the system output result corresponding to the current parameter values of the parameters in the key parameter set according to a preset optimization strategy, so as to obtain a performance evaluation result;

[0016] A parameter change trend update module, configured to determine the update weights of the current parameter change trends of the parameters in the key parameter set according to the performance evaluation results, and update the corresponding current parameter change trends according to the update weights;

[0017] A parameter value update module, configured to update the current parameter values of the corresponding parameters in the key parameter set according to the updated current parameter change trends of each parameter;

[0018] An optimal performance subspace determination module, configured to return and iterate the step of obtaining the system output result corresponding to the adjustable system according to the current parameter values of the parameters in the key parameter set, and determine the optimal performance subspace of the adjustable system according to the parameter values of the parameters in the key parameter set when the search termination condition is satisfied.

[0019] According to another aspect of the present invention, there is provided an electronic device, the electronic device comprising:

[0020] At least one processor; and

[0021] A memory communicatively connected to the at least one processor; wherein,

[0022] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the performance space search method of the adjustable system according to any embodiment of the present invention.

[0023] According to another aspect of the present invention, there is provided a computer-readable storage medium, the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the performance space search method of the adjustable system according to any embodiment of the present invention when executed.

[0024] According to another aspect of the present invention, there is provided a computer program product including a computer program which, when executed by a processor, implements the performance space search method of the adjustable system according to any embodiment of the present invention.

[0025] In the technical solution of the embodiment of the present invention, parameter initialization is performed according to the value ranges of the parameters in the key parameter set affecting the performance of the adjustable system to obtain the current parameter values of the parameters and the corresponding current parameter change trends; according to the current parameter values of the parameters in the key parameter set, the system output result corresponding to the adjustable system is obtained; performance evaluation is performed on the system output result corresponding to the current parameter value of each parameter in the key parameter set according to a preset optimization strategy to obtain a performance evaluation result; the update weights of the current parameter change trends of the parameters in the key parameter set are determined according to each performance evaluation result, and the corresponding current parameter change trends are updated according to the update weights; according to each updated current parameter change trend, the current parameter value of the corresponding parameter in the key parameter set is updated; return to the step of obtaining the system output result corresponding to the adjustable system according to the current parameter values of the parameters in the key parameter set for iteration, and according to the parameter values of the parameters in the key parameter set when the search termination condition is satisfied, determine the optimal performance subspace of the adjustable system, which solves the problem of searching for the optimal performance subspace of the adjustable system such as the processor architecture. By adjusting the update weights of the parameter change trends according to the performance evaluation results and then adjusting the parameters to search for the optimal performance subspace, the search efficiency and applicability of the optimal performance subspace can be improved, and the problem that manual parameter adjustment cannot achieve the search for the optimal performance subspace when the performance subspace has a high degree of discreteness is overcome.

[0026] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present invention, nor is it used to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. Description of the Drawings

[0027] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0028] Figure 1 is a flowchart of a performance space search method for an adjustable system according to Embodiment 1 of the present invention;

[0029] Figure 2 is a display diagram of a performance subspace according to Embodiment 1 of the present invention;

[0030] Figure 3 is a schematic flowchart of yet another performance space search method for an adjustable system provided according to Embodiment 1 of the present invention;

[0031] Figure 4 is a flowchart of a performance space search method for an adjustable system provided according to Embodiment 2 of the present invention;

[0032] Figure 5 is a schematic diagram showing the value of updating the weight when CT is greater than or equal to SBA provided according to Embodiment 2 of the present invention;

[0033] Figure 6 is a schematic diagram of the convergence result of an adjustable system provided according to Embodiment 2 of the present invention;

[0034] Figure 7 is a schematic diagram for comparing the search effects of multiple performance sub - spaces provided according to Embodiment 2 of the present invention;

[0035] Figure 8 is a schematic structural diagram of a performance space search device for an adjustable system provided according to Embodiment 3 of the present invention;

[0036] Figure 9 is a schematic structural diagram of an electronic device for implementing the performance space search method of the adjustable system according to the embodiments of the present invention. Detailed Embodiments

[0037] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0038] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above - mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described here can be implemented in an order different from those illustrated or described here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non - exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these process, method, product or device.

[0039] Embodiment 1

[0040] Figure 1 FIG. 5 is a flowchart of a method for searching the performance space of an adjustable system according to Embodiment 1 of the present invention. This embodiment is applicable to the case of searching for the optimal performance subspace of an adjustable system such as a processor architecture by parameter adjustment. This method can be executed by a performance space search device of the adjustable system. The performance space search device of the adjustable system can be implemented in the form of hardware and / or software. The performance space search device of the adjustable system can be configured in an electronic device, which can be a computer, a server, or the like.

[0041] Among them, the adjustable system Xsystem can be a hardware architecture such as a processor architecture, a model, or a server cluster. The performance space of the adjustable system is highly non-linear and complex. Its performance is affected by a large number of parameters with wide value ranges and discrete points, as well as their combinations, forming multiple discretely distributed optimal performance subspaces. Figure 2 FIG. 6 is a display diagram of a performance subspace according to Embodiment 1 of the present invention. As Figure 2 shown, the performance subspace of the adjustable system has a multi-peak and multi-valley structure. Based on the multi-peak and multi-valley nature of the performance subspace of the adjustable system, traditional performance subspace search methods or adjustment strategies relying on manual experience have limitations in dealing with the optimal performance subspace of the adjustable system. The technical solution provided by the embodiments of the present invention can solve the above problems and search for the optimal performance subspace of the adjustable system. As Figure 1 shown, the method includes:

[0042] Step 110: Initialize the parameters according to the value ranges of the parameters in the key parameter set that affects the performance of the adjustable system, and obtain the current parameter values of the parameters and the corresponding current parameter change trends.

[0043] Specifically, the adjustable system has a series of configuration parameters (System Config Parameters, SysCfgParam), and these configuration parameters can be adjusted to make the system work in the best state. In the adjustable system, there are a wide variety of configuration parameters. The configuration parameters include both key parameters that directly affect the performance of the adjustable system and some system default parameters that are irrelevant, do not affect, but must be configured.

[0044] In an embodiment of the present invention, all key parameters affecting the performance of an adjustable system can be identified based on a preset system operating state target in combination with a statistical analysis method. The key parameters and the system default parameters together constitute a key parameter set. Table 1 is an exemplary key parameter set. For example, when the adjustable system is a Register Transfer Level (RTL) simulation, taking the optimal performance in the Prefetch scenario in the hardware architecture as the optimization target of the adjustable system, the key parameters obtained through Key Impact Parameter Identification (KIPI) at this time. The key parameters and default parameters in the RTL simulation can together constitute the key parameter set shown in Table 1. Among them, the KIPI method is not specifically limited and can be implemented by means of neural network model analysis, parameter entity recognition, or manual experience analysis, etc.

[0045] For each parameter in the key parameter set, parameter initialization can be performed according to its value range to obtain the current parameter value of each parameter and the corresponding current parameter change trend. For example, the current parameter value can be randomly initialized within the value range of the parameter. The current parameter change trend can be randomly initialized according to the value range of the parameter and a preset parameter adjustment step size.

[0046] Table 1

[0047]

[0048] Optionally, parameter initialization is performed according to the value range of each parameter in the key parameter set affecting the performance of the adjustable system to obtain the current parameter value of each parameter and the corresponding current parameter change trend, including: performing random parameter initialization according to the value range of each parameter in the key parameter set affecting the performance of the adjustable system to obtain the current parameter value of each parameter; obtaining the current parameter change trend corresponding to each parameter according to the value range of each parameter, the preset parameter adjustment step size, and a random coefficient.

[0049] For example, the value range of the parameter is , the preset parameter adjustment step size can be set to 0.1, and the random coefficient is represented by , and its value is in the range of [-1, 1]. The current parameter change trend can be expressed as the product of , 0.1, and the length of the interval . Through adaptive parameter initialization, the optimization process can flexibly handle optimization problems in different numbers of parameters and scenarios.

[0050] In the embodiments of the present invention, in order to fully consider the discreteness and volatility of the adjustable system during subspace search, an Intelligent Discrete Multi-Peak Multi-Valley Transformer (IDMMV-Transformer) can be used to capture and expand the features of the configuration parameters. Among them, the IDMMV-Transformer can be composed of various functions that can express the performance change trend.

[0051] Optionally, before obtaining the system output result corresponding to the adjustable system according to the current parameter values of the parameters in the key parameter set, it further includes: using a variety of preset functions to perform discrete multi-peak and multi-valley feature capture and expansion on the current parameter values of the parameters in the key parameter set, and obtaining the updated current parameter values of the parameters in the key parameter set after expansion.

[0052] Among them, the preset function can be a mathematical transformation function in different forms. The preset functions include but are not limited to power functions, trigonometric functions, and exponential functions, etc. Performing discrete multi-peak and multi-valley feature capture and expansion on the current parameter values through the preset function can mean capturing the relationship between the parameters through the preset function, expanding the features of the parameters, and forming multi-dimensional eigenvalue. For example, for the parameters and , feature capture and expansion can be performed through the formula to obtain the updated current parameter values of each parameter after expansion. Through discrete multi-peak and multi-valley feature capture and expansion, it is possible to better capture and express the rich features in the discrete transformation of performance, enhance the expression ability of feature data, and thus be more conducive to searching for the optimal performance subspace.

[0053] Exemplarily, the key parameter set SysCfgParam Set includes N configuration parameters SysCfgParam, and there are a total of P key parameters that affect performance. Through discrete multi-peak and multi-valley feature capture and expansion, each configuration parameter SysCfgParam can be represented as a P-dimensional vector as the eigenvalue of the configuration parameter. Among them, each eigenvalue can have a corresponding value range. For example, for the current parameter value of the configuration parameter it can be represented as , where each eigenvalue in the vector can be randomly selected from the feature values. The current parameter change trend corresponding to the configuration parameter can be represented as a P-dimensional vector, such as . Assuming that the upper and lower boundaries of the jth feature are represented by , then the initialization position of can be determined by . Among them, is a random coefficient, and its value ranges within the interval [-1, 1]. 0.1 is the preset parameter adjustment step size. is the value range of the parameter.

[0054] In the present invention, the configuration parameters in the adjustable system can be discrete values. In order to be able to process discrete input values, and at the same time avoid precision loss and conversion complexity during the processing of discrete problems, the current parameter value can be trimmed and approximated to obtain a discrete current parameter value.

[0055] Exemplarily, the trimming and approximation of the current parameter value can be through an encoding and decoding framework to place all features in a unified dimensional space, ensuring the unity and coordination of the update of each configuration parameter in each feature dimension. For example, for the trimming and approximation of the current parameter value (0.2, 1.7, 13.2), it can be through the encoding and decoding framework to place all features in a unified dimensional space to obtain the discrete current parameter value (0, 2, 13). Thus, it can be ensured that the optimal performance subspace search of the adjustable system can directly process discrete input data.

[0056] Step 120: Obtain the system output result corresponding to the adjustable system according to the current parameter values of the parameters in the key parameter set.

[0057] The adjustable system corresponds to a system output result for the parameter configuration in the key parameter set. Inputting the current parameter values of the parameters in the key parameter set into the adjustable system can obtain the corresponding system output result (SystemOutput, SysOutput).

[0058] Taking RTL simulation as an example of the adjustable system, the system output results obtained through KPTI can be the chip read and write bandwidth range (Chip Read and Write Bandwidth Peak to Peak, ChipPtp), the cluster read and write bandwidth range (Cluster Read and Write Bandwidth Peak to Peak, ClusterPtp), and the total bandwidth (TotalBandwidth, TotalBw).

[0059] Step 130: Perform performance evaluation on the system output result corresponding to the current parameter values of the parameters in the key parameter set according to the preset optimization strategy to obtain a performance evaluation result.

[0060] Among them, the preset optimization policy (System Optimization Policy, SysOptPolicy) can be determined according to the specific adjustable system and the specific performance that the user is concerned about. Exemplarily, the preset optimization policy can be the determination of one or more performance targets in the system output result, that is, Key Performance Target Identification (KPTI). In different adjustable systems or different application scenarios, the preset optimization policy will have certain differences. When facing multiple interrelated or contradictory performance targets, it is necessary to balance the weights and priorities between the targets and flexibly respond to the performance optimization targets in different scenarios. For example, for the performance of the Prefetch scenario in the RTL simulation architecture, the preset optimization policy combined with the system output result can be . Through the preset optimization policy, the performance of the system output result can be evaluated to obtain the performance evaluation result.

[0061] Step 140: Determine the updated weight of the current parameter change trend of each parameter in the key parameter set according to each performance evaluation result, and update the corresponding current parameter change trend according to the updated weight.

[0062] Among them, by determining the updated weight of the current parameter change trend of the corresponding parameter according to the performance evaluation result, the cooperation relationship between the configuration parameters can be considered, the dynamic exploration of the performance space can be realized, the search range can be adjusted according to the performance evaluation result until approaching or reaching the optimal performance subspace.

[0063] When determining the updated weight of the current parameter change trend according to the performance evaluation result, it can be that when the performance evaluation result is large, a smaller updated weight is adopted to ensure full search of the parameters around the area close to or located at the global optimal solution and ensure the refined search ability; when the performance evaluation result is small, a larger updated weight is adopted to promote the acceleration of convergence to the global optimal solution.

[0064] The current parameter change trend can be updated according to the updated weight. For example, the product of the updated weight and the current parameter change trend is used as the updated current parameter change trend. By dynamically adjusting the updated weight, and then adjusting the current parameter change trend, it is ensured that the optimal performance subspace can be searched quickly and accurately.

[0065] Step 150: Update the current parameter value of the corresponding parameter in the key parameter set according to each updated current parameter change trend.

[0066] Exemplarily, the sum of the current parameter value of the corresponding parameter in the key parameter set and the updated current parameter change trend can be used as the updated current parameter value. That is, for the parameter , the updated current parameter value is . Among them, is the current parameter value in the t-th iteration. is the current parameter change trend at the (t + 1)-th iteration.

[0067] Step 160, return to Step 120 for iteration, and determine the optimal performance subspace of the adjustable system according to the parameter values of each parameter in the key parameter set when the search termination condition is met.

[0068] To ensure that the optimal performance subspace can be searched, Steps 120 to 150 can be iterated multiple times first. Determine the optimal performance subspace of the adjustable system from the system output results obtained from multiple iterations. For example, the search termination condition can be reaching a preset maximum number of iterations. When the maximum number of iterations is reached, the parameter values of each parameter in the key parameter set corresponding to the maximum value of the system output result can be used as the optimal performance subspace of the adjustable system.

[0069] In an alternative embodiment of the embodiment of the present invention, determining the optimal performance subspace of the adjustable system according to the parameter values of each parameter in the key parameter set when the search termination condition is met includes: when the maximum number of iterations is reached, determining whether the global best performance evaluation result of each parameter in the key parameter set during the iteration and / or the global best system output result during the iteration meet the preset acceptance condition; when the global best performance evaluation result and / or the global best system output result meet the preset acceptance condition, obtain the optimal performance subspace of the adjustable system according to the parameter values of each parameter in the key parameter set corresponding to the global best performance evaluation result and / or the global best system output result.

[0070] Among them, the preset acceptance condition can be the minimum requirement of the performance subspace. When the global best performance evaluation result and / or the global best system output result meet the preset acceptance condition, the adjustable system can be configured according to the parameter values of each parameter in the key parameter set corresponding to the global best performance evaluation result and / or the global best system output result to obtain the optimal performance subspace. If the global best performance evaluation result and / or the global best system output result do not meet the preset acceptance condition, then Step 110 can be returned to start a new round of iteration to search for the optimal performance subspace.

[0071] Figure 3 is a flowchart showing another method for searching the performance space of an adjustable system provided in Embodiment 1 of the present invention. As Figure 3As shown, the search process for the optimal performance subspace can be as follows: perform KIPI identification on the adjustable system to obtain the key parameters. Then, form a key parameter set from the key parameters and the default parameters. For each parameter in the key parameter set, a range definition can be performed to obtain the value range. Among them, the range definition includes the maximum value, minimum value, and value calculation method of each parameter, etc. According to the value range, each parameter can be adaptively initialized to obtain the current parameter value of each parameter and the corresponding current parameter change trend. The current parameter value of each parameter and the corresponding current parameter change trend can be input into the IDMMV-Transformer for discrete multi-peak and multi-valley feature capture and expansion to obtain the eigenvalue of the current parameter value of each parameter and the corresponding current parameter change trend. The current parameter value of each parameter after expansion and update can be input into the adjustable system to obtain the system output result corresponding to the adjustable system. The system output result can be evaluated through a preset optimization strategy to obtain a performance evaluation result. Among them, the preset optimization strategy can be comprehensively generated according to multiple key performance objectives. According to the performance evaluation result, the updated weight can be determined, and then the current parameter change trend and the current parameter value can be updated to perform the performance subspace search. After the performance subspace search, it is judged whether the search termination condition is satisfied. When the search termination condition is not satisfied, the iteration will continue. When the search termination condition is satisfied, the optimal parameter value corresponding to the optimal performance evaluation result can be obtained, and then the optimal performance subspace of the adjustable system can be determined. Through the performance space search method (Intelligent Search Method of Best Performance Subspace, ISMBPS) of the adjustable system shown in Figure 3 , according to the feedback of the system output result, automatically segment the words and iteratively generate the next round of key parameter set. Through continuous iteration, the system optimization goal can be achieved, which can comprehensively improve the applicability of the optimal performance subspace searched in the complex hardware architecture and improve the search efficiency.

[0072] In the technical solution of this embodiment, by initializing parameters according to the value ranges of the parameters in the key parameter set that affects the performance of the adjustable system, the current parameter values of the parameters and the corresponding current parameter change trends are obtained; according to the current parameter values of the parameters in the key parameter set, the system output result corresponding to the adjustable system is obtained; according to the preset optimization strategy, the performance evaluation of the system output result corresponding to the current parameter values of the parameters in the key parameter set is performed to obtain the performance evaluation result; according to each performance evaluation result, the update weight of the current parameter change trend of each parameter in the key parameter set is determined, and the corresponding current parameter change trend is updated according to the update weight; according to each updated current parameter change trend, the current parameter value of the corresponding parameter in the key parameter set is updated; return to the step of obtaining the system output result corresponding to the adjustable system according to the current parameter values of the parameters in the key parameter set for iteration. When the search termination condition is satisfied, the optimal performance subspace of the adjustable system is determined according to the parameter values of the parameters in the key parameter set, which solves the problem of searching for the optimal performance subspace of the adjustable system such as the processor architecture. By adjusting the update weight of the parameter change trend according to the performance evaluation result, and then adjusting the parameters to search for the optimal performance subspace, the search efficiency and applicability of the optimal performance subspace can be improved, and the problem that manual parameter adjustment cannot achieve the search for the optimal performance subspace when the performance subspace has high discreteness is overcome.

[0073] Embodiment 2

[0074] Figure 4 FIG. is a flowchart of a method for searching the performance space of an adjustable system according to Embodiment 2 of the present invention. This embodiment further refines the above technical solution, and the technical solution in this embodiment can be combined with each optional solution in one or more of the above embodiments. As Figure 4 shown, the method includes:

[0075] Step 410: Initialize parameters according to the value ranges of the parameters in the key parameter set that affects the performance of the adjustable system, and obtain the current parameter values of the parameters and the corresponding current parameter change trends.

[0076] In an optional implementation manner of the embodiment of the present invention, initializing parameters according to the value ranges of the parameters in the key parameter set that affects the performance of the adjustable system, and obtaining the current parameter values of the parameters and the corresponding current parameter change trends includes: performing random parameter initialization according to the value ranges of the parameters in the key parameter set that affects the performance of the adjustable system to obtain the current parameter values of the parameters; obtaining the current parameter change trends corresponding to the parameters according to the value ranges of the parameters, the preset parameter adjustment step size, and the random coefficient.

[0077] Step 420: Use multiple preset functions to perform feature capture and expansion of discrete multi-peaks and multi-valleys on the current parameter values of each parameter in the key parameter set, and obtain the updated current parameter values of each parameter in the key parameter set after expansion.

[0078] Step 430: Obtain the system output result corresponding to the adjustable system according to the current parameter values of each parameter in the key parameter set.

[0079] Step 440: Perform performance evaluation on the system output result corresponding to the current parameter values of each parameter in the key parameter set according to the preset optimization strategy, and obtain the performance evaluation result.

[0080] Step 450: Determine the update weight of the current parameter change trend of each parameter in the key parameter set according to each performance evaluation result.

[0081] In an alternative embodiment, determining the update weight of the current parameter change trend of each parameter in the key parameter set according to each performance evaluation result includes: when CT is greater than or equal to SBA and the current iteration is the first-stage iteration, determining the update weight of the current parameter change trend of the target parameter according to the weight threshold, CT, SBA, GB, and SA; when CT is greater than SBA and the current iteration is the second-stage iteration, determining the update weight of the current parameter change trend of the target parameter according to the weight threshold, the current iteration number, and the maximum iteration number; where the first-stage iteration is before the second-stage iteration; when SA is less than or equal to CT and CT is less than SBA, using the lower limit of the weight threshold as the update weight of the current parameter change trend of the target parameter; when CT is less than SA, using the upper limit of the weight threshold as the update weight of the current parameter change trend of the target parameter.

[0082] Wherein, CT is the performance evaluation result corresponding to the target parameter in the key parameter set in the current iteration (CurrentSysOptTarget, CT), SBA is the average value of the historical best performance evaluation results of each parameter in the key parameter set in each iteration (SysCfgParam Set Best Average, SBA), GB is the global best performance evaluation result of each parameter in the key parameter set in the iteration (Global Best SysCfgParam Set SysOptTarget, GB), and SA is the average performance evaluation result of each parameter in the key parameter set in each iteration (SysCfgParam Set SysOptTarget Average, SA).

[0083] In the embodiments of the present invention, the update weights of the current parameter change trend can be divided into three types according to the performance evaluation results, and four determination methods are carried out under these three types. Specifically, the key parameter set can be divided into an elite class, a medium fitness class, and a poor fitness class according to the performance evaluation results. Among them, the elite class key parameter set refers to a larger performance evaluation result. For example, CT is greater than or equal to SBA. The medium fitness class key parameter set refers to a performance evaluation result at a medium level. For example, SA is less than or equal to CT, and CT is less than SBA. The poor fitness class key parameter set refers to a lower performance evaluation result. For example, CT is less than SA.

[0084] In the embodiments of the present invention, when the performance evaluation result is larger, the update weight can be smaller; when the performance evaluation result is smaller, the update weight can be larger, so as to ensure global search ability while accelerating search convergence.

[0085] In the elite class key parameter set, the determination method of the update weight can be divided into two cases, namely the first half of the iteration and the second half of the iteration. Among them, the first half of the iteration is the first-stage iteration, and the second half of the iteration is the second-stage iteration. In the first-stage iteration, a larger update weight can be determined to maintain the global search ability and avoid premature convergence of the optimization process. The update weight of the current parameter change trend corresponding to the target parameter can be determined according to the weight threshold, CT, SBA, GB, and SA when CT is greater than or equal to SBA and the current iteration is the first-stage iteration. Exemplarily, when CT is greater than or equal to SBA and the current iteration is the first-stage iteration, the value of the update weight can be taken from 1 to 1.5.

[0086] Optionally, when CT is greater than or equal to SBA and the current iteration is the first-stage iteration, determining the update weight of the current parameter change trend corresponding to the target parameter according to the weight threshold, CT, SBA, GB, and SA includes: when CT is greater than or equal to SBA and the current iteration is the first-stage iteration, through the formula Determine the update weight of the current parameter change trend corresponding to the target parameter. Among them, Is the update weight, Is the preset minimum value of the update weight, Is the preset maximum value of the update weight.

[0087] In the second-stage iteration, the update weight can be gradually reduced to ensure that the update weight does not exceed the preset maximum value of the update weight and is not lower than the preset minimum value of the update weight, so as to enhance the local refinement search ability, promote more detailed search near the potential optimal solution, and thus increase the probability of finding the global optimal solution.

[0088] Optionally, when CT is greater than SBA and the current iteration is the second stage iteration, the update weight of the current parameter change trend corresponding to the target parameter is determined according to the weight threshold, the current iteration number and the maximum iteration number, including: when CT is greater than SBA and the current iteration is the second stage iteration, the update weight of the current parameter change trend corresponding to the target parameter is determined according to the formula , determine the update weight of the current parameter change trend corresponding to the target parameter. Among them, clip is a function that limits the value range, that is, Less than When , the update weight is taken as ;exist Greater than When , the update weight is taken as ;exist Between and When , the update weight is taken as .

[0089] Figure 5 FIG. 1 is a schematic diagram of updating the weight when CT is greater than or equal to SBA according to Embodiment 2 of the present invention. Figure 5 As shown in the first iteration, since the update weight value is relatively large and changes with its own parameter value, its fitness value may decrease when exploring more solution spaces, or even no longer meet the requirements. As the iteration enters the second stage, the update weight gradually decreases and stabilizes, accelerating the convergence to the global optimal solution.

[0090] When SA is less than or equal to CT, and CT is less than SBA, the lower limit of the weight threshold is As the update weight of the current parameter change trend corresponding to the target parameter, the local search accuracy can be improved, and these parameters can be quickly converged to a better fitness value, thus becoming part of the elite key parameter set.

[0091] When CT is less than SA, the upper limit of the weight threshold is The updated weight of the current parameter change trend corresponding to the target parameter can enhance the global search capability and avoid premature convergence to the local optimal solution, which helps the parameters escape the local optimal solution and explore a wider solution space, thereby increasing the possibility of finding the global optimal solution.

[0092] In an embodiment of the present invention, update weights are determined based on current performance evaluation results and historical performance evaluation results, which can provide a strong global search capability in the early stage and significantly enhance the local fine exploration capability in the later stage of optimization, thereby comprehensively improving the applicability and efficiency of searching in complex hardware performance optimization problems.

[0093] Step 460: Update the corresponding current parameter change trend according to the update weight.

[0094] In an alternative embodiment of the embodiment of the present invention, updating the corresponding current parameter change trend according to the update weight includes: determining a first update value of the current parameter change trend corresponding to the target parameter according to the update weight and the current parameter change trend corresponding to the target parameter; determining a second update value of the current parameter change trend corresponding to the target parameter according to the individual best parameter value of the target parameter in each iteration in the key parameter set, the current parameter value of the target parameter, and its own parameter adjustment factor; determining a third update value of the current parameter change trend corresponding to the target parameter according to the global best parameter value of the target parameter in each iteration in the key parameter set, the current parameter value of the target parameter, and the population parameter adjustment factor; and updating the current parameter change trend corresponding to the target parameter according to the first update value, the second update value, and the third update value.

[0095] Wherein, the first update value may be the product of the update weight and the current parameter change trend. The second update value may be the product of the difference between the individual best parameter value and the current parameter value of the target parameter and its own parameter adjustment factor. When determining the second update value, a random coefficient may also be added. The third update value may be the product of the difference between the global best parameter value and the current parameter value of the target parameter and the population parameter adjustment factor. When determining the third update value, a random coefficient may also be added. The sum of the first update value, the second update value, and the third update value may be used as the updated current parameter change trend corresponding to the target parameter.

[0096] Exemplarily, the updated current parameter change trend can be determined by the formula where is the current parameter change trend corresponding to the parameter at the (t + 1)-th iteration; is the individual best parameter value of the parameter ; is the current parameter value of the parameter at the t-th iteration; is the own parameter adjustment factor; is the population parameter adjustment factor; is the global best parameter value in the key parameter set; is the maximum number of iterations.

[0097] When updating the current parameter change trend, one can track one's own experience through the individual best parameter value, the current parameter value of the target parameter, and one's own parameter adjustment factor; one can track the group experience through the global best parameter value, the current parameter value of the target parameter, and the group parameter adjustment factor. By updating oneself according to one's own experience and the group experience, one can adjust the parameter values based on the overall information, deeply explore the interactions between individual parameters, and the comprehensive influence of the overall key parameter set, so as to achieve a comprehensive and detailed optimization of the adjustable system, thus fundamentally solving the problem that the traditional parameter adjustment method is prone to fall into local optimum and difficult to comprehensively consider the complex interaction between parameters.

[0098] Step 470: Update the current parameter values of the parameters corresponding to the key parameter set according to the updated current parameter change trends of each.

[0099] Step 480: Clip and approximate the current parameter values of the updated parameters to obtain discrete current parameter values.

[0100] Step 490: Return to step 430 for iteration, and determine the optimal performance subspace of the adjustable system according to the parameter values of each parameter in the key parameter set when the search termination condition is met.

[0101] Optionally, determining the optimal performance subspace of the adjustable system according to the parameter values of each parameter in the key parameter set when the search termination condition is met includes: when the maximum number of iterations is reached, determining whether the global best performance evaluation result and / or the global best system output result in the iteration meet the preset acceptance conditions; when the global best performance evaluation result and / or the global best system output result meet the preset acceptance conditions, obtaining the optimal performance subspace of the adjustable system according to the parameter values of each parameter in the key parameter set corresponding to the global best performance evaluation result and / or the global best system output result.

[0102] The technical solution of the embodiment of the present invention can better represent the characteristics of parameters by capturing and expanding the characteristics of discrete multi-peaks and multi-valleys of the current parameter values of each parameter, adapt to the discreteness and volatility of the adjustable system performance space, enhance the expression ability of feature data, and capture the characteristics of discrete multi-peaks and multi-valleys in the performance space by developing a more complex data generation and enhancement pipeline, making it easier to be utilized by subsequent analysis and models. By determining the update weight according to the size of the performance evaluation result and the stage of iteration, adaptively adjusting the update weight, a flexible search mechanism and an efficient optimization strategy are generated; in the initial stage of iterative optimization, in order to quickly locate the approximate area of the global optimal solution, a larger update weight value can be adopted to enhance the global search ability; as the iteration progresses, the weight value is gradually reduced, and instead, it focuses on local fine exploration to improve the accuracy and stability of the solution; the dynamic weight adjustment strategy ensures that both efficient global search ability and fine local optimization can be maintained during the optimization process. By considering the update weight, individual parameter influence, and group parameter influence when updating the current parameter change trend, the collaborative working mode of the group can be simulated, and through the interaction and cooperation between parameters, comprehensive and detailed optimization of the complex adjustable system can be achieved, fundamentally solving the problems that the traditional system parameter tuning method is prone to falling into local optimality and difficult to comprehensively consider the complex interaction between parameters. Through the upper-layer verification of the maximum number of iterations and the preset acceptance condition, the reliability of the optimal performance subspace can be guaranteed. By trimming and approximating the current parameter values, discrete input parameter values can be processed to achieve the unification of multi-dimensional parameters, avoiding the accuracy loss or conversion complexity that may occur in traditional continuous optimization methods when dealing with discrete problems. Thus, the technical solution of the embodiment of the present invention solves the problem of searching for the optimal performance subspace of adjustable systems such as processor architectures, can improve the search efficiency and applicability of the optimal performance subspace, and overcome the problem that manual parameter tuning cannot achieve the search for the optimal performance subspace when the performance subspace has a high degree of discreteness.

[0103] Through the above-mentioned performance space search method of the adjustable system, after multiple iterations, the adjustable system reaches convergence. Figure 6 It is a schematic diagram of the convergence result of an adjustable system provided according to Embodiment 2 of the present invention. As Figure 6 shown, the adjacent local optimum close to the optimal subspace is successfully identified and approached through the performance space search method of the adjustable system, reflecting the stability and excellent performance of the performance subspace search.

[0104] The performance space search method of the adjustable system provided by the embodiment of the present invention has excellent performance compared with the manual tuning method in the prior art. Figure 7 It is a schematic diagram for comparing the search effects of multiple performance subspaces provided according to Embodiment 2 of the present invention. As Figure 7As shown, the vertical axis is TotalBw, and the horizontal axis is the ratio of the maximum to minimum chip read and write bandwidth (Chip Read and Write Bandwidth Peak to Peak Ratio, ChipPtpRatio). A higher TotalBw value and a lower ChipPtpRatio indicate better performance of the adjustable system.

[0105] Figure 7 The distribution of the blue data points corresponding to manual parameter tuning in [description] shows a convergence bottleneck when approaching the optimal performance region in the upper left corner, indicating that manual parameter tuning faces significant challenges when attempting to optimize peak performance and coverage simultaneously, and often can only achieve optimization effects for a single target. In contrast, the distribution of the orange data points obtained by tuning using the optimal performance subspace search method provided in the embodiments of the present invention generally shows superiority over manual parameter tuning in terms of peak performance. Moreover, the method of the embodiments of the present invention has successfully discovered numerous performance points that can break through the bottleneck of manual parameter tuning and found an optimization configuration scheme that can balance peak performance and coverage. This comparison result not only highlights the significant advantages of the method of the embodiments of the present invention and its necessity in practical applications, but also further emphasizes the urgent need when solving large-scale engineering optimization problems.

[0106] The performance space search method of the adjustable system provided by the embodiments of the present invention shows extremely high practicality and innovation in the field of performance optimization. It can be used to guide multiple links such as system architecture tuning and system performance evaluation, providing strong support for enhancing the market competitiveness of system products. At the same time, its high efficiency and flexibility also provide new ideas and methods for solving complex optimization problems in other fields. Especially for hardware architecture tuning, it solves problems that cannot be solved by traditional methods and alleviates the need for a large amount of investment in equipment and manpower.

[0107] Embodiment 3

[0108] Figure 8 is a schematic structural diagram of a performance space search device for an adjustable system provided according to Embodiment 3 of the present invention. As Figure 8 shown, the device includes: a parameter initialization module 810, a system output result determination module 820, a performance evaluation module 830, a parameter change trend update module 840, a parameter value update module 850, and an optimal performance subspace determination module 860. Among them:

[0109] The parameter initialization module 810 is used to perform parameter initialization according to the value ranges of the parameters in the key parameter set affecting the performance of the adjustable system to obtain the current parameter values of the parameters and the corresponding current parameter change trends;

[0110] The system output result determination module 820 is configured to obtain the system output result corresponding to the adjustable system according to the current parameter values of the parameters in the key parameter set;

[0111] The performance evaluation module 830 is configured to perform performance evaluation on the system output result corresponding to the current parameter values of the parameters in the key parameter set according to the preset optimization strategy to obtain a performance evaluation result;

[0112] The parameter change trend update module 840 is configured to determine the update weight of the current parameter change trend of each parameter in the key parameter set according to each performance evaluation result, and update the corresponding current parameter change trend according to the update weight;

[0113] The parameter value update module 850 is configured to update the current parameter value of the corresponding parameter in the key parameter set according to each updated current parameter change trend;

[0114] The optimal performance subspace determination module 860 is configured to return and perform iteration on the step of obtaining the system output result corresponding to the adjustable system according to the current parameter values of the parameters in the key parameter set, and determine the optimal performance subspace of the adjustable system according to the parameter values of the parameters in the key parameter set when the search termination condition is satisfied.

[0115] Optionally, the apparatus further includes:

[0116] The feature capture and expansion module is configured to, before obtaining the system output result corresponding to the adjustable system according to the current parameter values of the parameters in the key parameter set, perform feature capture and expansion of discrete multi-peaks and multi-valleys on the current parameter values of the parameters in the key parameter set by using a plurality of preset functions to obtain the updated current parameter values of the parameters in the key parameter set.

[0117] Optionally, the parameter change trend update module 840 includes:

[0118] The first update weight determination unit is configured to determine the update weight of the current parameter change trend corresponding to the target parameter according to the weight threshold, CT, SBA, GB, and SA when CT is greater than or equal to SBA and the current iteration is the first-stage iteration;

[0119] where CT is the performance evaluation result corresponding to the target parameter in the key parameter set in the current iteration, SBA is the average value of the historical best performance evaluation results of the parameters in the key parameter set in each iteration, GB is the global best performance evaluation result of the parameters in the key parameter set in the iteration, and SA is the average performance evaluation result of the parameters in the key parameter set in each iteration;

[0120] The second update weight determination unit is configured to determine the update weight of the current parameter change trend corresponding to the target parameter according to the weight threshold, the current iteration number, and the maximum iteration number when CT is greater than SBA and the current iteration is the second-stage iteration; wherein, the first-stage iteration is before the second-stage iteration.

[0121] The third update weight determination unit is configured to use the lower limit of the weight threshold as the update weight of the current parameter change trend corresponding to the target parameter when SA is less than or equal to CT and CT is less than SBA.

[0122] The fourth update weight determination unit is configured to use the upper limit of the weight threshold as the update weight of the current parameter change trend corresponding to the target parameter when CT is less than SA.

[0123] Optionally, the parameter change trend update module 840 includes:

[0124] The first update value determination unit is configured to determine the first update value of the current parameter change trend corresponding to the target parameter according to the update weight and the current parameter change trend corresponding to the target parameter.

[0125] The second update value determination unit is configured to determine the second update value of the current parameter change trend corresponding to the target parameter according to the individual best parameter values of the target parameter in each iteration in the key parameter set, the current parameter value of the target parameter, and its own parameter adjustment factor.

[0126] The third update value determination unit is configured to determine the third update value of the current parameter change trend corresponding to the target parameter according to the global best parameter values of the target parameter in each iteration in the key parameter set, the current parameter value of the target parameter, and the population parameter adjustment factor.

[0127] The parameter change trend update unit is configured to update the current parameter change trend corresponding to the target parameter according to the first update value, the second update value, and the third update value.

[0128] Optionally, the optimal performance subspace determination module 860 includes:

[0129] The preset acceptance condition verification unit is configured to determine whether the global best performance evaluation result of each parameter in the key parameter set in the iteration and / or the global best system output result in the iteration meet the preset acceptance condition when the maximum iteration number is reached.

[0130] The optimal performance subspace determination unit is configured to obtain the optimal performance subspace of the adjustable system according to the parameter values of each parameter in the key parameter set corresponding to the global best performance evaluation result and / or the global best system output result when the global best performance evaluation result and / or the global best system output result meet the preset acceptance condition.

[0131] Optionally, the device further includes:

[0132] A clipping and approximation module, configured to perform clipping and approximation on the current parameter values of the updated parameters after updating the current parameter values of the corresponding parameters in the key parameter set according to the change trends of the updated current parameters, so as to obtain discrete current parameter values.

[0133] Optionally, the parameter initialization module 810 includes:

[0134] A parameter value initialization unit, configured to perform random parameter initialization according to the value ranges of the parameters in the key parameter set affecting the performance of the adjustable system, so as to obtain the current parameter values of the parameters;

[0135] A parameter change trend initialization unit, configured to obtain the current parameter change trends of the parameters according to the value ranges of the parameters, a preset parameter adjustment step, and a random coefficient.

[0136] The performance space search device of the adjustable system provided by the embodiment of the present invention can execute the performance space search method of the adjustable system provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects for executing the method.

[0137] Embodiment 4

[0138] Figure 9 FIG. shows a schematic structural diagram of an electronic device 10 that can be used to implement the embodiments of the present invention. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0139] As Figure 9As shown, the electronic device 10 includes at least one processor 11 and a memory communicatively connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc. Among them, the memory stores a computer program executable by the at least one processor. The processor 11 can execute various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 into the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. The input / output (I / O) interface 15 is also connected to the bus 14.

[0140] Multiple components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disc, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0141] The processor 11 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the processor 11 include but are not limited to a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as the performance space search method of the adjustable system.

[0142] In some embodiments, the performance space search method of the adjustable system can be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as the storage unit 18. Optionally, part or all of the computer program can be loaded and / or installed onto the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the performance space search method of the adjustable system described above can be executed. Alternatively, in other embodiments, the processor 11 can be configured to execute the performance space search method of the adjustable system in any other appropriate way (for example, by means of firmware).

[0143] The various embodiments of the systems and techniques described above in this specification can be implemented in digital electronic circuitry, integrated circuit systems, field programmable gate arrays (FPGA), application specific integrated circuits (ASIC), application specific standard products (ASSP), systems-on-chip (SOC), complex programmable logic devices (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include: being implemented in one or more computer programs that may be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a special-purpose or general-purpose programmable processor that receives data and instructions from, and transmits data and instructions to, a storage system, at least one input device, and at least one output device.

[0144] The computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the computer programs, when executed by the processor, cause the functions / operations specified in the flowchart and / or block diagram to be implemented. The computer programs may execute entirely on the machine, partly on the machine, as a stand-alone software package partly on the machine and partly on a remote machine or entirely on the remote machine or server.

[0145] In the context of the present invention, a computer-readable storage medium may be a tangible medium that can contain, or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. The computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. Alternatively, the computer-readable storage medium may be a machine-readable signal medium. More specific examples of the machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0146] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and a pointing device (e.g., a mouse or a trackball) through which the user can provide input to the electronic device. Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, speech input, or tactile input).

[0147] The systems and techniques described herein can be implemented in a computing system including backend components (e.g., as a data server), or a computing system including middleware components (e.g., an application server), or a computing system including frontend components (e.g., a user computer having a graphical user interface or a web browser through which the user can interact with an implementation of the systems and techniques described herein), or a computing system including any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected to each other by digital data communication in any form or medium (e.g., a communication network). Examples of communication networks include: local area network (LAN), wide area network (WAN), blockchain network, and the Internet.

[0148] The computing system can include a client and a server. The client and the server are generally far from each other and often interact through a communication network. The relationship between the client and the server is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability existing in traditional physical hosts and VPS services.

[0149] It should be understood that various forms of the processes shown above can be used, with steps reordered, added, or deleted. The steps recited in the present invention can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved, and no limitation is imposed herein.

[0150] The above specific embodiments do not constitute a limitation to the protection scope of the present invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for searching the performance space of an adjustable system, characterized in that, Including: Performing parameter initialization according to the value ranges of the parameters in the key parameter set that affects the adjustable system performance, to obtain the current parameter values of the parameters and the corresponding current parameter change trends; wherein, the key parameter set includes key parameters that directly affect the adjustable system performance and system default parameters that are not related to performance but must be configured; Obtaining the system output result corresponding to the adjustable system according to the current parameter values of the parameters in the key parameter set; Performing performance evaluation on the system output result corresponding to the current parameter values of the parameters in the key parameter set according to a preset optimization strategy, to obtain a performance evaluation result; Determining the update weights of the current parameter change trends of the parameters in the key parameter set according to the performance evaluation results, and updating the corresponding current parameter change trends according to the update weights; Updating the current parameter values of the corresponding parameters in the key parameter set according to the updated current parameter change trends; Returning to the step of obtaining the system output result corresponding to the adjustable system according to the current parameter values of the parameters in the key parameter set for iteration, and determining the optimal performance subspace of the adjustable system according to the parameter values of the parameters in the key parameter set when the search termination condition is met; wherein, the search termination condition is reaching a preset maximum number of iterations; Determining the update weights of the current parameter change trends of the parameters in the key parameter set according to the performance evaluation results, including: dividing the key parameter set into an elite class, a medium fitness class, and a poor fitness class according to the performance evaluation results; wherein, the elite class key parameter set refers to a larger performance evaluation result, the medium fitness class key parameter set refers to a performance evaluation result at a medium level, and the poor fitness class key parameter set refers to a lower performance evaluation result; in the elite class key parameter set, in the first-stage iteration, determining that the update weight is greater than the upper limit of the weight threshold, and in the second-stage iteration, making the update weight gradually decrease within the range from the upper limit of the weight threshold to the lower limit of the weight threshold; wherein, the first-stage iteration is before the second-stage iteration; in the medium fitness class key parameter set, taking the lower limit of the weight threshold as the update weight; in the poor fitness class key parameter set, taking the upper limit of the weight threshold as the update weight.

2. The method according to claim 1, characterized in that, Before obtaining the system output result corresponding to the adjustable system according to the current parameter values of the parameters in the key parameter set, further including: Performing discrete multi-peak and multi-valley feature capture and expansion on the current parameter values of the parameters in the key parameter set by using a variety of preset functions, to obtain the updated current parameter values of the parameters in the key parameter set after expansion; 3. The method according to claim 1, characterized in that, Determining the update weights of the current parameter change trends of the parameters in the key parameter set according to the performance evaluation results, including: When CT is greater than or equal to SBA and the current iteration is the first-stage iteration, determining the update weight of the current parameter change trend corresponding to the target parameter according to the weight threshold, CT, SBA, GB, and SA; Among them, CT is the performance evaluation result corresponding to the target parameter in the current iteration in the key parameter set, SBA is the average of the historical best performance evaluation results of each parameter in each iteration in the key parameter set, GB is the global best performance evaluation result of each parameter in the iteration in the key parameter set, and SA is the average performance evaluation result of each parameter in each iteration in the key parameter set; When CT is greater than SBA and the current iteration is the second-stage iteration, determine the update weight of the current parameter change trend corresponding to the target parameter according to the weight threshold, the current iteration number, and the maximum iteration number; When SA is less than or equal to CT and CT is less than SBA, use the lower limit of the weight threshold as the update weight of the current parameter change trend corresponding to the target parameter; When CT is less than SA, use the upper limit of the weight threshold as the update weight of the current parameter change trend corresponding to the target parameter.

4. The method according to claim 1, wherein Update the corresponding current parameter change trend according to the update weight, including: Determine the first update value of the current parameter change trend corresponding to the target parameter according to the update weight and the current parameter change trend corresponding to the target parameter; Determine the second update value of the current parameter change trend corresponding to the target parameter according to the individual best parameter value of the target parameter in each iteration in the key parameter set, the current parameter value of the target parameter, and its own parameter adjustment factor; Determine the third update value of the current parameter change trend corresponding to the target parameter according to the global best parameter value of the target parameter in each iteration in the key parameter set, the current parameter value of the target parameter, and the population parameter adjustment factor; Update the current parameter change trend corresponding to the target parameter according to the first update value, the second update value, and the third update value.

5. The method according to claim 1, wherein Determine the optimal performance subspace of the adjustable system according to the parameter values of each parameter in the key parameter set when the search termination condition is satisfied, including: When the maximum iteration number is reached, determine whether the global best performance evaluation result of each parameter in the iteration in the key parameter set and / or the global best system output result in the iteration satisfy the preset acceptance condition; When the global best performance evaluation result and / or the global best system output result satisfy the preset acceptance condition, obtain the optimal performance subspace of the adjustable system according to the parameter values of each parameter in the key parameter set corresponding to the global best performance evaluation result and / or the global best system output result.

6. The method according to claim 1, wherein After updating the current parameter values of the corresponding parameters in the key parameter set according to each updated current parameter change trend, it also includes: Perform clipping and approximation processing on the updated current parameter values of each parameter to obtain discrete current parameter values.

7. The method according to claim 1, characterized in that, Initialize the parameters according to the value ranges of each parameter in the key parameter set that affect the performance of the adjustable system to obtain the current parameter values of each parameter and the corresponding current parameter change trends, including: Perform random parameter initialization according to the value ranges of each parameter in the key parameter set that affect the performance of the adjustable system to obtain the current parameter values of each parameter; Obtain the current parameter change trends of each parameter according to the value ranges of the parameters, the preset parameter adjustment step sizes, and the random coefficients.

8. A performance space search device for an adjustable system, characterized in that, Including: A parameter initialization module, configured to perform parameter initialization according to the value ranges of the parameters in the key parameter set that affects the performance of the adjustable system, to obtain the current parameter values of the parameters and the corresponding current parameter change trends; wherein, the key parameter set includes the key parameters directly affecting the performance of the adjustable system and the system default parameters that are not related to the performance but must be configured. A system output result determination module, configured to obtain the system output result corresponding to the adjustable system according to the current parameter values of the parameters in the key parameter set. A performance evaluation module, configured to perform performance evaluation on the system output results corresponding to the current parameter values of the parameters in the key parameter set according to a preset optimization strategy, to obtain a performance evaluation result. A parameter change trend update module, configured to determine the update weights of the current parameter change trends of the parameters in the key parameter set according to the performance evaluation results, and update the corresponding current parameter change trends according to the update weights. A parameter value update module, configured to update the current parameter values of the corresponding parameters in the key parameter set according to the updated current parameter change trends of each parameter. An optimal performance subspace determination module, configured to return and iterate the step of obtaining the system output result corresponding to the adjustable system according to the current parameter values of the parameters in the key parameter set, and determine the optimal performance subspace of the adjustable system according to the parameter values of the parameters in the key parameter set when the search termination condition is met; wherein, the search termination condition is reaching a preset maximum number of iterations. A parameter change trend update module, configured to divide the key parameter set into an elite class, a medium fitness class, and a poor fitness class according to the performance evaluation results; wherein, the elite class key parameter set refers to the key parameter set with a relatively large performance evaluation result, the medium fitness class key parameter set refers to the key parameter set with a medium-level performance evaluation result, and the poor fitness class key parameter set refers to the key parameter set with a relatively low performance evaluation result; in the elite class key parameter set, in the first-stage iteration, it is determined that the update weight is greater than the upper limit of the weight threshold, and in the second-stage iteration, within the range from the upper limit of the weight threshold to the lower limit of the weight threshold, the update weight is gradually decreased; wherein, the first-stage iteration is before the second-stage iteration; in the medium fitness class key parameter set, the lower limit of the weight threshold is used as the update weight; in the poor fitness class key parameter set, the upper limit of the weight threshold is used as the update weight.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor, so that the at least one processor can execute the performance space search method of the adjustable system according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the processor to implement the performance space search method of the adjustable system according to any one of claims 1-7 when executed.

Images