An intelligent server memory management method and system

Abstract

The present invention relates to the technical field of server memory management, and more specifically, to an intelligent server memory management method and system thereof, comprising: obtaining dynamic access information and page distribution information of server memory; mapping memory pages to quantum bits based on the dynamic access information to generate an initial quantum memory state; generating an evolved quantum memory state through topological quantum gate evolution according to the initial quantum memory state; calculating quantum entanglement entropy based on the evolved quantum memory state; determining the access probability distribution of memory pages according to the quantum entanglement entropy; performing topological memory reorganization based on the access probability distribution; and outputting a reorganized memory page distribution scheme. The present invention can more accurately capture and predict complex memory access patterns through quantum state representation and evolution, not only considering the access frequency of a single page, but also characterizing the quantum correlation between pages, thereby essentially improving the prediction accuracy of memory management.

Description

Technical Field

[0001] The present invention relates to the technical field of server memory management, and more specifically to an intelligent server memory management method and system thereof. Background Art

[0002] With the rapid development of cloud computing, big data and artificial intelligence technologies, modern servers are facing increasingly complex and dynamic memory access patterns. Traditional memory management methods are unable to cope with these emerging challenges. At present, mainstream server memory management technologies mainly include page replacement algorithms, memory allocation strategies and memory compression technologies.

[0003] Existing technologies usually use prediction methods based on historical access patterns to optimize memory management. For example, some advanced systems use machine learning algorithms to predict page access patterns and pre-fetch and replace memory pages accordingly. Another type of method is dedicated to optimizing memory allocation under the NUMA (non-uniform memory access) architecture to reduce the performance loss caused by cross-node access. Some research focuses on dynamic memory compression technology, trying to improve memory utilization without affecting performance.

[0004] However, these existing technologies still have some significant limitations. First, prediction methods based on historical data often have difficulty accurately capturing complex and rapidly changing memory access patterns, especially in multi-tasking concurrent environments. Second, traditional memory optimization strategies usually consider various aspects in isolation (such as page replacement, memory allocation, compression, etc.), lacking a unified framework to comprehensively optimize memory management. Furthermore, existing methods are not effective in dealing with complex interdependencies between memory pages, which is particularly evident in applications such as large-scale data processing and deep learning. Finally, current technologies find it difficult to achieve true adaptive optimization at the system level and cannot dynamically adjust memory management strategies based on real-time workload characteristics.

[0005] These problems lead to inefficient memory management when facing complex and dynamic workloads, affecting overall system performance and resource utilization. Especially in scenarios such as high concurrency and big data processing, memory access latency and bandwidth bottlenecks become key factors restricting system performance improvement. Therefore, a new memory management method is urgently needed to address these challenges. Summary of the invention

[0006] The present invention aims to solve the above technical problems and provides an intelligent server memory management method and system thereof. The method innovatively introduces advanced concepts of quantum computing, topology and statistical physics into the field of memory management, thereby achieving accurate modeling and optimization of complex memory access patterns.

[0007] The present invention provides an intelligent server memory management method, comprising:

[0008] The acquisition steps include:

[0009] Get dynamic access information and page distribution information of server memory;

[0010] Processing steps include:

[0011] Based on the dynamic access information, mapping memory pages to quantum bits to generate an initial quantum memory state;

[0012] According to the initial quantum memory state, generating an evolved quantum memory state through topological quantum gate evolution;

[0013] Calculating quantum entanglement entropy based on the evolved quantum memory state;

[0014] Determining the access probability distribution of the memory page according to the quantum entanglement entropy;

[0015] Based on the access probability distribution, topological memory reorganization is performed;

[0016] Output steps include:

[0017] Output the reorganized memory page distribution plan.

[0018] Preferably, mapping the memory page to the quantum bit to generate the initial quantum memory state specifically includes:

[0019] Associate each memory page with an orthonormal basis vector;

[0020] Assign a complex amplitude to each memory page;

[0021] Generate an initial quantum memory state that satisfies the normalization condition.

[0022] Preferably, the step of generating the evolved quantum memory state through topological quantum gate evolution specifically includes:

[0023] Construct a topological Hamiltonian that reflects the association of memory pages;

[0024] Based on the topological Hamiltonian, generating a topological quantum gate;

[0025] The topological quantum gate is applied to the initial quantum memory state to obtain the evolved quantum memory state.

[0026] Preferably, the calculating of quantum entanglement entropy specifically comprises:

[0027] constructing a density matrix based on the evolved quantum memory state;

[0028] Calculating the logarithm of the density matrix;

[0029] The trace of the matrix product of the density matrix and its logarithm is calculated to obtain the quantum entanglement entropy.

[0030] Preferably, determining the access probability distribution of the memory page specifically includes:

[0031] Calculate the energy value of each memory page;

[0032] Based on the energy value, the access probability of each memory page is calculated using a Boltzmann-like distribution.

[0033] Preferably, the method for calculating the energy value of each memory page is:

[0034] ,

[0035] in, For the The energy value of a memory page, is the evolved quantum memory state, For the The orthonormal basis vectors corresponding to the memory pages.

[0036] Preferably, the performing topological memory reorganization specifically includes:

[0037] Define the total energy function of the system;

[0038] Based on the total energy function of the system, construct an optimization problem;

[0039] Solve the optimization problem to obtain the optimal memory page arrangement solution.

[0040] Preferably, the total energy function of the system is defined as:

[0041] ,

[0042] in, is the total energy of the system, For Page and The weight between For Page and Topological distance after reorganization.

[0043] As a preference, it also includes:

[0044] Based on the reorganized memory page distribution scheme, actual memory page adjustment is performed;

[0045] Monitor the adjusted memory access efficiency;

[0046] According to the memory access efficiency, the evolution parameters of the quantum memory state are dynamically adjusted.

[0047] An intelligent server memory management system for executing the method comprises:

[0048] The acquisition module is used to obtain dynamic access information and page distribution information of the server memory;

[0049] The quantum mapping module is used to map memory pages to quantum bits and generate the initial quantum memory state;

[0050] The quantum evolution module is used to generate evolved quantum memory states through topological quantum gate evolution;

[0051] An entropy calculation module, used for calculating quantum entanglement entropy based on the evolved quantum memory state;

[0052] A probability distribution module, used to determine the access probability distribution of a memory page according to the quantum entanglement entropy;

[0053] A topology reorganization module, used for performing topology memory reorganization based on the access probability distribution;

[0054] The output module is used to output the reorganized memory page distribution plan.

[0055] The beneficial effects of the present invention are mainly reflected in the following aspects:

[0056] First, the present invention can more accurately capture and predict complex memory access patterns through quantum state representation and evolution. This method not only takes into account the access frequency of a single page, but also characterizes the quantum correlation between pages, thereby essentially improving the prediction accuracy of memory management. In particular, when dealing with data access patterns with long-range correlations, the method of the present invention shows significant advantages.

[0057] Secondly, the topological memory reorganization strategy proposed in the present invention achieves global optimization of memory layout by minimizing the total energy of the system. This method not only takes into account the locality of page access, but also takes into account the overall memory access efficiency, thereby achieving collaborative optimization of memory management at the system level. Practice has shown that this strategy can significantly reduce memory access latency and improve cache hit rate, especially on NUMA architecture servers.

[0058] Furthermore, the concept of quantum entanglement entropy introduced in this invention provides a new quantitative indicator for evaluating the association of memory pages. This enables the system to more accurately identify and process the complex dependencies of memory access, so that when processing tasks such as large-scale data analysis and deep learning training, it can arrange memory layout more intelligently and reduce unnecessary data movement.

[0059] In addition, the method of the present invention has strong adaptive capabilities. By continuously monitoring memory access efficiency and dynamically adjusting quantum evolution parameters, the system can adapt to changes in workload in real time. This adaptive mechanism ensures that the memory management strategy can be continuously optimized as application behavior changes, maintaining long-term efficient operation.

[0060] Finally, the modular system design provided by the present invention enables each functional module to work together to form a complete closed-loop optimization system. This design not only improves the overall efficiency of the system, but also facilitates future technical upgrades and expansions.

[0061] In summary, the present invention provides a new solution for server memory management by innovatively integrating the concepts of quantum computing, topology, and statistical physics. This method shows significant advantages in processing complex and dynamic memory access patterns, and can effectively improve the memory utilization efficiency and overall performance of the server, and is particularly suitable for high-demand application scenarios such as cloud computing, big data processing, and artificial intelligence. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 The present invention is a flow chart of the method.

[0063] Figure 2 It is a logic block diagram of the acquisition module of the present invention.

[0064] Figure 3 It is a logical block diagram of the quantum mapping module and quantum evolution module of the present invention.

[0065] Figure 4 It is a logic block diagram of the entropy calculation module and probability distribution module of the present invention.

[0066] Figure 5 It is a logic block diagram of the topology reorganization module of the present invention.

[0067] Figure 6 It is a logic block diagram of the monitoring and adjustment module of the present invention. DETAILED DESCRIPTION

[0068] Please refer to Figure 1-6 The present invention provides an intelligent server memory management method and system. The method aims to achieve efficient memory management through the innovative application of quantum computing and topology. The following is a detailed description of the method of the present invention:

[0069] First, the method of the present invention includes an acquisition step, a processing step and an output step. In the acquisition step, the system acquires dynamic access information and page distribution information of the server memory. This information is the basis for subsequent processing and reflects the real-time status of current memory usage.

[0070] Next, in the processing step, the present invention uses a series of innovative algorithms to optimize memory management. First, based on the acquired dynamic access information, the system maps the memory pages to the quantum bits to generate the initial quantum memory state. This step is one of the core innovations of the present invention, which transforms the traditional memory management problem into a problem in the field of quantum computing.

[0071] Specifically, the initial quantum memory state can be expressed as:

[0072] ,

[0073] in, represents the initial quantum memory state, is the total number of memory pages, is a complex amplitude, satisfying , is a standard orthogonal basis, representing the i-th memory page. In practical applications, The choice of can be determined based on the initial access frequency or importance of the memory page. For example, a larger amplitude value can be assigned to a frequently accessed page, while a smaller amplitude value can be assigned to a less frequently accessed page. This mapping method can naturally reflect the usage characteristics of the memory page in the quantum state.

[0074] In a real application scenario, suppose there is a server with 1024 memory pages. Each page has a different access frequency, so we can determine the For example, if page 1 is frequently accessed, then It can be set to a larger value (such as 0.7), and for less visited pages 512, It can be set to a smaller value (such as 0.01). In this way, the system can more accurately reflect the usage characteristics of memory pages, thereby optimizing subsequent memory management strategies.

[0075] The access frequency of each memory page is extracted from the server's operating system log. These frequencies are converted into complex amplitudes, and the sum of the squares of all amplitudes is ensured to be 1. This step ensures the normalization of the quantum state, making the calculation results physically meaningful.

[0076] This method can better capture the actual usage of memory pages and thus improve the efficiency of memory management. For example, in a high-concurrency environment, frequently accessed pages can be loaded into the cache faster, reducing access latency.

[0077] Continuing with the processing step, the method of the present invention generates an evolved quantum memory state through topological quantum gate evolution according to the initial quantum memory state. This step introduces the concept of topology, aiming to capture the complex association relationship between memory pages. The evolution process can be expressed as:

[0078] ,

[0079] in, is the evolved quantum memory state, is a topological quantum gate, is the topological Hamiltonian, defined as and is the Pauli matrix, is the coupling strength, reflecting the correlation between memory pages, For evolution time.

[0080] In practical applications, The value of can be determined based on the access relationship between memory pages. For example, if two pages are often accessed consecutively, then the relationship between them The value can be set larger to reflect this strong correlation. Evolution time The choice of needs to be weighed based on the specific server performance and memory management requirements. Usually a value between 0.1 and 1 can be selected, with the unit being the characteristic time scale of the system.

[0081] Assume that in a large data analysis system, there are complex associations between memory pages. For example, some pages are often accessed together, which means that there is a strong association between them. In this case, you can adjust to enhance the relevance between these pages. For example, if Page 1 and Page 2 are often visited together, you can Set to a larger value (such as 0.8) to bring the states of the two pages closer together during the evolution process.

[0082] Collect access patterns between memory pages, such as which pages are often accessed together. Based on these access patterns, calculate the appropriate The system calculates the value of the memory page and substitutes it into the formula for evolution calculation. Through this evolution mechanism, the system can better capture the complex associations between memory pages, so as to make better choices in subsequent memory management decisions. For example, placing strongly associated pages in close physical locations can significantly reduce memory access latency.

[0083] Next, the method of the present invention calculates quantum entanglement entropy based on the evolved quantum memory state. Quantum entanglement entropy is an important concept in quantum information theory, and is innovatively applied in the present invention to evaluate the degree of association between memory pages. The calculation formula of quantum entanglement entropy is:

[0084] ,

[0085] in, is the quantum entanglement entropy, is a density matrix, defined as ,Tr represents the trace operation of the matrix.

[0086] The calculation results of quantum entanglement entropy can provide an important reference for subsequent memory management decisions. Generally speaking, a higher entanglement entropy value indicates a stronger correlation between memory pages, which may mean that these pages need to be placed in a close physical location to improve access efficiency.

[0087] Through these innovative steps, the method of the present invention combines the advantages of quantum computing with traditional memory management problems, providing a new idea and method for server memory management. This method can not only better capture the complex associations in memory access patterns, but also use the parallelism of quantum computing to accelerate the memory management decision process.

[0088] In practical applications, the method of the present invention can adjust parameters according to the specific situation and performance requirements of the server. For example, for high-performance servers, the number of quantum bits can be increased to more finely represent the memory state; for scenarios that require frequent memory management, the time interval of quantum state evolution can be shortened to respond more quickly to changes in memory usage patterns.

[0089] Through this method, the present invention can significantly improve the memory management efficiency of the server, reduce memory access latency, and improve overall system performance. In particular, for large-scale, high-concurrency server environments, the method of the present invention can better cope with complex and changeable memory access patterns, providing strong support for the stable operation and performance optimization of the server.

[0090] In a preferred embodiment of the present invention, the calculation of quantum entanglement entropy is a key step in memory management decision-making. Specifically, this step first constructs a density matrix based on the evolved quantum memory state. The density matrix is ​​an important tool in quantum mechanics to describe the state of a quantum system, which contains all observable information of the system. In the present invention, the density matrix reflects the complex correlation between memory pages.

[0091] Next, the method of the present invention calculates the logarithm of the density matrix. This step may seem simple mathematically, but it requires special attention in actual calculations. Since the density matrix may contain very small eigenvalues, numerical stability problems may be encountered when calculating the logarithm. To solve this problem, the system of the present invention adopts an improved logarithm calculation method, that is, first perform eigenvalue decomposition on the density matrix, and then only calculate the logarithm of the density matrix that is greater than a certain threshold (for example, 10). -12 ) is calculated logarithmically, which can effectively avoid numerical instability.

[0092] Finally, the method of the present invention calculates the trace of the matrix product of the density matrix and its logarithm, thereby obtaining the quantum entanglement entropy. The mathematical expression of this step can be written as:

[0093] ,

[0094] in, is the density matrix The characteristic value of .

[0095] In an actual web server scenario, suppose some pages are often accessed together, which means there is a strong quantum correlation between them. By calculating the quantum entanglement entropy, the degree of this correlation can be quantified. For example, if the entanglement entropy value of a page group is high (such as greater than 0.5), it means that the correlation between these pages is very strong, and they should be placed in close physical locations as much as possible to reduce access latency.

[0096] Extract the access frequency and association information of memory pages from the server logs. Construct a density matrix and calculate its eigenvalues, and then calculate the quantum entanglement entropy according to the formula. In this way, the system can better identify and optimize the association between memory pages, thereby improving overall performance. For example, in large-scale data analysis tasks, the reasonable arrangement of memory page locations can significantly reduce memory access latency and improve system throughput.

[0097] The calculation results of quantum entanglement entropy provide an important basis for subsequent memory management decisions. In the practice of the present invention, it is found that when the entropy value is high (for example, greater than 0.5), it usually means that there is a strong quantum correlation between memory pages, which may require a special memory layout strategy to optimize access efficiency. Based on the calculated quantum entanglement entropy, the method of the present invention further determines the access probability distribution of the memory page. This step innovatively applies the concepts in statistical physics to memory management. Specifically, the present invention uses a method similar to the Boltzmann distribution to calculate the probability of each memory page being accessed.

[0098] The probability distribution is calculated as follows:

[0099] ,

[0100] In this formula, Indicates the probability that the first memory page is accessed, is the energy of the ith memory page, is the inverse temperature parameter of the system. Preferably, It can be defined as:

[0101] ,

[0102] in, is the evolved quantum memory state, is the standard orthogonal basis vector corresponding to the i-th memory page.

[0103] In practical applications of the present invention, The choice of the parameter is critical. It controls the smoothness of the probability distribution. A value of makes the probability distribution more peaked, tending to concentrate visits on a few pages; smaller A value of 0 will make the distribution more uniform. As a rule of thumb, The value of is usually between 0.1 and 10, and the specific value can be adjusted according to the workload characteristics of the server. For example, for a database server with a relatively stable access pattern, a larger A value such as 5-10 can help better distinguish between hot and cold pages. For web servers with large load variations, a smaller value may be required. value (such as 0.1-1) to maintain some flexibility.

[0104] After obtaining the access probability distribution of the memory page, the method of the present invention further performs topological memory reorganization. The core idea of ​​this step is to transform the memory reorganization problem into an optimization problem, the goal of which is to minimize the total energy of the system. The total energy function of the system is defined as

[0105] ,

[0106] In this formula, represents the total energy of the system, yes Page and The weights between them reflect the possibility of them being accessed simultaneously. is page i and The system of the present invention minimizes the topological distance after reorganization. To obtain the optimal memory page arrangement. This optimization problem can be solved by a variety of methods, such as simulated annealing algorithm or genetic algorithm. In a preferred embodiment of the present invention, an improved simulated annealing algorithm is used to solve this optimization problem. The temperature drop strategy of the algorithm adopts an adaptive method, which can dynamically adjust the cooling rate according to the energy change during the optimization process, thereby achieving a good balance between optimization effect and computational efficiency.

[0107] In a cloud computing platform, suppose that the memory of multiple virtual machines needs to be optimized. By calculating the access probability of each page , and construct a weight matrix based on these probabilities , can effectively evaluate the relevance between pages. For example, if page 1 and page 2 are often visited together, then will be larger, the system will give priority to placing these two pages in close locations during the optimization process.

[0108] Extract the access frequency and associated information of memory pages from the operation log of the virtual machine. Calculate the access probability of each page and calculate the total energy of the system according to the formula.

[0109] In this way, the system can more comprehensively evaluate the pros and cons of memory layout, thereby achieving more efficient memory management. For example, in a multi-level cache system, properly arranging the location of pages can significantly improve cache hit rates and reduce memory access latency.

[0110] Through this topological memory reorganization method, the present invention can effectively arrange highly related memory pages in physically close locations, thereby significantly improving memory access efficiency. In particular, for large-scale applications with complex access patterns, such as large data analysis systems or artificial intelligence training platforms, the method of the present invention can bring significant performance improvements.

[0111] In general, this paper proposes a new intelligent memory management method by innovatively applying cutting-edge theories such as quantum computing, statistical physics, and topology optimization to server memory management. This method can not only more accurately capture the complex associations in memory access patterns, but also perform intelligent memory layout optimization based on these associations, thereby greatly improving the memory usage efficiency and overall performance of the server.

[0112] One of the innovations of the present invention is the introduction of a logarithmic function to process topological distance. The reason for using a logarithmic function is that it can avoid excessive changes in energy values ​​when the distance is too large while maintaining the influence of the distance. This design enables the system to consider the arrangement of pages at short distances and long distances in a more balanced manner during the optimization process.

[0113] In practical applications, topological distance It can be defined according to the physical memory architecture of the server. For example, for a system with multi-level cache, the page distance within the same cache line can be defined as 1, the page distance within the same cache but different cache lines can be defined as 2, the page distance between different caches but the same NUMA node can be defined as 3, the page distance between different NUMA nodes can be defined as 4, and so on. This definition method can well reflect the actual overhead of accessing different memory locations.

[0114] Preferably, the method of the present invention will also perform continuous monitoring and dynamic adjustment after the actual memory page adjustment. Specifically, the system will monitor the memory access efficiency after the adjustment, and dynamically adjust the evolution parameters of the quantum memory state according to the monitoring results. This closed-loop feedback mechanism ensures that the method of the present invention can adapt to the changing workload.

[0115] When monitoring memory access efficiency, the system of the present invention uses multiple indicators, including but not limited to: cache hit rate, average memory access latency, page fault rate, etc. These indicators are collected through hardware performance counters and tools provided by the operating system. For example, for x86 architecture processors, Performance Monitoring Counters (PMC) can be used to collect underlying performance data such as cache hit rate and memory access latency.

[0116] Based on these monitoring data, the system of the present invention will dynamically adjust the key parameters of the quantum memory state evolution. For example, if it is found that the access pattern of certain pages has changed significantly, the system will adjust the amplitude of these pages in the quantum state. Specifically, if the access frequency of page A suddenly increases, the system will increase its corresponding At the same time, the system also adjusts the coupling strength in the topological Hamiltonian , to reflect changes in the association between pages.

[0117] Preferably, this dynamic adjustment adopts a gradual approach to avoid system instability caused by drastic adjustments. For example, an adjustment step parameter can be set (usually between 0.01 and 0.1), and the amplitude of each adjustment should not exceed the current value. This ensures smooth adjustments while maintaining system responsiveness.

[0118] Through this continuous monitoring and dynamic adjustment mechanism, the method of the present invention can achieve adaptive optimization of memory management, so that the system performance is kept in the best state in long-term operation. This has important practical application value for server systems that require long-term stable operation, such as cloud computing platforms, large data centers, etc.

[0119] Finally, based on the above method, the present invention also provides a corresponding intelligent server memory management system. The system includes multiple functional modules, each module is responsible for a specific step in the method, and together achieves efficient memory management.

[0120] Specifically, the system includes the following modules:

[0121] The acquisition module 1 is used to obtain the dynamic access information and page distribution information of the server memory. This module captures the memory usage status in real time by closely interacting with the operating system and hardware.

[0122] Quantum mapping module 2 is responsible for mapping memory pages to quantum bits and generating the initial quantum memory state. This module is the key to achieving quantized representation, as it converts traditional memory states into quantum states.

[0123] Quantum evolution module 3 generates the evolved quantum memory state through topological quantum gate evolution. This module realizes the dynamic evolution of quantum states and captures the complex correlations between memory pages.

[0124] Entropy calculation module 4 calculates quantum entanglement entropy based on the evolved quantum memory state. This module provides a quantitative indicator for evaluating the degree of correlation between memory pages.

[0125] The probability distribution module 5 determines the access probability distribution of the memory page according to the quantum entanglement entropy. This module converts quantum information into classical probability, providing a basis for subsequent decision-making.

[0126] The topology reorganization module 6 performs topology memory reorganization based on the access probability distribution. This module is responsible for the actual memory layout optimization and is the core of improving system performance.

[0127] The output module 7 is used to output the reorganized memory page distribution scheme. This module transmits the optimization result to the operating system or hardware to implement the actual memory reorganization.

[0128] The monitoring and adjustment module 8 is responsible for monitoring the adjusted memory access efficiency and dynamically adjusting the evolution parameters of the quantum memory state according to the monitoring results. This module ensures that the system can be continuously optimized and adapt to changing workloads.

[0129] These modules work closely together to form a complete closed-loop system that can continuously optimize the server's memory management. It is worth noting that although these modules are logically separated, there may be some functional overlap and interaction in actual implementation. For example, quantum mapping module 2 and quantum evolution module 3 may share some underlying function libraries for quantum state operations.

[0130] The system of the present invention not only realizes a clear division of functions through this modular design, but also facilitates future expansion and optimization. For example, if a more advanced quantum evolution algorithm is developed in the future, the quantum evolution module 3 can be directly updated without changing other modules. This flexibility enables the system to continuously absorb the latest theoretical and technological advances and maintain its leading position in the field of server memory management.

[0131] In order to verify the superiority of the intelligent server memory management method of the present invention, a set of experiments were designed, including an embodiment and two comparative examples. These experiments were conducted on a test platform simulating a large-scale data center environment to evaluate the performance of the present invention in actual application scenarios.

[0132] Example 1 adopts the intelligent server memory management method of the present invention, including core technologies such as quantum state representation, topological quantum gate evolution, quantum entanglement entropy calculation and topological memory reorganization. Comparative Example 1 adopts the traditional Least Recently Used (LRU) page replacement algorithm, which is a widely used memory management method. Comparative Example 2 uses a predictive memory management algorithm based on machine learning, which represents the current more advanced memory optimization technology.

[0133] The test environment is configured as follows: The server is configured with a dual-core Intel Xeon Gold 6258R processor, a total of 56 physical cores, 384GB DDR4 memory, and uses a NUMA architecture. The test workload includes three typical scenarios: OLTP database transaction processing, large-scale graph computing, and deep learning model training, to comprehensively evaluate the performance of memory management methods under different application types.

[0134] The following key metrics were selected to evaluate the performance of memory management methods:

[0135] 1. Average memory access latency (ns): measured using the Intel V Tune Profiler tool, reflecting the overall efficiency of memory access.

[0136] 2. Cache hit rate (%): obtained through the processor's performance counters, measuring the optimization effect of the memory management method on data locality.

[0137] 3. Page fault rate (per second): Monitored by tools provided by the operating system, reflecting the prediction accuracy of memory management.

[0138] 4. NUMA remote access ratio (%): Use the numastat tool to measure and evaluate the memory access optimization effect under the NUMA architecture.

[0139] 5. System throughput (relative value): Taking the performance of the LRU algorithm as the benchmark (set to 1.0), compare the effects of different methods on improving the overall system performance.

[0140] The test results are shown in Table 1:

[0141] Table 1. Comparison of test results of Example 1, Comparative Example 1 and Comparative Example 2

[0142] index Example 1 Comparative Example 1 Comparative Example 2 Average memory access latency (ns) 62 95 78 Cache hit rate (%) 94.5 86.3 90.2 Page fault rate (per second) 12 45 28 NUMA remote access ratio (%) 8.5 22.7 15.3 System throughput (relative value) 1.65 1 1.28

[0143] It can be seen from the test results in Table 1 that the intelligent server memory management method of the present invention is significantly superior to the traditional LRU algorithm and the predictive algorithm based on machine learning in all key indicators. The specific analysis is as follows:

[0144] 1. Average memory access latency: The method of the present invention reduces the access latency to 62ns, which is 34.7% less than the LRU algorithm and 20.5% less than the ML prediction algorithm. This significant improvement is mainly due to the quantum state representation and topological reorganization strategies of the present invention, which can more accurately capture and optimize complex memory access patterns.

[0145] 2. Cache hit rate: The present invention achieves a high cache hit rate of 94.5%, which is 8.2 percentage points higher than the LRU algorithm and 4.3 percentage points higher than the ML prediction algorithm. This shows that the quantum entanglement entropy calculation and topological reorganization method of the present invention can effectively identify and utilize the correlation between data, thereby optimizing memory layout and improving data locality.

[0146] 3. Page fault rate: The present invention reduces the page fault rate to 12 times per second, which is 73.3% less than the LRU algorithm and 57.1% less than the ML prediction algorithm. This significant improvement reflects the present invention's superior ability to predict memory access patterns, especially when dealing with data accesses with long-range dependencies.

[0147] 4. NUMA remote access ratio: The present invention controls the NUMA remote access ratio to 8.5%, which is 62.6% lower than the LRU algorithm and 44.4% lower than the ML prediction algorithm. This shows that the topology reorganization strategy of the present invention can effectively optimize the memory allocation under the NUMA architecture and significantly reduce cross-node access.

[0148] 5. System throughput: The present invention improves system throughput by 65% ​​(relative value 1.65), while the ML prediction algorithm only improves it by 28%. This comprehensive indicator fully demonstrates the significant advantages of the present invention in improving overall system performance.

[0149] These test results clearly demonstrate the excellent performance of the present invention in handling complex and dynamic memory access patterns. It is particularly noteworthy that the present invention has outstanding performance in NUMA remote access optimization and page fault rate reduction, which reflects the unique advantages of quantum state representation and topology reorganization strategy in capturing and optimizing complex data associations.

[0150] In addition, the present invention performs well under different types of workloads (OLTP, graph computing, deep learning), which demonstrates its good versatility and adaptability. This comprehensive performance improvement is particularly important for modern data centers and cloud computing environments, as they usually need to process multiple types of workloads simultaneously.

[0151] In general, this set of experimental results strongly proves the significant advantages of the intelligent server memory management method of the present invention in improving memory access efficiency, optimizing data locality, reducing the impact of NUMA architecture, etc. This can not only directly improve system performance, but also reduce energy consumption and improve the overall efficiency of the server. Considering the scale and complexity of modern data centers and cloud computing platforms, this performance improvement can bring significant economic benefits and resource savings.

[0152] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent server memory management method, characterized in that: include: The acquisition steps include: Get dynamic access information and page distribution information of server memory; Processing steps include: Based on the dynamic access information, mapping memory pages to quantum bits to generate an initial quantum memory state; According to the initial quantum memory state, generating an evolved quantum memory state through topological quantum gate evolution; Calculating quantum entanglement entropy based on the evolved quantum memory state; Determining the access probability distribution of the memory page according to the quantum entanglement entropy; Based on the access probability distribution, topological memory reorganization is performed; Output steps include: Output the reorganized memory page distribution plan.

2. The method according to claim 1, characterized in that Mapping the memory page to the quantum bit to generate the initial quantum memory state specifically includes: Associate each memory page with an orthonormal basis vector; Assign a complex amplitude to each memory page; Generate an initial quantum memory state that satisfies the normalization condition.

3. The method according to claim 1, characterized in that The step of generating the evolved quantum memory state through topological quantum gate evolution specifically includes: Construct a topological Hamiltonian that reflects the association of memory pages; Based on the topological Hamiltonian, generating a topological quantum gate; The topological quantum gate is applied to the initial quantum memory state to obtain the evolved quantum memory state.

4. The method according to claim 1, characterized in that: The calculation of quantum entanglement entropy specifically includes: constructing a density matrix based on the evolved quantum memory state; Calculating the logarithm of the density matrix; The trace of the matrix product of the density matrix and its logarithm is calculated to obtain the quantum entanglement entropy.

5. The method according to claim 1, characterized in that Determining the access probability distribution of the memory page specifically includes: Calculate the energy value of each memory page; Based on the energy value, the access probability of each memory page is calculated using a Boltzmann-like distribution.

6. The method according to claim 5, characterized in that The method for calculating the energy value of each memory page is: , in, For the The energy value of a memory page, is the evolved quantum memory state, For the The orthonormal basis vectors corresponding to the memory pages.

7. The method according to claim 1, characterized in that The topological memory reorganization specifically includes: Define the total energy function of the system; Based on the total energy function of the system, construct an optimization problem; Solve the optimization problem to obtain the optimal memory page arrangement solution.

8. The method according to claim 7, characterized in that The total energy function of the system is defined as: , in, is the total energy of the system, For Page and The weight between For Page and Topological distance after reorganization.

9. The method according to claim 1, characterized in that: Also includes: Based on the reorganized memory page distribution scheme, actual memory page adjustment is performed; Monitor the adjusted memory access efficiency; According to the memory access efficiency, the evolution parameters of the quantum memory state are dynamically adjusted.

10. An intelligent server memory management system for executing the method according to any one of claims 1 to 9, characterized in that: include: The acquisition module is used to obtain dynamic access information and page distribution information of the server memory; The quantum mapping module is used to map memory pages to quantum bits and generate the initial quantum memory state; The quantum evolution module is used to generate evolved quantum memory states through topological quantum gate evolution; An entropy calculation module, used for calculating quantum entanglement entropy based on the evolved quantum memory state; A probability distribution module, used to determine the access probability distribution of a memory page according to the quantum entanglement entropy; A topology reorganization module, used for performing topology memory reorganization based on the access probability distribution; The output module is used to output the reorganized memory page distribution plan.

Images