Performance optimization method for Ceph cache system based on persistent memory

Abstract

The invention relates to a performance optimization method for a Ceph cache system based on a persistent memory, and the method comprises the steps: enabling a Ceph storage system based on the persistent memory to adjust the related parameter configuration of a cache and the persistent memory through a parameter optimization system, feeding back performance data through performance monitoring, completing the adjustment and optimization of configuration parameters, and finally obtaining an optimal performance parameter combination. Comprising the following steps: S1, configuring a persistent memory device; step S2, configuring a Ceph OSD (On Screen Display); s3, performance monitoring and sampling are carried out, and a sample set is generated; s4, dynamically adjusting the parameter optimization system; step S5, testing and verifying; the problem of slow data storage performance of a storage system constructed by a common HDD hard disk is solved, data caching is accelerated by utilizing the characteristics of high bandwidth, high throughput and data persistence of the persistent memory, and then the overall performance of the Ceph storage system is improved by integrally optimizing the parameters of the storage system after the persistent memory device is used.

Description

Technical Field

[0001] The present invention belongs to the technical field of distributed storage optimization, and in particular relates to a distributed storage system optimized by a persistent memory device. Background Art

[0002] Distributed storage systems are widely used in mass data storage as a reliable solution. Ceph distributed storage system is one of the software, and the performance requirements for it are also increasing. Parameter tuning, as an effective performance tuning method, has also received widespread attention.

[0003] Persistent memory is a non-volatile memory technology that can retain data in the event of a power outage. Using persistent memory to optimize data storage, metadata management, logging, etc. can significantly improve the performance of the storage system, especially in scenarios with frequent writes and high I / O loads, improving the responsiveness and reliability of the storage system.

[0004] As a non-volatile storage technology, persistent memory has similar persistence to traditional disks and similar access speed to RAM. It can be configured as a block device or a memory-mapped device. In Ceph cache optimization scenarios, persistent memory can be used as an efficient and reliable storage medium.

[0005] Ceph storage systems generally use ordinary HDD mechanical hard disks to build data disks. When the storage system faces high-speed data storage, due to the read and write limitations of HDD mechanical hard disks, Ceph is inefficient when processing a large number of high-speed read and write requests. Even if SSD acceleration is used, there is still a bottleneck problem of data reading and writing under a large number of concurrent read and write requests.

[0006] The Ceph storage system uses SSD as the cache of the storage system. When faced with high-speed data read and write processing, it may not be able to meet the high-speed read and write requirements. At the same time, after using the cache settings, reasonable configuration parameter optimization is required to improve the overall performance of the system. Summary of the invention

[0007] In order to solve the above problems, the present invention aims at the problem of slow data storage performance of the storage system built with ordinary HDD hard disks, utilizes the high bandwidth, high throughput and data persistence characteristics of persistent memory to accelerate data caching, and then optimizes the storage system parameters after using persistent memory devices to improve the overall performance of the Ceph storage system.

[0008] The technical solution of the present invention is:

[0009] A method for optimizing the performance of a Ceph cache system based on persistent memory, wherein the Ceph storage system based on persistent memory adjusts cache and persistent memory related parameter configurations through a parameter optimization system, and feeds back performance data through performance monitoring to complete the tuning of configuration parameters, and finally obtains the optimal performance parameter combination, including the following steps:

[0010] Step S1, configuring a persistent memory device;

[0011] Configure the persistent memory device to file system mode; use the libpmem library to directly mount the persistent memory device as a file system;

[0012] Step S2: Configure Ceph OSD;

[0013] Specify the persistent memory device as the cache device in the Ceph configuration file, enable the relevant cache options in ceph.conf and keep the default values; create OSD and configure the cache layer through the ceph command;

[0014] Step S3: monitor performance and perform sampling to generate a sample set;

[0015] After deployment, continuously monitor Ceph's performance indicators, record the initial default values ​​of relevant parameters of Ceph storage system cache and Ceph cluster performance indicators under this configuration; use the open source monitoring and alarm tool Prometheus to collect and monitor various indicators of Ceph cluster, connect Prometheus data source through the open source data visualization tool Grafana, and visualize the results of collection and monitoring; the displayable monitoring indicators include Ceph cluster status, read and write performance indicators, component quantity, PG, and OSD information;

[0016] Step S4, dynamic adjustment of parameter optimization system;

[0017] Based on the sample set consisting of parameter configuration values ​​and corresponding monitoring results, a genetic algorithm GA model is established, and the cache size and other parameters are dynamically adjusted using the genetic algorithm. The cluster management tool Ansible is used to apply the parameter configuration combinations generated during the execution of the genetic algorithm to obtain system load and iops performance indicators, and the genetic algorithm GA is used for tuning.

[0018] Step S5: testing and verification;

[0019] Before actual deployment, conduct detailed performance testing and verification; use FIO tools to simulate load, test performance bottlenecks, and verify parameter tuning effects;

[0020] Open the fio configuration file, set the read-write mode rw parameter, and run the fio. / write.fio command to run the fio test on the OSD node to obtain the iops performance results of the current Ceph system and verify whether the tuning parameters are effective.

[0021] Furthermore, the specific method for generating the sample set in step S3 is:

[0022] Step S3.1, for the relevant parameters of the Ceph storage system cache based on persistent memory, use the random function to randomly select values ​​within a reasonable and effective range according to the parameter type and value range; the target tuning parameters are: osd_map_cache_size, cache-mode, hit_set_type, hit_set_count, hit_set_period, target_max_bytes, target_max_objects, cache_target_dirty_ratio, cache_target_dirty_high_ratio, cache_target_full_ratio, osd_journal_size, compression_mode;

[0023] Step S3.2: Apply the default values ​​of the target tuning parameters to the Ceph cluster, test the corresponding system performance, and collect the performance monitoring results iops through the monitoring tool i ;

[0024] Step S3.3: Set the parameter configuration value set {conf 1 ,conf 2 ,......,conf 12} i And the corresponding performance monitoring results iops i Combination, forming a sample data item ({conf 1 ,conf 2 ,......,conf 12} i , iops i ), through loops S3.1 and S3.2, when the number of collected sample data items reaches a preset value, a sample set for building a performance tuning model of the Ceph cache system based on persistent memory is formed.

[0025] Furthermore, the specific method for the monitoring tool to collect performance monitoring results in step S3.2 is:

[0026] Step S3.2.1: Select a node in the ceph cluster to deploy the Ceph MGR management service, and enable the ceph prometheus module built into the Ceph MGR on the node. After successful enablement, the performance of the ceph cluster can be obtained through the default listening port of ceph_exporter;

[0027] Step S3.2.2: Install and configure the Prometheus service in S3 and configure it as a system service for management. Later, the software can use the ceph_exporter port as a data source to obtain the status and performance monitoring indicators of the Ceph cluster;

[0028] Step S3.2.3: Modify the configuration file prometheus.yml of the Prometheus service in S3 and configure the data source so that the Prometheus service can obtain monitoring data from ceph_exporter;

[0029] Step S3.2.4: Install and configure the Grafana service and configure it as a system service for management; after starting, fill in the access address of Prometheus in S3, connect to the Prometheus data source, and then create a dashboard through Grafana for visual monitoring;

[0030] Step S3.2.5: In the Ceph system under the current parameter configuration, execute the stress test script through the Rados Bench tool that comes with Ceph, and obtain the Ceph cluster load and corresponding iops value under the current parameter configuration combination through Prometheus and Grafana.

[0031] Furthermore, in step S4, a genetic algorithm GA model is established, and the specific method of using the genetic algorithm to dynamically adjust the cache size and other parameters is as follows:

[0032] GA simulates the natural selection and reproduction process of biological evolution to find the optimal solution to the problem. Assume that the population size is M and the maximum number of iterations is T. In the parameter tuning problem of Ceph cache system based on persistent memory, the number of Ceph system parameter configuration combinations is used as the population size, and the parameter configuration combination {conf 1 ,conf 2 ,......,conf 12} i As an individual in the population, each specific parameter value represents a gene of the individual, P(t) represents the t-th generation population; GA will select the fitness f({conf 1 ,conf 2 ,......,conf 12}i ) A larger set of individuals {E}, that is, the parameter configuration combination with better system performance at the moment, saves its information, obtains new individuals, that is, new parameter configuration combinations after crossover and mutation operations, and uses {E} to replace the last M / 2 individuals with smaller fitness in the new population, that is, the new parameter configuration combination, so that {E} is retained in the next generation population; until the termination condition is reached, the optimal parameter configuration group is obtained;

[0033] In the above Ceph cache system based on persistent memory, the performance indicator is the number of I / O requests processed per unit time (iops); when the performance indicator is iops, the above fitness f is expressed as:

[0034] f({conf 1 ,conf 2 ,......,conf 12} i )=(iops cur -iops exp ) / iops exp

[0035] Among them, iops cur Indicates the number of I / O requests processed per unit time by the system under the current parameter configuration combination, iops exp Indicates the number of I / O requests processed per unit time by the system under the empirical parameter configuration combination, that is, the parameter values ​​are taken as default values.

[0036] Furthermore, the main steps of the process of applying the GA algorithm to the parameter tuning of the Ceph cache system based on persistent memory include:

[0037] Initialize the population: Initialize a set of parameter configurations based on experience and historical data;

[0038] Obtain samples: Obtain samples of performance indicators under each parameter configuration combination;

[0039] Calculate fitness: Calculate the fitness value corresponding to each parameter configuration combination;

[0040] Selection operation: keep the parameter configuration combination with high fitness and replace the parameter configuration combination with low fitness;

[0041] Genetic operation: recombine the parameter configuration combination according to the crossover probability and mutation probability to obtain a new parameter configuration combination;

[0042] Determine whether the maximum number of iterations has been reached: if not, return to continue executing the genetic operation; otherwise, output the optimal parameter configuration;

[0043] Through continuous iteration and optimization of the above steps, the optimal parameter configuration is finally obtained.

[0044] Preferably, the read / write mode rw parameter in step S5 may be write sequential write, randwrite random write, read sequential read, or randread random read.

[0045] The beneficial effects of the present invention are:

[0046] By applying persistent memory to the ceph distributed storage system, real-time optimization of system performance can be achieved, bringing significant results in many aspects. The Ceph system based on persistent memory has a speed close to that of memory DRAM, and can provide lower latency and higher IOPS throughput compared to traditional SSDs. For high-speed data processing scenarios, using persistent memory as a cache layer can greatly improve read and write performance. At the same time, the use of persistent memory can provide higher data reliability. In the event of a power failure, the non-volatility of persistent memory will ensure that data will not be lost, greatly improving application reliability and reducing the risk of data loss. In addition, combined with the storage parameter optimization of persistent memory, the system can maximize resource utilization and ensure that the system maintains efficient and stable operation under various load requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 The main principle diagram of the Ceph storage system optimized based on persistent memory devices of the present invention;

[0048] Figure 2 It is a schematic diagram of the overall framework of the performance optimization method based on persistent memory of the present invention;

[0049] Figure 3 The present invention is a flowchart of applying the GA algorithm of the present invention to parameter tuning of Ceph cache system based on persistent memory. DETAILED DESCRIPTION

[0050] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0051] The present invention first uses persistent memory as a cache layer to improve read performance. The core idea is that by caching hot data in persistent memory, the number of accesses to the backend storage device can be reduced, thereby improving overall performance.

[0052] Persistent memory devices can be configured as block devices and can be used like ordinary hard disks or solid-state drives (SSDs). In the distributed storage system Ceph, persistent memory can be used as a cache layer to improve performance. In the Ceph storage system, the cache tiering mechanism refers to building a dedicated Ceph pool, the cache pool, on a higher-performance disk. The cache pool is located before the regular pool, so that all I / O operations from the client are initially processed through the cache pool. Subsequently, the data is asynchronously layered and written back to the back-end data pool. The client can enjoy high performance on the cache pool, while the data will eventually be written to the regular pool.

[0053] However, since Ceph itself has thousands of parameters for configuring cache, memory, CPU, and data write-related properties, among which cache configuration includes multiple parameters, different parameter configurations will lead to different cache performance, thus affecting the overall performance of the Ceph storage system.

[0054] The present invention aims at the performance optimization problem of persistent memory as a Ceph cache optimization device, and proposes a method for tuning the main parameters affecting the persistent memory cache through a genetic algorithm, thereby ultimately improving the overall performance of the Ceph storage system. Figure 1 As shown in the figure, the following two aspects are optimized for the Ceph storage system: first, persistent memory devices are introduced for cache optimization; second, the configuration parameters of the Ceph storage system based on persistent memory devices are optimized through algorithms.

[0055] The present invention configures the persistent memory device as a block device and uses it as a Ceph cache. The specific technical approach includes configuring the persistent memory device, configuring the Ceph OSD, optimizing cache performance, adjusting Ceph OSD parameters, and continuous monitoring and tuning. Through reasonable parameter adjustment and monitoring, the performance of the Ceph storage system is improved to ensure that the system performance reaches the optimal state.

[0056] A method for optimizing Ceph cache system performance based on persistent memory, such as Figure 2 As shown in the figure, the Ceph storage system based on persistent memory adjusts the cache and persistent memory related parameter configurations through the parameter optimization system, and feedbacks the performance data through performance monitoring to complete the tuning of the configuration parameters, and finally obtains the optimal performance parameter combination, including the following steps:

[0057] S1. Configure persistent memory devices.

[0058] Configure the persistent memory device as a file system. Use the libpmem library to mount the persistent memory device directly as a file system.

[0059] S2. Configure Ceph OSD.

[0060] Specify the persistent memory device as the cache device in the Ceph configuration file, enable the relevant cache options in ceph.conf and keep the default values. Use the ceph command to create OSD and configure the cache layer.

[0061] S3: Monitor performance and perform sampling to generate a sample set.

[0062] After deployment, continuously monitor Ceph's performance indicators, record the initial default values ​​of the relevant parameters of the Ceph storage system cache and the Ceph cluster performance indicators under this configuration. Specifically, use the open source monitoring and alarm tool Prometheus to collect and monitor various indicators of the Ceph cluster, connect the Prometheus data source through the open source data visualization tool Grafana, and visualize the results of collection and monitoring. The monitoring indicators that can be displayed include the status of the Ceph cluster, read and write performance indicators, component quantity, PG, and OSD information. Here, we focus on the cluster status and read and write performance indicators.

[0063] The specific method for generating sample sets in S3 is:

[0064] S3.1. For the relevant parameters of the Ceph storage system cache based on persistent memory, according to the parameter type and value range, use the random function to randomly select values ​​within a reasonable and effective range. The target tuning parameters of the present invention are shown in the following table.

[0065] Table 1 Tuning parameter list

[0066]

[0067]

[0068] S3.2. Apply the default values ​​of the parameters listed in Table 1 to the Ceph cluster, test the corresponding system performance, and collect the performance monitoring results iops through the monitoring tool i ;

[0069] S3.3, set the parameter configuration value set {conf 1 ,conf 2 ,……,conf 12} i And the corresponding performance monitoring results iops i Combination, forming a sample data item ({conf 1 ,conf 2 ,……,conf 12} i , iops i), through loops S3.1 and S3.2, when the number of collected sample data items reaches a preset value, a sample set for building a performance tuning model of the Ceph cache system based on persistent memory is formed.

[0070] The specific method for the monitoring tool in S3.2 to collect performance monitoring results is:

[0071] Step S3.2.1: Select a node in the ceph cluster to deploy the Ceph MGR management service, and enable the ceph prometheus module built into the Ceph MGR on the node. After successful enablement, the performance of the ceph cluster can be obtained through the default listening port of ceph_exporter;

[0072] Step S3.2.2: Install and configure the Prometheus service in S3 and configure it as a system service for management. Later, the software can use the ceph_exporter port as a data source to obtain the status and performance monitoring indicators of the Ceph cluster;

[0073] Step S3.2.3: Modify the configuration file prometheus.yml of the Prometheus service in S3 and configure the data source so that the Prometheus service can obtain monitoring data from ceph_exporter;

[0074] Step S3.2.4: Install and configure the Grafana service and configure it as a system service for management. After starting, fill in the access address of Prometheus in S3, connect to the Prometheus data source, and then create a dashboard through Grafana for visual monitoring;

[0075] Step S3.2.5: In the Ceph system under the current parameter configuration, execute the stress test script through the Rados Bench tool that comes with Ceph, and obtain the Ceph cluster load and corresponding iops value under the current parameter configuration combination through Prometheus and Grafana.

[0076] S4. Dynamic adjustment of parameter optimization system.

[0077] Based on the sample set composed of parameter configuration values ​​and corresponding monitoring results, a genetic algorithm (GA) model is established, and the cache size and other parameters are dynamically adjusted using the genetic algorithm. The cluster management tool Ansible is used to apply the parameter configuration combination generated during the execution of the genetic algorithm to obtain system load and iops performance indicators, and the genetic algorithm GA is used for tuning.

[0078] A genetic algorithm GA model is established in S4. The specific method of using the genetic algorithm to dynamically adjust the cache size and other parameters is as follows:

[0079] GA simulates the natural selection and reproduction process of biological evolution to find the optimal solution to the problem. Let the population size be M, the maximum number of iterations be T, and in the parameter tuning problem of Ceph cache system based on persistent memory, the number of Ceph system parameter configuration combinations is used as the population size, and the parameter configuration combination {conf 1 ,conf 2 ,......,conf 12} i As an individual in the population, each specific parameter value represents a gene of the individual, and P(t) represents the t-th generation population. GA will select the fitness f({conf 1 ,conf 2 ,......,conf 12} i ) The largest individual set {E}, that is, the parameter configuration combination with the best system performance at the moment, saves its information, obtains new individuals, that is, new parameter configuration combinations after crossover and mutation operations, and uses {E} to replace the last M / 2 individuals with smaller fitness in the new population (that is, the new parameter configuration combination), so that {E} is retained in the next generation population. Until the termination condition is reached, the optimal parameter configuration group is obtained.

[0080] In the above Ceph cache system based on persistent memory, the performance indicator is the number of I / O requests processed per unit time (iops). When the performance indicator is iops, the above fitness f is expressed as:

[0081] f({conf 1 ,conf 2 ,......,conf 12}i)=(iops cur -iops exp ) / iops exp

[0082] Among them, iops cur Indicates the number of I / O requests processed per unit time by the system under the current parameter configuration combination, iops exp Indicates the number of I / O requests processed per unit time by the system under the empirical parameter configuration combination (that is, the parameter values ​​are the default values).

[0083] The flowchart and pseudocode of GA algorithm applied to parameter tuning of Ceph cache system based on persistent memory are as follows: Figure 3 , as shown in Table 2.

[0084] Figure 3The following is a flowchart of applying the GA algorithm of the present invention to parameter tuning of Ceph cache system based on persistent memory. The main steps of the process include:

[0085] Initialize the population: Initialize a set of parameter configurations based on experience and historical data.

[0086] Get samples: Get samples of performance indicators (such as IOPS values) under each parameter configuration combination.

[0087] Calculate fitness: Calculate the fitness value corresponding to each parameter configuration combination.

[0088] Select operation: keep the parameter configuration combination with high fitness and replace the parameter configuration combination with low fitness.

[0089] Genetic operation: Recombine the parameter configuration combination according to the crossover probability and mutation probability to obtain a new parameter configuration combination.

[0090] Determine whether the maximum number of iterations has been reached: if not, return to continue executing the genetic operation; otherwise, output the optimal parameter configuration.

[0091] Through continuous iteration and optimization of the above steps, the optimal parameter configuration is finally obtained.

[0092] Table 2 Pseudo-code representation of the GA algorithm in the Ceph cache system parameter tuning scenario based on persistent memory

[0093]

[0094]

[0095] S5. Testing and verification.

[0096] Before actual deployment, conduct detailed performance testing and verification. Use the FIO (Flexible I / O Tester) tool to simulate load, test performance bottlenecks, and verify the effect of parameter tuning.

[0097] Open the fio configuration file, set the read / write mode rw parameter, and select write sequential write, randwrite random write, read sequential read, and randread random read. Run the fio. / write.fio command to run the fio test on the OSD node to obtain the iops performance results of the current Ceph system and verify whether the tuning parameters are effective.

[0098] The above-mentioned embodiment only expresses one implementation mode of the present invention, and its description is relatively specific and detailed, but it cannot be understood as limiting the scope of the invention patent. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be based on the attached claims.

Claims

1. A method for optimizing Ceph cache system performance based on persistent memory, characterized in that: The Ceph storage system based on persistent memory adjusts the cache and persistent memory related parameter configurations through the parameter optimization system, and uses performance monitoring to feedback performance data to complete the tuning of configuration parameters and finally obtain the optimal performance parameter combination, including the following steps: Step S1, configuring a persistent memory device; Configure the persistent memory device to file system mode; use the libpmem library to directly mount the persistent memory device as a file system; Step S2: Configure Ceph OSD; Specify the persistent memory device as the cache device in the Ceph configuration file, enable the relevant cache options in ceph.conf and keep the default values; create OSD and configure the cache layer through the ceph command; Step S3: monitor performance and perform sampling to generate a sample set; After deployment, continuously monitor Ceph's performance indicators, record the initial default values ​​of relevant parameters of the Ceph storage system cache and the Ceph cluster performance indicators under this configuration; use the open source monitoring and alarm tool Prometheus to collect and monitor various indicators of the Ceph cluster, connect the Prometheus data source through the open source data visualization tool Grafana, and visualize the results of collection and monitoring; the displayable monitoring indicators include the status of the Ceph cluster, read and write performance indicators, component quantity, PG, and OSD information; Step S4, parameter optimization system dynamic adjustment; Based on the sample set consisting of parameter configuration values ​​and corresponding monitoring results, a genetic algorithm GA model is established, and the cache size and other parameters are dynamically adjusted using the genetic algorithm. The cluster management tool Ansible is used to apply the parameter configuration combinations generated during the execution of the genetic algorithm to obtain system load and iops performance indicators, and the genetic algorithm GA is used for tuning. Step S5: testing and verification; Before actual deployment, conduct detailed performance testing and verification; use FIO tools to simulate load, test performance bottlenecks, and verify parameter tuning effects; Open the fio configuration file, set the read-write mode rw parameter, and run the fio. / write.fio command to run the fio test on the OSD node to obtain the iops performance results of the current Ceph system and verify whether the tuning parameters are effective.

2. The method for optimizing Ceph cache system performance based on persistent memory according to claim 1, characterized in that: The specific method of generating the sample set in step S3 is: Step S3.1, for the relevant parameters of the Ceph storage system cache based on persistent memory, use the random function to randomly select values ​​within a reasonable and effective range according to the parameter type and value range; the target tuning parameters are: osd_map_cache_size, cache-mode, hit_set_type, hit_set_count, hit_set_period, target_max_bytes, target_max_objects, cache_target_dirty_ratio, cache_target_dirty_high_ratio, cache_target_full_ratio, osd_journal_size, compression_mode; Step S3.2: Apply the default values ​​of the target tuning parameters to the Ceph cluster, test the corresponding system performance, and collect the performance monitoring results iops through the monitoring tool i ; Step S3.3: Set the parameter configuration value set {conf1, conf2, ..., conf 12 } i And the corresponding performance monitoring results iops i Combination, forming a sample data item ({conf1,conf2,...,conf 12 } i ,iops i ), through loops S3.1 and S3.2, when the number of collected sample data items reaches a preset value, a sample set for building a performance tuning model of the Ceph cache system based on persistent memory is formed.

3. The method for optimizing Ceph cache system performance based on persistent memory according to claim 2, characterized in that: The specific method for the monitoring tool to collect performance monitoring results in step S3.2 is: Step S3.2.1: Select a node in the ceph cluster to deploy the Ceph MGR management service, enable the ceph prometheus module built into CephMGR on the node, and after successful enablement, obtain the performance of the ceph cluster through the default listening port of ceph_exporter; Step S3.2.2: Install and configure the Prometheus service in S3 and configure it as a system service for management. Later, the software can use the ceph_exporter port as a data source to obtain the status and performance monitoring indicators of the Ceph cluster; Step S3.2.3: Modify the configuration file prometheus.yml of the Prometheus service in S3 and configure the data source so that the Prometheus service can obtain monitoring data from ceph_exporter; Step S3.2.4: Install and configure the Grafana service and configure it as a system service for management; after starting, fill in the access address of Prometheus in S3, connect to the Prometheus data source, and then create a dashboard through Grafana for visual monitoring; Step S3.2.5: In the Ceph system under the current parameter configuration, execute the stress test script through the RadosBench tool that comes with Ceph, and obtain the Ceph cluster load and corresponding iops value under the current parameter configuration combination through Prometheus and Grafana.

4. The method for optimizing Ceph cache system performance based on persistent memory according to claim 1, characterized in that: In step S4, a genetic algorithm GA model is established, and the specific method of using the genetic algorithm to dynamically adjust the cache size and other parameters is as follows: GA simulates the natural selection and reproduction process of biological evolution to find the optimal solution to the problem. Assume that the population size is M and the maximum number of iterations is T. In the parameter tuning problem of Ceph cache system based on persistent memory, the number of Ceph system parameter configuration combinations is used as the population size, and the parameter configuration combination {conf1, conf2, ..., conf 12 } i As an individual in the population, each specific parameter value represents a gene of the individual, P(t) represents the t-th generation population; GA will select the fitness f({conf1, conf2, ..., conf 12 } i ) A larger set of individuals {E}, that is, the parameter configuration combination with better system performance at the moment, saves its information, obtains new individuals, that is, new parameter configuration combinations after crossover and mutation operations, and uses {E} to replace the last M / 2 individuals with smaller fitness in the new population, that is, the new parameter configuration combination, so that {E} is retained in the next generation population; Until the termination condition is reached, the optimal parameter configuration group is obtained; In the above Ceph cache system based on persistent memory, the performance indicator is the number of I / O requests processed per unit time, lops; when the performance indicator is iops, the above fitness f is expressed as: f({conf1,conf2,......,conf 12 } i )=(iops cur -iops exp ) / iops exp Among them, iops cur Indicates the number of I / O requests processed per unit time by the system under the current parameter configuration combination, iops exp Indicates the number of I / O requests processed per unit time by the system under the empirical parameter configuration combination, that is, the parameter values ​​are taken as default values.

5. The method for optimizing Ceph cache system performance based on persistent memory according to claim 4, characterized in that: The main steps of the GA algorithm application process for tuning parameters of the Ceph cache system based on persistent memory include: Initialize the population: Initialize a set of parameter configurations based on experience and historical data; Obtain samples: Obtain samples of performance indicators under each parameter configuration combination; Calculate fitness: Calculate the fitness value corresponding to each parameter configuration combination; Selection operation: keep the parameter configuration combination with high fitness and replace the parameter configuration combination with low fitness; Genetic operation: recombine the parameter configuration combination according to the crossover probability and mutation probability to obtain a new parameter configuration combination; Determine whether the maximum number of iterations has been reached: if not, return to continue executing the genetic operation; otherwise, output the optimal parameter configuration; Through continuous iteration and optimization of the above steps, the optimal parameter configuration is finally obtained.

6. The method for optimizing Ceph cache system performance based on persistent memory according to claim 1, characterized in that: The read / write mode rw parameter in step S5 may be write sequential write, randwrite random write, read sequential read, or randread random read.

Images