A memory hotspot elimination method and device based on memory pooling technology

By dividing the load into delay-sensitive and non-sensitive loads, and using memory pooling technology for memory bandwidth control or cold data migration according to environmental support, the performance loss and migration failure caused by memory hotspots are solved, achieving more efficient Memory resource management.

CN118550846BActive Publication Date: 2024-11-26CHINA TELECOM CLOUD TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411026252.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-30
Publication Date
2024-11-26
Estimated Expiration
2044-07-30

AI Technical Summary

Technical Problem

When facing memory hotspots, the prior art usually mitigates by suppressing or migrating the jamming of neighbor loads, but this will lead to performance losses or migration failures, affecting load performance.

Method used

Using a method based on memory pooling technology, the load is divided into delay-sensitive and delay-insensitive loads, and determine whether the delay-insensitive load has memory hotspots, and use the memory bandwidth resource allocation sub-tool based on whether the Intel-RDT environment is supported. Or page error handlers perform control or cold data migration, reducing memory bandwidth usage and alleviating memory hotspots.

Benefits of technology

It effectively reduces memory hotspots, avoids the impact on load performance, solves the problem of migration failure, and improves the system's memory access efficiency and resource utilization.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118550846B_ABST
    Figure CN118550846B_ABST
Patent Text Reader

Abstract

The present invention relates to a method and device for eliminating memory hotspots based on memory pooling technology, which belongs to the field of cloud computing technology, including: obtaining the load of local memory, dividing the load into delay-sensitive load and delay-insensitive load, judging whether the delay-insensitive load has a memory hotspot; if so, entering the next step; otherwise, ending the process; judging whether the current environment supports the Intel-RDT environment, if so, controlling the bandwidth of the delay-insensitive load with memory hotspots through the memory bandwidth resource allocation sub-tool, otherwise, performing cold data migration on the delay-insensitive load with memory hotspots through the page fault handler in the kernel. When there are not only delay-sensitive loads, they are arranged in ascending order by access frequency, and a preset proportion of memory pages are migrated to the CXL memory pool to alleviate local memory hotspots, avoid affecting the performance of the load, and solve the problem of migration failure.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of cloud computing, and in particular relates to a method and device for eliminating memory hot spots based on memory pooling technology. Background Art

[0002] At present, tenant businesses on the cloud that use VMs and containers as operating carriers have differences in their use of memory bandwidth resources. In addition, due to the lack of reasonable management and control of these resources, latency-sensitive and memory bandwidth resource-sensitive loads are easily disturbed by memory-intensive loads, resulting in performance impairment. For some businesses that are sensitive to memory access latency, their performance will be volatile, and in severe cases, they may even fail to meet their QoS goals.

[0003] In a high-density co-location cloud scenario, the impact of this QoS default on tenant services is sometimes catastrophic. The services of a few tenants have long generated a large amount of memory bandwidth traffic consumption that is disproportionate to the memory resources allocated to them, resulting in memory channel traffic saturation, which further causes a significant increase in memory access latency for other co-located tenant workloads, resulting in performance degradation.

[0004] When facing the problem of memory hotspots, existing technologies mostly consider suppressing the disturbing neighbor load or migrating the disturbing neighbor load to alleviate the memory hotspot phenomenon. However, suppressing the disturbing neighbor load requires sacrificing the performance of the current load, which will lead to memory hotspots. At the same time, migrating the disturbing neighbor load faces the problem of migration failure due to non-convergence of the migration process in high-pressure scenarios, which affects the performance of the load. Summary of the invention

[0005] In view of the above deficiencies in the prior art, the purpose of the invention is to provide a method and device for eliminating memory hotspots based on memory pooling technology, by dividing the load into latency-sensitive loads and latency-insensitive loads, determining whether the latency-insensitive load has a memory hotspot, and determining whether the current environment supports the Intel-RDT environment; if so, controlling the bandwidth of the latency-insensitive load with memory hotspots through a memory bandwidth resource allocation sub-tool; otherwise, performing cold data migration on the latency-insensitive load with memory hotspots through a page fault handler in the kernel; effectively reducing the occupation of memory bandwidth resources by disturbing neighbor loads, alleviating local memory hotspots, and avoiding the impact on load performance.

[0006] A first aspect of the present invention provides a method for eliminating memory hot spots based on memory pooling technology, comprising:

[0007] S1, obtain the load of local memory;

[0008] S2, dividing the load into a delay-sensitive load and a delay-insensitive load;

[0009] S3, determining whether the latency-insensitive load has a memory hotspot; if so, proceeding to the next step; otherwise, ending the process;

[0010] S4, determine whether the current environment supports the Intel-RDT environment; if so, control the bandwidth of latency-insensitive loads with memory hotspots through the memory bandwidth resource allocation sub-tool; otherwise, perform cold data migration on latency-insensitive loads with memory hotspots through the page fault handler in the kernel.

[0011] Furthermore, after S2, the method further includes:

[0012] S5: When only latency-sensitive loads exist on the non-uniform memory access node, the memory pages are sorted in ascending order by access frequency and a preset proportion of memory pages are migrated to the CXL memory pool.

[0013] Further, the determining in S3 whether the latency-insensitive load has a memory hotspot specifically includes:

[0014] In an environment supporting Intel-RDT, the memory bandwidth of the load is monitored by using an RDT tool;

[0015] Determine whether the memory bandwidth exceeds a first threshold; if so, determine that a memory hotspot exists in the load.

[0016] Further, the determining in S3 whether the latency-insensitive load has a memory hotspot specifically includes:

[0017] In an environment that does not support Intel-RDT, real-time memory access latency is obtained by periodically sending a small number of non-temporal access instructions to the local memory;

[0018] Determine whether the real-time memory access latency exceeds a second threshold; if so, determine that a memory hotspot exists in the load.

[0019] Furthermore, the page table entry accessed again by the latency-insensitive load with a memory hotspot in S4 is processed by a page fault handler in the kernel, specifically including:

[0020] S401, periodically sampling the load memory access through PEBS;

[0021] S402, modifying the memory page table of the load according to the memory address recorded by the sample, and marking the corresponding page table entry as an invalid page table entry;

[0022] S403, when the load accesses the memory page again, access processing is performed through the page fault handler in the kernel;

[0023] S404, sorting in ascending order according to access frequency, and classifying the memory data in a preset number of memory addresses as cold data;

[0024] S405: Migrate the cold data to the CXL memory pool.

[0025] A second aspect of the present invention provides a memory hotspot elimination device based on memory pooling technology, comprising:

[0026] Acquisition module, used to obtain the load of local memory;

[0027] A division module, used for dividing the load into a delay-sensitive load and a delay-insensitive load;

[0028] The first judgment module is used to judge whether the latency-insensitive load has a memory hotspot; if so, enter the second judgment module; otherwise, end the process;

[0029] The second judgment module is used to determine whether the current environment supports the Intel-RDT environment; if so, the bandwidth of the latency-insensitive load with memory hotspots is controlled through the memory bandwidth resource allocation sub-tool; otherwise, cold data migration is performed on the latency-insensitive load with memory hotspots through the page fault handler in the kernel.

[0030] Furthermore, the device also includes:

[0031] The migration module is used to migrate a preset proportion of memory pages to the CXL memory pool in ascending order of access frequency when only latency-sensitive loads exist on the non-uniform memory access node.

[0032] Furthermore, the first determination module is used for:

[0033] In an environment supporting Intel-RDT, the memory bandwidth of the load is monitored by using an RDT tool;

[0034] Determine whether the memory bandwidth exceeds a first threshold; if so, determine that a memory hotspot exists in the load.

[0035] Furthermore, the first determination module is used for:

[0036] In an environment that does not support Intel-RDT, real-time memory access latency is obtained by periodically sending a small number of non-temporal access instructions to the local memory;

[0037] Determine whether the real-time memory access latency exceeds a second threshold; if so, determine that a memory hotspot exists in the load.

[0038] Furthermore, the second judgment module is used for:

[0039] Periodically sample load memory accesses through PEBS;

[0040] According to the memory address recorded by the sample, the memory page table of the load is modified, and the corresponding page table entry is marked as an invalid page table entry;

[0041] When the load accesses the memory page again, the access is processed by the page fault handler in the kernel;

[0042] Arrange in ascending order according to access frequency, and classify the memory data in a preset number of memory addresses as cold data;

[0043] Migrate the cold data to the CXL memory pool.

[0044] A third aspect of the present invention proposes a memory hotspot elimination system based on memory pooling technology, comprising: a memory and one or more processors.

[0045] One or more application programs are stored in the memory, and the one or more application programs are suitable for being executed by one or more processors to implement a memory hotspot elimination method based on memory pooling technology according to the first aspect.

[0046] In a fourth aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored. The computer program can be loaded and executed by a processor to provide a memory hotspot elimination method based on memory pooling technology according to the first aspect.

[0047] The beneficial effects of the present invention are as follows:

[0048] The method and device of the present invention divide the load into latency-sensitive load and latency-insensitive load. Determine whether the latency-insensitive load has a memory hotspot, and determine whether the current environment supports the Intel-RDT environment; if so, control the bandwidth of the latency-insensitive load with memory hotspot through the memory bandwidth resource allocation sub-tool; otherwise, perform cold data migration on the latency-insensitive load with memory hotspot through the page fault handler in the kernel to alleviate local memory hotspots and solve the problem of migration failure. When only latency-sensitive loads exist on non-uniform memory access nodes, they are sorted in ascending order by access frequency, and a preset proportion of memory pages are migrated to the CXL memory pool, thereby reducing memory hotspots and avoiding affecting load performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The accompanying drawings are only used to illustrate specific embodiments and are not considered to limit the present invention. In the entire drawings, the same reference symbols represent the same components. Obviously, the drawings described below are only some embodiments recorded in the embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0050] Figure 1 A flowchart of a method for eliminating memory hot spots based on memory pooling technology provided by an embodiment of the present invention;

[0051] Figure 2 Another schematic diagram of a flow chart of a method for eliminating memory hot spots based on memory pooling technology provided by an embodiment of the present invention;

[0052] Figure 3 A schematic diagram of the structure of a memory hotspot elimination device based on memory pooling technology provided by an embodiment of the present invention;

[0053] Figure 4 A structural schematic diagram of a memory hotspot elimination system based on memory pooling technology provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. It should be understood that these descriptions are only exemplary and are not intended to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in the field without making creative work should fall within the scope of protection of the present invention.

[0055] Furthermore, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessarily obscuring the concepts disclosed in the present invention.

[0056] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the orientation or position relationship indicated by the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inside", "outside", etc. is based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation of the present invention. In addition, the terms "first", "second", and "third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. The terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be a connection between the two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0057] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of methods and devices consistent with some aspects of the present invention as detailed in the appended claims.

[0058] The present invention proposes a memory hotspot elimination method and device based on memory pooling technology, in order to solve the problem that the existing technology, when facing the memory hotspot problem, mostly considers suppressing the disturbing neighbor load or migrating the disturbing neighbor load to alleviate the memory hotspot phenomenon. However, suppressing the disturbing neighbor load needs to sacrifice the performance of the current load as a premise, which will lead to the occurrence of memory hotspots; at the same time, migrating the disturbing neighbor load faces the problem of migration failure due to non-convergence of the migration process in high-pressure scenarios, which affects the performance of the load.

[0059] Method Embodiment

[0060] Reference Manual Attached Figure 1 , showing a flow chart of a memory hotspot elimination method based on memory pooling technology provided by an embodiment of the present invention.

[0061] Reference Manual Attached Figure 2 , shows another flow chart of a method for eliminating memory hot spots based on memory pooling technology provided by an embodiment of the present invention.

[0062] An embodiment of the present invention provides a method for eliminating memory hot spots based on memory pooling technology, comprising:

[0063] Specifically, the method includes steps S1 to S4.

[0064] S1, obtain the load of local memory.

[0065] S2, divides the load into delay-sensitive load and delay-insensitive load.

[0066] It should be noted that the specific basis for classification can be determined by external direct input labels or by load resource usage pattern analysis based on machine learning technology. Delay-sensitive loads usually have periodic changes in resource consumption during the day and night, and there is a certain degree of interactivity, that is, a mode in which "network traffic" drives "business", that is, during the period when network traffic enters the load, the load has synchronous CPU computing resource consumption, especially user-mode CPU resource consumption. Through the correlation analysis of network and user-mode CPU resource consumption, loads that show negative correlation can be marked as delay-insensitive.

[0067] For example, when the video transcoding FFMpeg load and the key-value memory cache Memcached load are co-located, the FFMpeg load will consume a large amount of memory bandwidth resources when running. First, the FFMpeg and Memcached loads are classified. FFMpeg will be classified as a latency-insensitive load, and Memcached will be classified as a latency-sensitive load.

[0068] In the present invention, after classifying different types of loads, the performance requirements and bottlenecks of the system can be predicted more accurately. For delay-sensitive loads, the system can take targeted tuning measures to meet their specific performance requirements. At the same time, by effectively managing and allocating resources, the utilization rate of system resources can be improved. Delay-insensitive loads usually have more flexible requirements for resource use, so they can be processed during resource idle or off-peak hours to avoid wasting resources.

[0069] In a possible implementation manner, after S2, the method further includes:

[0070] S5: When only latency-sensitive loads exist on the non-uniform memory access node, the memory pages are sorted in ascending order by access frequency and a preset proportion of memory pages are migrated to the CXL memory pool.

[0071] Among them, CXL (Compute-Express-Link) is an open, standard high-speed interconnect technology for connecting CPUs, memory, accelerators and other devices. CXL was jointly developed by major technology companies in computing and storage systems to improve the performance and efficiency of data centers.

[0072] In the present invention, the CXL memory pool generally has higher bandwidth and lower access latency than the host memory. By migrating some memory pages of latency-sensitive loads to the CXL memory pool, the data access speed of these loads can be accelerated, significantly improving the responsiveness of the application.

[0073] Furthermore, CXL technology allows high-speed interconnection between CPU, memory and accelerators, effectively utilizing high-bandwidth communications between these devices. Migrating memory pages to the CXL memory pool can optimize the allocation of system resources and ensure that mission-critical applications or applications that require high-performance data access can obtain sufficient bandwidth support.

[0074] S3, determine whether there is a memory hotspot in the latency-insensitive load. If so, proceed to the next step. Otherwise, end the process.

[0075] In a possible implementation manner, determining whether there is a memory hotspot in the latency-insensitive load in S3 specifically includes:

[0076] In an environment that supports Intel-RDT, use the RDT tool to monitor the memory bandwidth of the load.

[0077] It is determined whether the memory bandwidth exceeds a first threshold. If so, it is determined that the load has a memory hotspot.

[0078] In a possible implementation manner, determining whether there is a memory hotspot in the latency-insensitive load in S3 specifically includes:

[0079] In an environment that does not support Intel-RDT, real-time memory access latency is obtained by periodically sending a small number of non-temporal access instructions to the local memory.

[0080] Determine whether the real-time memory access latency exceeds a second threshold. If so, determine that the load has a memory hotspot.

[0081] Among them, RDT (Resource-Director-Technology) is a technology provided by Intel, which aims to manage and regulate the resources inside the processor, such as memory bandwidth and cache capacity, through hardware support. It is mainly used to optimize and control the allocation of shared resources to improve the performance and predictability of the system.

[0082] Specifically, the present invention will periodically start a local memory access latency detection script to monitor the memory access latency. If the detected local memory access latency is greater than a certain threshold, such as 120ns, it is determined that a memory hotspot phenomenon has occurred. In an environment where the CPU enables Intel-RDT, the RDT tool is used to monitor the memory bandwidth. If the detected memory bandwidth is greater than a certain threshold, such as 60% of the total bandwidth, it is determined that a memory hotspot phenomenon has occurred.

[0083] In the present invention, by monitoring memory bandwidth or real-time memory access latency, memory hotspots in latency-insensitive loads can be discovered and identified in a timely manner. This helps predict performance issues that may occur in the system and take preventive measures to avoid potential performance degradation or system failures. At the same time, memory hotspots may cause resource competition and conflicts between different tasks, thereby affecting the stability and reliability of the system. Timely identification and processing of memory hotspots helps reduce these competitions and conflicts and improve the operational stability of the system.

[0084] Furthermore, by effectively managing memory access patterns and resource allocation, the system's energy consumption and operating costs can be reduced. Avoiding memory hot spots can reduce unnecessary resource waste and improve the energy efficiency of a data center or server environment.

[0085] S4, determine whether the current environment supports the Intel-RDT environment. If so, control the bandwidth of latency-insensitive loads with memory hotspots through the memory bandwidth resource allocation sub-tool. Otherwise, perform cold data migration on latency-insensitive loads with memory hotspots through the page fault handler in the kernel.

[0086] Among them, the Page Fault Handler is an important part of the operating system, responsible for handling page faults that occur when a program accesses memory. A page fault is triggered when a program attempts to access a virtual page that has not yet been loaded into physical memory.

[0087] Specifically, in an RDT environment, memory bandwidth restrictions are imposed on FFMpeg. For example, the bandwidth restrictions on FFMpeg loads are adjusted in steps of 50%, 40%...10%, and at the same time, whether the memory hotspot phenomenon is alleviated is observed.

[0088] In the present invention, in an environment that supports Intel-RDT, the memory bandwidth resource allocation sub-tool can be used to accurately control the memory bandwidth usage of latency-insensitive loads. By setting different bandwidth limits, such as 50%, 40%, etc., the generation of memory hotspots can be effectively avoided while ensuring the normal operation of other tasks in the system. At the same time, by observing and adjusting the usage of memory bandwidth in real time, it is possible to respond to changes in load and system requirements in a timely manner, and optimize resource allocation and utilization efficiency. This method can effectively manage and balance the competition for memory bandwidth from different loads without affecting the overall performance of the system.

[0089] In a possible implementation, in S4, a page fault handler in the kernel processes a page table entry accessed again by a latency-insensitive load having a memory hotspot, specifically including:

[0090] S401, periodically sampling the load memory access through PEBS.

[0091] In the present invention, the memory access pattern of latency-insensitive loads can be dynamically identified and processed through periodic sampling and management of page table entries.

[0092] S402, modify the memory page table of the load according to the memory address recorded by the sample, and mark the corresponding page table entry as an invalid page table entry.

[0093] S403, when the load accesses the memory page again, the access is processed by the page fault handler in the kernel.

[0094] In the present invention, the implementation of the page fault handling program does not rely on specific hardware support and is applicable to a variety of system environments. By effectively managing and processing memory access, the overall performance of the system can be optimized, and the user experience and service quality can be improved.

[0095] S404, sorting in ascending order according to access frequency, and classifying the memory data in a preset number of memory addresses as cold data.

[0096] S405, migrating cold data to the CXL memory pool.

[0097] Specifically, PEBS is used to periodically sample FFMpeg's memory access behavior, and the samples are counted to obtain memory within a range with low access frequency, such as a set of memory addresses with access frequency below the 50% percentile, and this part of memory data is migrated to the CXL memory pool. In an environment that does not support RDT, PEBS is used to periodically sample FFMpeg's memory access behavior, and the page table corresponding to the FFMepg process is found in the OS. All memory addresses within the sampling range are invalidated, causing related memory page accesses to cause page faults. During the page fault handle process, the page table entries are synchronously refilled in order. These refilled page table entries may be re-invalidated by subsequent PEBS sampling to intercept FFMpeg's normal access to memory and weaken FFMpeg's squeeze on memory bandwidth. At the same time, similar to the actions in an environment that supports RDT, cold data within the address range needs to be migrated to the CXL memory pool to expand the memory access bandwidth in the CXL memory pool.

[0098] In the present invention, memory data with low access frequency is marked as cold data and migrated to the CXL memory pool, which can effectively release host memory resources and improve the overall memory access efficiency of the system. At the same time, according to different environments and hardware support, choosing appropriate methods to handle memory hotspots of latency-insensitive loads can effectively improve the performance, stability and resource utilization efficiency of the system, thereby optimizing the overall performance effect.

[0099] The beneficial effects of the present invention are as follows:

[0100] By dividing the load into latency-sensitive load and latency-insensitive load. Determine whether the latency-insensitive load has a memory hotspot, and determine whether the current environment supports the Intel-RDT environment; if so, control the bandwidth of the latency-insensitive load with memory hotspot through the memory bandwidth resource allocation sub-tool; otherwise, perform cold data migration on the latency-insensitive load with memory hotspot through the page fault handler in the kernel to alleviate local memory hotspots and solve the problem of migration failure. When only latency-sensitive loads exist on the non-uniform memory access node, they are sorted in ascending order by access frequency, and a preset proportion of memory pages are migrated to the CXL memory pool, thereby reducing memory hotspots and avoiding affecting the performance of the load.

[0101] Device Embodiment

[0102] Reference Manual Attached Figure 3 , showing a structural schematic diagram of a memory hotspot elimination device based on memory pooling technology provided by an embodiment of the present invention.

[0103] An embodiment of the present invention provides a memory hotspot elimination device 20 based on memory pooling technology, comprising:

[0104] The acquisition module 201 is used to acquire the load of the local memory;

[0105] A division module 202, configured to divide the load into a delay-sensitive load and a delay-insensitive load;

[0106] The first judgment module 203 is used to judge whether the latency-insensitive load has a memory hotspot; if so, enter the second judgment module; otherwise, end the process;

[0107] The second judgment module 204 is used to determine whether the current environment supports the Intel-RDT environment; if so, the bandwidth of the latency-insensitive load with memory hotspots is controlled through the memory bandwidth resource allocation sub-tool; otherwise, cold data migration is performed on the latency-insensitive load with memory hotspots through the page fault handler in the kernel.

[0108] In a possible implementation, the device further includes:

[0109] The migration module is used to migrate a preset proportion of memory pages to the CXL memory pool in ascending order of access frequency when only latency-sensitive loads exist on the non-uniform memory access node.

[0110] In a possible implementation manner, the first determining module 203 is used to:

[0111] In an environment supporting Intel-RDT, the memory bandwidth of the load is monitored by using an RDT tool;

[0112] Determine whether the memory bandwidth exceeds a first threshold; if so, determine that a memory hotspot exists in the load.

[0113] In a possible implementation manner, the first determining module 203 is used to:

[0114] In an environment that does not support Intel-RDT, real-time memory access latency is obtained by periodically sending a small number of non-temporal access instructions to the local memory;

[0115] Determine whether the real-time memory access latency exceeds a second threshold; if so, determine that a memory hotspot exists in the load.

[0116] In a possible implementation manner, the second determining module 204 is configured to:

[0117] Periodically sample load memory accesses through PEBS;

[0118] According to the memory address recorded by the sample, the memory page table of the load is modified, and the corresponding page table entry is marked as an invalid page table entry;

[0119] When the load accesses the memory page again, the access is processed by the page fault handler in the kernel;

[0120] Arrange in ascending order according to access frequency, and classify the memory data in a preset number of memory addresses as cold data;

[0121] Migrate the cold data to the CXL memory pool.

[0122] A memory hotspot elimination device 20 based on memory pooling technology provided by the present invention can implement each of the memory hotspot elimination methods based on memory pooling technology in the above-mentioned method embodiments, and can obtain the same technical effect. To avoid repetition, the present invention will not go into details.

[0123] The beneficial effects of the present invention are as follows:

[0124] By dividing the load into latency-sensitive load and latency-insensitive load. Determine whether the latency-insensitive load has a memory hotspot, and determine whether the current environment supports the Intel-RDT environment; if so, control the bandwidth of the latency-insensitive load with memory hotspot through the memory bandwidth resource allocation sub-tool; otherwise, perform cold data migration on the latency-insensitive load with memory hotspot through the page fault handler in the kernel to alleviate local memory hotspots and solve the problem of migration failure. When only latency-sensitive loads exist on the non-uniform memory access node, they are sorted in ascending order by access frequency, and a preset proportion of memory pages are migrated to the CXL memory pool, thereby reducing memory hotspots and avoiding affecting the performance of the load.

[0125] System Example

[0126] Reference Manual Attached Figure 4 , showing a structural schematic diagram of a memory hotspot elimination system based on memory pooling technology provided by an embodiment of the present invention.

[0127] The present invention proposes a memory hotspot elimination system 30 based on memory pooling technology, including: a memory 303 and one or more processors 301.

[0128] One or more application programs are stored in the memory 303 , and the one or more application programs are suitable for being executed by the one or more processors 301 to implement a memory hotspot elimination method based on memory pooling technology described in a method embodiment.

[0129] A memory hotspot elimination system 30 based on memory pooling technology includes: a processor 301 and a memory 303. The processor 301 and the memory 303 are connected, for example, through a bus 302.

[0130] The structure of the memory hotspot elimination system 30 based on the memory pooling technology does not constitute a limitation on the embodiment of the present invention.

[0131] The beneficial effects of the present invention are as follows:

[0132] By dividing the load into latency-sensitive load and latency-insensitive load. Determine whether the latency-insensitive load has a memory hotspot, and determine whether the current environment supports the Intel-RDT environment; if so, control the bandwidth of the latency-insensitive load with memory hotspot through the memory bandwidth resource allocation sub-tool; otherwise, perform cold data migration on the latency-insensitive load with memory hotspot through the page fault handler in the kernel to alleviate local memory hotspots and solve the problem of migration failure. When only latency-sensitive loads exist on the non-uniform memory access node, they are sorted in ascending order by access frequency, and a preset proportion of memory pages are migrated to the CXL memory pool, thereby reducing memory hotspots and avoiding affecting the performance of the load.

[0133] Processor 301 may be a CPU, a general purpose processor, a DSP, an ASIC, an FPGA or other programmable logic device, a transistor logic device, a hardware component or any combination thereof. It may implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of the present invention. Processor 301 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0134] The bus 302 may include a path to transmit information between the above components. The bus 302 may be a PCI bus or an EISA bus, etc. The bus 302 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0135] The memory 303 can be a ROM or other type of static storage device that can store static information and instructions, a RAM or other type of dynamic storage device that can store information and instructions, or an EEPROM, a CD-ROM or other optical disk storage, an optical disk storage (including a compressed optical disk, a laser disk, an optical disk, a digital versatile disk, a Blu-ray disk, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited to these.

[0136] Computer Readable Storage Medium Embodiments

[0137] The present invention proposes a computer-readable storage medium on which a computer program is stored. The computer program can be loaded and executed by a processor to implement a memory hotspot elimination method based on memory pooling technology as described in the first aspect.

[0138] The applicant of the present invention has made a detailed explanation and description of the implementation examples of the present invention in conjunction with the drawings in the specification. However, those skilled in the art should understand that the above implementation examples are only preferred implementation schemes of the present invention, and the detailed description is only to help readers better understand the spirit of the present invention, but not to limit the scope of protection of the present invention. On the contrary, any improvements or modifications based on the inventive spirit of the present invention should fall within the scope of protection of the present invention.

[0139] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention, rather than to limit them. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the above embodiments, or replace some of the technical features therein by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Any changes or replacements that can be easily thought of by those skilled in the art within the technical scope disclosed in the present invention should be covered within the protection scope of the present invention.

Claims

1. A memory hotspot elimination method based on memory pooling technology, characterized in that: include: S1, obtain the load of local memory; S2, dividing the load into a delay-sensitive load and a delay-insensitive load; S3, determining whether the latency-insensitive load has a memory hotspot; if so, proceeding to the next step; Otherwise, end the process; S4, determine whether the current environment supports the Intel-RDT environment; if so, control the bandwidth of latency-insensitive loads with memory hotspots through the memory bandwidth resource allocation sub-tool; otherwise, perform cold data migration on latency-insensitive loads with memory hotspots through the page fault handler in the kernel.

2. According to the memory hotspot elimination method based on memory pooling technology according to claim 1, it is characterized in that: After S2, the method further includes: S5: When only latency-sensitive loads exist on the non-uniform memory access node, the memory pages are sorted in ascending order by access frequency and a preset proportion of memory pages are migrated to the CXL memory pool.

3. The memory hotspot elimination method based on memory pooling technology according to claim 1 is characterized in that: The determining in S3 whether the latency-insensitive load has a memory hotspot specifically includes: In an environment supporting Intel-RDT, the memory bandwidth of the load is monitored by using an RDT tool; Determine whether the memory bandwidth exceeds a first threshold; if so, determine that a memory hotspot exists in the load.

4. The memory hotspot elimination method based on memory pooling technology according to claim 1 is characterized in that: The determining in S3 whether the latency-insensitive load has a memory hotspot specifically includes: In an environment that does not support Intel-RDT, real-time memory access latency is obtained by periodically sending a small number of non-temporal access instructions to the local memory; Determine whether the real-time memory access latency exceeds a second threshold; if so, determine that a memory hotspot exists in the load.

5. The memory hotspot elimination method based on memory pooling technology according to claim 4 is characterized in that: In S4, the page table entry accessed again by the latency-insensitive load with the memory hotspot is processed by the page fault handler in the kernel, specifically including: S401, periodically sampling the load memory access through PEBS; S402, modifying the memory page table of the load according to the memory address recorded by the sample, and marking the corresponding page table entry as an invalid page table entry; S403, when the load accesses the memory page again, access processing is performed through the page fault handler in the kernel; S404, sorting in ascending order according to access frequency, and classifying the memory data in a preset number of memory addresses as cold data; S405: Migrate the cold data to the CXL memory pool.

6. A memory hotspot elimination device based on memory pooling technology, characterized in that: The device comprises: Acquisition module, used to obtain the load of local memory; A division module, used for dividing the load into a delay-sensitive load and a delay-insensitive load; The first judgment module is used to judge whether the latency-insensitive load has a memory hotspot; if so, enter the second judgment module; otherwise, end the process; The second judgment module is used to determine whether the current environment supports the Intel-RDT environment; if so, the bandwidth of the latency-insensitive load with memory hotspots is controlled through the memory bandwidth resource allocation sub-tool; otherwise, cold data migration is performed on the latency-insensitive load with memory hotspots through the page fault handler in the kernel.

7. The memory hotspot elimination device based on memory pooling technology according to claim 6 is characterized in that: The device also includes: The migration module is used to migrate a preset proportion of memory pages to the CXL memory pool in ascending order of access frequency when only latency-sensitive loads exist on the non-uniform memory access node.

8. The memory hotspot elimination device based on memory pooling technology according to claim 6 is characterized in that: The first determination module is used for: In an environment supporting Intel-RDT, the memory bandwidth of the load is monitored by using an RDT tool; Determining whether the memory bandwidth exceeds a first threshold; If so, it is determined that a memory hotspot exists in the load.

9. The memory hotspot elimination device based on memory pooling technology according to claim 6, characterized in that: The first determination module is used for: In an environment that does not support Intel-RDT, real-time memory access latency is obtained by periodically sending a small number of non-temporal access instructions to the local memory; Determining whether the real-time memory access latency exceeds a second threshold; If so, it is determined that the load has a memory hotspot.

10. The memory hotspot elimination device based on memory pooling technology according to claim 9, characterized in that: The second judgment module is used for: Periodically sample load memory accesses through PEBS; According to the memory address recorded by the sample, the memory page table of the load is modified, and the corresponding page table entry is marked as an invalid page table entry; When the load accesses the memory page again, the access is processed by the page fault handler in the kernel; Arrange in ascending order according to access frequency, and classify the memory data in a preset number of memory addresses as cold data; Migrate the cold data to the CXL memory pool.