WO2014206078A1

WO2014206078A1 - Memory access method, device and system

Info

Publication number: WO2014206078A1
Application number: PCT/CN2014/071252
Authority: WO
Inventors: 褚力行
Original assignee: 华为技术有限公司
Priority date: 2013-06-25
Filing date: 2014-01-23
Publication date: 2014-12-31
Also published as: CN103365717B; CN103365717A

Abstract

Provided are a memory access method, device and system. The memory access method comprises: a node controller receiving monitoring information sent by an operating system, wherein the monitoring information carries information about a monitored memory in a first node to which the node controller belongs, the monitored memory is a memory resource occupied by a target process on the first node, and the target process is a process which operates on a central processing unit (CPU) of the first node and accesses a memory of an accessed node other than the first node in a server system; if it is monitored that the frequency of access to the memory of the accessed node by the target process occupying the monitored memory is greater than or equal to a threshold value, then sending information about the accessed node to the operating system, so as to migrate the target process to the accessed node according to the information about the accessed node; and converting the remote memory access into local memory access or adjacent memory access, so that the time it takes for the target process to access the memory can be reduced, thereby effectively improving the performance of the server system.

Description

Memory access method, device and system

TECHNICAL FIELD Embodiments of the present invention relate to computer technologies, and in particular, to a memory access method, apparatus, and system. BACKGROUND OF THE INVENTION With the development of computer technology, a server system can be composed of one or more servers, and each server acts as a node to form a non-uniform memory access (NUMA) architecture.

In the server system of the N medical A architecture, each node may include one or more central processing units (CPUs), and each CPU may be configured with a certain memory resource in advance. The process running on the CPU can use the following three methods to access the memory resources in the server system, and the other ^one is local (local) memory access, 4 near (buddy) memory access, and remote (remote) memory access.

The method of accessing the CPU's own memory resources by the process running on the CPU, that is, the local memory access; the method running on the CPU accesses the memory resources of other CPUs in the node to which the CPU belongs, that is, the adjacent memory access; The running process accesses the memory resources of the CPUs in other nodes other than the node to which the CPU belongs, that is, the remote memory access.

Among the three memory access methods, the process running on the CPU takes a long time to access the remote memory, which may be 3-20 times longer than the time required for local memory access. Therefore, the CPU in the existing server system When a running process accesses a memory resource by using a remote memory access method, the overall performance of the system is degraded. SUMMARY OF THE INVENTION Embodiments of the present invention provide a memory access method, apparatus, and system for improving overall performance of a server system. A first aspect of the embodiments of the present invention provides a memory access method, including: receiving, by a node controller, monitoring information sent by an operating system, where the monitoring information carries a monitored memory in a first node to which the node controller belongs The monitored memory is a memory resource occupied by the target process on the first node, and the operating system runs in a server system consisting of at least two nodes including the first node. And the target process is a process running on a central processing unit CPU of the first node and accessing a memory of the accessed node other than the first node in the server system;

If the node controller monitors that the frequency of accessing the memory of the accessed node by the target process occupying the monitored memory is greater than or equal to a threshold, sending the information of the accessed node to the An operating system, such that the operating system migrates the target process to the accessed node according to information of the accessed node.

With reference to the memory access method provided by the first aspect, in a first possible implementation manner, the threshold is a memory that is running on all CPUs in the first node in a preset time, and is in memory of other nodes. The ratio of the number of accesses to the total number of CPUs in the first node.

With reference to the first aspect or the first possible implementation manner, in a second possible implementation manner, the node controller is a node controller NC chip;

Correspondingly, the receiving, by the node controller, the monitoring information sent by the operating system includes:

The node controller NC chip receives the monitoring information sent by the operating system by using a motherboard management control unit of the first node;

Correspondingly, if the node controller monitors that the frequency of accessing the memory of the accessed node by the target process occupying the monitored memory is greater than or equal to a threshold, the information of the accessed node is Sending to the operating system includes:

The node controller NC chip, if the frequency of accessing the memory of the accessed node by the target process occupying the monitored memory is greater than or equal to the threshold, the node being accessed The information of the accessed memory is sent to the operating system by the mainboard management control unit.

A second aspect of the embodiments of the present invention provides a node controller, including: a receiving unit, configured to receive monitoring information sent by an operating system, where the monitoring information carries information about monitored memory in a first node to which the node controller belongs, where the monitored memory is a target process at the first node An occupied memory resource, the operating system running on each node in a server system consisting of at least two nodes including the first node, where the target process is in the middle of the first node a process running on the processor CPU and accessing memory of the accessed node other than the first node in the server system;

a monitoring unit, configured to send, when the frequency of accessing the memory of the accessed node by the target process that monitors the monitored memory is greater than or equal to a threshold, sending information of the accessed node to the An operating system, to cause the operating system to migrate the target process to the accessed node according to information of the accessed node.

In conjunction with the node controller provided in the second aspect, in a first possible implementation manner, the threshold is a memory that is running on all CPUs in the first node in a preset time, and is in memory of other nodes. The ratio of the number of accesses to the total number of CPUs in the first node.

With reference to the second aspect or the first possible implementation manner, in a second possible implementation manner, the receiving unit is specifically configured to:

Receiving, by the mainboard management control unit of the first node, the monitoring information sent by the operating system;

The monitoring unit is specifically configured to:

When the frequency of accessing the memory of the accessed node by the target process occupying the monitored memory is greater than or equal to the threshold, the information of the accessed memory in the accessed node is The main board management control unit sends to the operating system.

A third aspect of the embodiments of the present invention provides a server system, including at least two nodes including a node controller; and an operating system running in the server system;

If the operating system determines that the target process exists, the monitoring information is sent to the first node where the target process is running, where the monitoring information carries the monitored memory in the first node to which the node controller belongs. The monitored memory is a memory resource occupied by the target process on the first node, and the target process is on a central processing unit CPU of the first node. a process running, and accessing memory of the accessed node other than the first node in the server system;

After the operating system receives the information of the accessed node that is accessed by the target process by the first node, the operating system migrates the target process to the accessed node.

In conjunction with the server system provided by the third aspect, in a first possible implementation manner, if the operating system determines that the CPU of the accessed node belongs to the CPU to which the accessed memory belongs, the memory resource has the target process running Capabilities, the target process is migrated to the CPU to which the accessed memory belongs; if it is determined that the CPU to which the accessed memory belongs, the memory resource does not have the ability to run the target process, and the accessed node The memory resources of other CPUs in the CPU have the ability to run the target process, and then migrate the target process to other CPUs of the accessed node.

The memory access method, device and system provided by the embodiment of the present invention, after receiving the monitoring information sent by the operating system, the node controller monitors the frequency of the remote access of the memory of the accessed node by the process occupying the monitored memory. If the value is greater than or equal to the threshold, the information of the accessed node is sent to the operating system, so that the operating system migrates the target process to the accessed node; and the remote memory access is converted into a local memory access or a neighboring memory access, thereby enabling Reduce the time that the target process accesses memory, effectively improving the performance of the server system. BRIEF DESCRIPTION OF DRAWINGS FIG. 1 is a flowchart of a memory access method according to an embodiment of the present invention;

2 is a flowchart of another memory access method according to an embodiment of the present invention;

3 is a schematic structural diagram of a node controller according to an embodiment of the present invention;

4 is a schematic structural diagram of another node controller according to an embodiment of the present invention;

FIG. 5 is a schematic structural diagram of a server system according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS In order to improve the performance of a server system when a remote memory access method is used in a process of the present invention, a memory access method for migrating a process that satisfies certain conditions is used. The way memory access is converted to local memory access or proximity memory access.

Embodiments of the present invention can be applied to a multi-server system composed of at least two nodes, each of which includes a node controller according to various embodiments of the present invention. Each node can be a server, and each server can include one or more CPUs, and each CPU can be allocated a part of memory resources correspondingly. The operating system (Operational Edition) can be run on the server system.

FIG. 1 is a flowchart of a memory access method according to an embodiment of the present invention. As shown in FIG. 1, the method includes:

1 01. The node controller receives the monitoring information sent by the operating system.

The monitoring information carries information about the monitored memory in the first node to which the node controller belongs, where the monitored memory is a memory resource occupied by the target process on the first node, and the operating system runs. On each of the server systems consisting of at least two nodes including the first node, the target process is run on a central processor CPU of the first node, and the server is The process of accessing the memory of the accessed node other than the first node in the system.

Specifically, the operating system runs on each node of the server system, and each node may run one or more processes on the CPU, and the processes may use local memory access, adjacent memory access, or remote memory access for memory access. .

The embodiments of the present invention are mainly directed to the process of performing remote memory access. Therefore, the "target process" described in the embodiments of the present invention refers to the process of performing remote memory access. That is, the target process accesses memory resources in other nodes other than the node to which the CPU it is running belongs to. The "first node" described in the embodiments of the present invention is a node to which the node controller of the execution subject belongs, so as to be distinguished from other nodes to avoid confusion.

The operating system can manage each node and the processes running in each node from a global perspective, so the operating system can know the running status of the processes in each node. For example, which processes are running on the node, which CPU the processes are running on, and which part of the memory is occupied by which CPU. The operating system can know the status of each process accessing the memory, including which part of the memory of which node the process accesses. Therefore, when the process performs remote memory access The operating system is aware of this situation.

The operating system sends monitoring information to the node controller on the first node running the target process when it knows that there is a target process in the process running by the server system. The monitoring information includes information about the memory resources occupied by the target process on the first node. In the embodiments of the present invention, the part of the memory resources is referred to as “monitored memory”.

When the node controller monitors that the frequency of accessing the memory of the accessed node by the target process occupying the monitored memory is greater than or equal to a threshold, the information of the accessed node is sent to the operating system. Step 102 is performed to enable the operating system to migrate the target process to the accessed node according to the information of the accessed node.

Specifically, after receiving the monitoring information, the node controller monitors the monitored memory indicated in the monitoring information. The first node is able to know the situation in which the process running on it accesses the memory. The node controller may not directly monitor the process, because the process running on the first node will locally occupy part of the memory on the running CPU, so the node controller can monitor this part of the memory.

Moreover, when the process on the first node accesses the memory on the other node, the node controller can also learn the information of the accessed memory on the other node, and the memory accessed by the target process in the embodiments of the present invention. The node to which it belongs is called the "accessed node"; the memory resource that is located on other nodes and accessed by the target process is called "accessed memory".

The node controller monitors the progress of the processes running in the monitored memory and the memory of other nodes. Specifically, it can monitor the frequency at which the target process accesses the memory of other nodes.

The statistical period and threshold can be preset in the node controller. The node controller monitors the number of times the target node accesses the memory of the accessed node during each statistical period, and obtains the frequency value. If the node controller determines that the frequency value is greater than or equal to the threshold, it indicates that the target process currently uses the remote memory access mode. In order to improve the performance of the server system, in this case, the node controller sends the information of the accessed node to the operating system. The information of the accessed node may include information of the accessed memory accessed by the target process in the accessed node. However, due to operation The system can know which memory resources are accessed when the target process performs remote access. Therefore, it is not necessary to carry the information of the accessed memory in the information of the accessed node.

After receiving the information of the accessed node, the operating system may migrate the target process to the accessed node, so that after the migration, the target process can access the memory of the accessed node by using local access or proximity access. Thereby, the system performance can be effectively improved.

Further, the operating system may not directly migrate the target process. After determining the remaining condition of the memory resource in the accessed node, when the remaining memory resource is sufficient to run the target process, the target process is migrated. To better improve system performance.

The method for the operating system to migrate the process may use an implementation similar to that in the prior art, and will not be repeated here.

The memory access method provided by the embodiment of the present invention, after receiving the monitoring information sent by the operating system, if the target controller monitors the target process occupying the monitored memory, the frequency of remote access to the accessed node's memory is greater than or equal to The threshold value, the information of the accessed node is sent to the operating system, so that the operating system migrates the target process to the accessed node; the remote memory access is converted into a local memory access or a neighboring memory access, thereby reducing the target The time that the process accesses the memory effectively improves the performance of the server system.

2 is a flowchart of another memory access method according to an embodiment of the present invention. As shown in FIG. 2, the method includes:

201. Node controller in the first node The NC chip receives the monitoring information sent by the operating system through the main board management control unit.

Specifically, based on the implementation manner in step 101, communication between the NC chip and the operating system is implemented by a motherboard management control unit disposed on the first node.

202. The NC chip in the first node, when the frequency of accessing the memory of the accessed node by the target process occupying the monitored memory is greater than or equal to the threshold, is accessed. Information of the accessed memory in the node is sent to the operating system by the mainboard management control unit.

For details, refer to the implementation manner described in step 102. Further, in the server system of the NUMA architecture, when the number of nodes is large, if the nodes are connected by direct interconnection, the connection relationship between the nodes will be limited by the number of interfaces or interconnections on the nodes. The communication protocol allows the nodes to be interconnected using a Node Controller (NC) chip. Each node is provided with an NC chip, and the nodes can be communicably connected through respective NC chips. The CPUs on the two nodes can realize high-speed interconnection through the NC chip, and the connections between the nodes are not required. It is then limited by the number of interfaces on the node. The NC chip has a function of caching, such as causality checking and message forwarding.

The operating system can read and write to the registers on the NC chip on the node through the Baseboard Management Controller (BMC) on the node. BMC supports industry-standard intelligent platform management interfaces (Intelligent Platform Management

Interface, IPMI) specification. This specification describes the management features already built into the motherboard, including local and remote diagnostics, console support, configuration management, hardware management, and troubleshooting. IPMI is an open, free standard for server management system design. I P M I provides a variety of system interfaces, and the Keyboard Controller Style (KCS) is the most widely used IPMI system interface. The operating system can communicate with the BMC through the KCS. The BMC can access the registers of the NC chip through the Inter-Integrated Circuit (IIC) to perform read and write operations, thereby monitoring or controlling the state of the NC chip.

The NC chip on the first node monitors the monitored memory and monitors the access of the target process running on the monitored memory to the memory of other nodes.

Further, the threshold is a ratio of the number of times the memory of the other node is accessed to the total number of CPUs in the first node in the process running on all the CPUs in the first node in a preset time.

Specifically, the threshold can be set as needed, and a typical setting is as follows.

Presetting a time length in the first node, counting the processes running on all the CPUs connected to the NC chip on the first node in the preset time, and using the number of remote memory accesses, the number of times is included in other nodes. Access to shared memory also includes access to non-shared memory in other nodes.

The ratio of the number of times to the number of CPUs on the first node is taken as the threshold. That is, If the NC chip of the first node monitors the target process running on the monitored memory, the number of remote memory accesses is greater than or equal to the average value of the remote memory access on a single CPU, indicating that the target process performs remote memory access. Visits are more frequent and the target process needs to be migrated.

For example, for example, the number of CPUs in a node is eight, and the number of requests for remote memory access through the node's NC chip is 1000 times in a preset time, that is, all processes running on eight CPUs are performed in total. 10,000 remote memory accesses. It can be calculated that the average value of remote access per CPU within a preset time is 1 25 times.

Thus, the threshold is set to 1 25 . When the number of remote memory accesses in the statistics period is greater than or equal to 1 25, the target process accesses the remote memory more frequently and needs to be migrated.

FIG. 3 is a schematic structural diagram of a node controller according to an embodiment of the present invention. As shown in FIG. 3, the node controller includes:

The receiving unit 1 1 is configured to receive monitoring information sent by the operating system, where the monitoring information carries information about the monitored memory in the first node to which the node controller belongs, where the monitored memory is the target process in the first a memory resource occupied by a node, the operating system running on each node in a server system consisting of at least two nodes including the first node, the target process is at the first node a process running on the central processing unit CPU and accessing the memory of the accessed node other than the first node in the server system;

The monitoring unit 1 2 is configured to: when the frequency of accessing the memory of the accessed node by the target process that monitors the monitored memory is greater than or equal to a threshold, send information of the accessed node to the An operating system, such that the operating system migrates the target process to the accessed node according to information of the accessed node.

Further, the threshold value of the monitoring unit 12 for comparing the frequency of accessing the memory of the accessed node by the target process occupying the monitored memory, the threshold is The ratio of the number of times the memory of the other node is accessed to the total number of CPUs in the first node in the processes running on all the CPUs in the first node within a preset time.

Further, the receiving unit 1 1 is specifically configured to: Receiving, by the mainboard management control unit of the first node, the monitoring information sent by the operating system;

The monitoring unit 12 is specifically configured to:

4 is a schematic structural diagram of another node controller according to an embodiment of the present invention. As shown in FIG. 4, the node controller includes:

The processor 21, the memory 22, the bus 23, and the communication interface 24. The processor 21, the memory 22 and the communication interface 24 are connected by a bus 23 and communicate with each other.

The processor 21 may be a single core or multi-core central processing unit (CPU), or an Application Specific Integrated Circuit (ASIC), or one or more configured to implement the embodiments of the present invention. integrated circuit.

The memory 11 can be a high speed RAM memory or a nonvolatile memory.

(non-volatile memory), such as at least one disk storage.

The memory 22 is used to store the program 221. Specifically, the program 221 may include program code, where the program code includes computer operation instructions.

The communication interface 24 is configured to receive monitoring information sent by the operating system, where the monitoring information carries information about the monitored memory in the first node to which the node controller belongs, where the monitored memory is the target process at the first node. An occupied memory resource, the operating system running on each node in a server system consisting of at least two nodes including the first node, where the target process is in the middle of the first node A process running on a processor CPU and accessing memory of a visited node other than the first node in the server system.

The processor 21 runs the program 221 to execute:

Sending information of the accessed node to the operating system, so that the frequency of accessing the memory of the accessed node by the target process of the monitored memory is greater than or equal to a threshold The operating system migrates the target process to the location according to the information of the accessed node Said access node.

Specifically, the method for performing the memory access by the node controller provided by the embodiments of the present invention may be referred to the operation steps described in the foregoing method embodiments, and details are not described herein again.

FIG. 5 is a schematic structural diagram of a server system according to an embodiment of the present invention. As shown in FIG. 5, the server system includes at least two nodes, respectively, node 1 and node 2 including a node controller as shown in FIG. 3 or FIG. .

An operating system is run in the server system; if the operating system determines that the target process exists, the operating system sends monitoring information to the first node controller where the target process is running, where the monitoring information carries the node Information about the monitored memory in the node to which the first node controller belongs to the controller, the monitored memory is a memory resource occupied by the target process on the first node, and the target process is in the first node a process running on the central processing unit CPU and accessing the memory of the accessed node other than the first node in the server system;

For details, refer to the implementations described in steps 1 01 - 1 02 or 2 01 - 02. Further, in practical applications, the server system can implement access to the memory in the following manner.

When the operating system is running, if it is determined that a process has created shared memory, that is, the shared memory is used as monitored memory, the shared memory monitoring module is started. The shared memory monitoring module is configured to determine whether the process accesses memory resources of other nodes other than the node to which it belongs; if yes, the operating system starts the OS information reporting module. The OS information reporting module is configured to send monitoring information to the BMC.

The monitoring information includes a CPU number, a node number, and information of the monitored memory. The CPU number is the unique identifier of the CPU in the entire server system; the node number is the unique identifier of the node in the entire server system; the information of the monitored memory is the description of the memory area that the NC chip needs to monitor, usually several groups of memory segments. The start and end addresses of the address. For example, 1 KB memory space, there may be 2 segments in physical memory addressing, one segment is 0x 0- 0x2 00, and the other segment is 0x400- 0x6 00, then the description letter The information should be 2 sets of descriptions, ie 0x 0- 0x 200 and 0x4 00- 0x6 00.

The communication channel between the operating system and the BMC can be selected as a KCS or a single-block transmission (BT). In the monitoring process, the OS information reporting module sends information to the BMC, and the BMC feeds back information and migrates to the operating system. Instructions, etc., are sent by KCS or BT.

After receiving the monitoring information, the BMC converts the monitoring information into the value of a specific register in the NC chip by using the conversion module, and writes the information into the monitoring target register of the NC chip through the register reading and writing module, and is monitored by the NC chip.

Based on these register information, the NC chip monitors the frequency of access to the memory on other nodes by the process running on the monitored memory through the monitoring module of its network subsystem. When the frequency exceeds the threshold, the feedback module alarm is started, and the CPU number of the monitored CPU and the monitored memory information are sent to the BMC, which is forwarded to the operating system by the BMC.

Among them, in the NC chip, it is necessary to increase the threshold register for storing the threshold; the monitoring register group is used to store the monitored CPU number and node number corresponding to the OS information reporting module, and the memory segment register is used to store The memory segment information of the memory may also be included, and may also include a target memory access counter for counting the number of times the process performs remote memory access; adding an alarm register for indicating whether migration is required; adding a feedback register for providing to the BMC The CPU number, node number, and memory segment information corresponding to the process of the migration, and the content information indicated therein corresponds to the contents of the monitoring register.

In addition, the operating system can set the threshold stored in the threshold register through the BMC. The NC chip can periodically compare the target memory access counter with the value in the threshold register to determine whether the number of times the process performs remote memory access exceeds the threshold. The period of time can be set as needed, such as 60 seconds or 120 seconds, but the optional values are not limited to this.

The BMC reads the value in the register of the NC chip by timing. If it is determined that the alarm register is set, the value in the feedback register is read, and the value of the register is converted into the corresponding CPU number and memory section by the conversion module. Information, combined with instructions that need to be migrated, is sent to the operating system. The time interval for the BMC to read the alarm register can be set as needed, for example, 60 seconds or 120 seconds, but the optional values are not limited to this. After receiving the migration instruction, the operating system determines which process needs to be migrated according to the received CPU number and the memory segment information, and implements the migration of the process through the process migration module.

Further, if the operating system determines that the accessed memory in the accessed node belongs to

a CPU having a memory resource having the capability of running the target process, and migrating the target process to a CPU to which the accessed memory belongs; if it is determined that the CPU to which the accessed memory belongs, the memory resource does not have a running location The capability of the target process, and the memory resources of other CPUs in the accessed node have the ability to run the target process, and then migrate the target process to other CPUs of the accessed node.

Specifically, in an actual application, after receiving the migration instruction, the operating system may further determine whether the destination node, that is, the accessed node, has sufficient memory resources to be allocated to the target process.

If the remaining memory resources in the accessed memory in the destination node are sufficient for allocation to the target process, i.e., the ability to run the target process, the operating system may migrate the target process to the portion of memory of the destination node.

If the remaining memory resources in the accessed memory in the destination node are not sufficient for allocation to the target process, that is, the capability to run the target process, the migration may not be performed temporarily. The memory may also be detected periodically, and when the remaining memory resources are sufficient, the operating system may migrate the target process to the portion of the memory of the destination node.

The operating system can also determine whether the remaining memory resources in the memory of other CPUs in the destination node are sufficient for allocation to the target process. If yes, the target process can be temporarily migrated to the partial memory, and the memory is periodically accessed. Detecting, when the remaining memory resources in the accessed memory are sufficient for allocation to the target process, the target process is migrated to the accessed memory.

The memory access method, device and system provided by the embodiments of the present invention can fully utilize the CPU and memory resources of the migration destination node through process migration, thereby improving resource utilization; and at the same time, due to process migration, processes in the node are If the CPU that has been moved out and the process is moved out has no other processes that need to be run, the energy consumption of the node can be reduced and energy can be saved.

One of ordinary skill in the art can understand that all or part of the steps of the foregoing method embodiments are implemented. The steps can be completed by the relevant hardware of the program instructions. The aforementioned program can be stored in a computer readable storage medium. When the program is executed, the steps including the foregoing method embodiments are performed; and the foregoing storage medium includes: various media that can store program codes, such as ROM, RAM, disk or optical disk.

Finally, it should be noted that the above embodiments are only for explaining the technical solutions of the present invention, and are not intended to be limiting thereof; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art will understand that The technical solutions described in the foregoing embodiments may be modified, or some or all of the technical features may be equivalently replaced; and the modifications or substitutions do not deviate from the technical solutions of the embodiments of the present invention. range.

Claims

Rights request

1. A memory access method, characterized by including:

The node controller receives monitoring information sent by the operating system. The monitoring information carries information about the monitored memory in the first node to which the node controller belongs. The monitored memory is occupied by the target process on the first node. memory resources, the operating system runs on each node in the server system consisting of at least two nodes including the first node, and the target process is the central processor of the first node A process running on the CPU and accessing the memory of an accessed node other than the first node in the server system;

If the node controller monitors that the frequency of accessing the memory of the accessed node by the target process occupying the monitored memory is greater than or equal to a threshold, then the information of the accessed node is sent to the accessed node. The operating system is configured to enable the operating system to migrate the target process to the visited node according to the information of the visited node.

2. The memory access method according to claim 1, wherein the threshold value is the number of times the memory of other nodes is accessed by processes running on all CPUs in the first node within a preset time. The ratio to the total number of CPUs in the first node.

3. The memory access method according to claim 1 or 2, characterized in that the node controller is a node controller NC chip;

Correspondingly, the node controller receiving the monitoring information sent by the operating system includes: the node controller NC chip receives the monitoring information sent by the operating system through the mainboard management control unit of the first node;

Correspondingly, if the node controller monitors that the frequency of the target process occupying the monitored memory accessing the memory of the accessed node is greater than or equal to the threshold, the information of the accessed node will be Sent to the operating system includes:

The node controller NC chip monitors that the frequency of the target process occupying the monitored memory accessing the memory of the accessed node is greater than or equal to the threshold, then the accessed node is The information of the accessed memory is sent to the operating system through the mainboard management control unit.

4. A node controller, characterized by including:

A receiving unit, configured to receive monitoring information sent by the operating system. The monitoring information carries information about the monitored memory in the first node to which the node controller belongs. The monitored memory is the location of the target process on the first node. The operating system runs on each node in the server system consisting of at least two nodes including the first node, and the target process is in the center of the first node. A process running on the processor CPU and accessing the memory of an accessed node other than the first node in the server system;

A monitoring unit configured to send information about the accessed node to the accessed node when it is monitored that the target process of the monitored memory accesses the memory of the accessed node at a frequency greater than or equal to a threshold. Operating system, so that the operating system migrates the target process to the visited node according to the information of the visited node.

5. The node controller according to claim 4, wherein the threshold value is the number of times the memory of other nodes is accessed by processes running on all CPUs in the first node within a preset time. The ratio to the total number of CPUs in the first node.

6. The node controller according to claim 4 or 5, characterized in that the receiving unit is specifically used for:

Receive the monitoring information sent by the operating system through the mainboard management control unit of the first node;

The monitoring unit is specifically used for:

When it is monitored that the frequency of the target process occupying the monitored memory accessing the memory of the accessed node is greater than or equal to the threshold, the information of the accessed memory in the accessed node is passed through the The mainboard management control unit sends it to the operating system.

7. A server system, characterized in that it includes at least two nodes including the node controller according to any one of claims 4-6; an operating system runs in the server system; if the operating system determines If the target process exists, the monitoring information is sent to the first node where the target process runs. The monitoring information carries the information of the monitored memory in the first node to which the node controller belongs. Monitoring memory is the target process described in Section 1 The memory resources occupied by the node, the target process is a process running on the central processing unit CPU of the first node and accessing the memory of the accessed node other than the first node in the server system ;

After receiving the information of the visited node sent by the target process from the first node, the operating system migrates the target process to the visited node.

8. The server system according to claim 7, wherein if the operating system determines that the memory resource of the CPU to which the accessed memory in the accessed node belongs has the ability to run the target process, it will The target process migrates to the CPU to which the accessed memory belongs; if it is determined that the CPU to which the accessed memory belongs does not have the memory resources to run the target process, and other CPUs in the accessed node If the memory resource has the ability to run the target process, the target process will be migrated to other CPUs of the visited node.