WO2016119618A1

WO2016119618A1 - Remote memory allocation method, device and system

Info

Publication number: WO2016119618A1
Application number: PCT/CN2016/071483
Authority: WO
Inventors: 李辉
Original assignee: 华为技术有限公司
Priority date: 2015-01-27
Filing date: 2016-01-20
Publication date: 2016-08-04
Also published as: CN105988871B; CN105988871A

Abstract

A remote memory allocation method, and a corresponding device and system, the method comprising: receiving, by a memory node (305), a request transmitted by a computing node (301); allocating, by the memory node (305), a first kernel memory of XB to the computing node (301) from an idle kernel memory of the memory node (305); after the memory node (305) allocates the first kernel memory of XB to the computing node (301) from the idle kernel memory of the memory node (305), and before the computing node (301) reads and writes data from the remote memory, the method further comprising: mapping, by the memory node (305), the first kernel memory to a first storage space of the memory node (305), so as to allocate the remote memory to the computing node (301). The technical solution is free from a physical memory upper limit of the memory node (305) when allocating remote memory to the computing node (301), and allocates a larger remote memory to the computing node (301).

Description

Remote memory allocation method, device and system

This application claims priority to Chinese Patent Application No. 201510041110.7, entitled "A Remote Memory Allocation Method, Apparatus and System", filed on January 27, 2015, the entire contents of which are incorporated by reference. In this application.

Technical field

The present invention relates to the field of computer technologies, and in particular, to a remote memory allocation method, apparatus, and system.

Background technique

Physical memory is one of the important components in a computer, and it is a bridge to communicate with the CPU. All programs in the computer run in memory, so the performance of physical memory has a huge impact on the computer.

Currently, when the physical memory of the first terminal device is too small to meet the requirement, the first terminal device usually requests remote memory from other terminal devices to expand the physical memory of the first terminal device. In the prior art, other terminal devices (for the sake of convenience, the following "second terminal device" replaces "other terminal device") will divide the physical memory of the second terminal device into a part according to the request of the first terminal device. A terminal device to serve as the remote memory of the first terminal device. The disadvantage of this scheme is that since the physical memory size of the second terminal device is fixed, the remote memory allocated to the computing node cannot exceed the upper limit of the free physical memory in the second terminal device.

Summary of the invention

The present invention provides a remote memory allocation method, which is used to limit the allocation of remote memory to a computing node without being limited by the upper limit of physical memory in the memory node.

In a first aspect, an embodiment of the present invention provides a remote memory allocation method, where the method includes:

The memory node receives a request sent by the computing node, the request instructing the computing node to apply for remote memory of the XB to the memory node, where the memory node and the computing node are connected by network communication, the X Greater than zero and less than 2 ⁿ , the n is the data width of the memory node CPU general purpose register;

The memory node allocates the said node from the free kernel memory of the memory node The first kernel memory of XB;

After the memory node allocates the first kernel memory of the XB from the free kernel memory of the memory node for the computing node, and before the computing node reads and writes data to the remote memory, the method Also includes:

The memory node maps the first kernel memory to the first storage space of the memory node, so as to allocate the remote memory to the computing node, so that the computing node is to the first storage space Read and write data, the size of the first storage space is the XB; the first storage space is physical memory, or the first storage space is virtual memory, or the first storage space is physical memory And virtual memory.

In conjunction with the first aspect, in a first embodiment of the first aspect,

The request further indicates that the remote memory is a remote physical memory; or

The request further indicates that the remote memory is a remote virtual memory; or

The request further indicates that the remote memory is remote physical memory and remote virtual memory.

In conjunction with the first embodiment of the first aspect, in the second embodiment of the first aspect,

In the case where the request further indicates that the remote memory is remote physical memory,

The memory node maps the first kernel memory to the first storage space of the memory node, so as to allocate the remote memory to the computing node, so that the computing node is to the first storage space Reading and writing data, the size of the first storage space is the XB, and specifically includes:

The memory node maps the first kernel memory to a first physical memory of the memory node to implement allocating the remote physical memory to the computing node, thereby causing the computing node to the first physical The memory reads and writes data, and the size of the first physical memory is the XB.

With reference to the first aspect, the first embodiment of the first aspect or the second embodiment of the first aspect, in a third embodiment of the first aspect,

The request further indicates a first start address, where the first start address refers to an address of the remote memory start location facing the computing node;

The method further includes:

The memory node establishes a mapping relationship between the first start address and a second start address, where the second start address refers to an address of the first kernel memory start position in the free core memory.

With reference to the first aspect, the first embodiment of the first aspect or the third embodiment of the first aspect, in a fourth implementation manner of the first aspect,

The memory node allocates the first kernel memory of the XB to the computing node from the free kernel memory of the memory node, and specifically includes:

In a case where the free kernel memory of the memory node is greater than or equal to the XB, the memory node allocates the first kernel memory from the free core memory to the computing node.

In a second aspect, an embodiment of the present invention provides a remote memory allocation method, where the method includes:

The memory node receives a request sent by the computing node, where the request indicates that the computing node requests the remote memory of the XB from the memory node, where the X is greater than zero and less than 2 ⁿ , and the n is the memory node CPU a data width of the general purpose register, wherein the memory node and the computing node are connected by a network communication;

The memory node allocates, according to the request, the first logical memory of the XB from the idle logical memory of the first target process to the computing node, where the first target process refers to running in the memory node, a first interface that interacts with the computing node and can allocate a logical memory to the computing node through the first interface, where the idle logical memory of the first target process is greater than or equal to the XB;

After the memory node allocates the first logical memory of the XB from the idle logical memory of the first target process to the computing node, and before the computing node reads and writes data to the remote memory, The method also includes:

Mapping, by the memory node, the first logical memory to a first storage space of the memory node, to implement allocating the remote memory to the computing node, thereby causing the computing node to move to the first storage space Reading and writing data; wherein the size of the first storage space is the XB, the first storage space is physical memory and virtual memory, or the first storage space is physical memory, or the first The storage space is virtual memory.

In conjunction with the second aspect, in a first embodiment of the second aspect,

After the memory node receives the request sent by the computing node, and the memory node allocates the first logical memory of the XB from the idle logical memory of the first target process to the computing node according to the request, the method Also includes:

The memory node selects the first target process from among all processes running on the memory node.

In conjunction with the second aspect, in a second embodiment of the second aspect,

The memory node creates the first target process and allocates logical memory for the first target process.

With reference to the second aspect, the first embodiment of the second aspect, or the second embodiment of the second aspect, in the third embodiment of the second aspect,

With reference to the third implementation manner of the second aspect, in a fourth implementation manner of the second aspect, in the case that the request further indicates that the remote memory is a remote physical memory,

Mapping, by the memory node, the first logical memory to a first storage space of the memory node, to implement allocating the remote memory to the computing node, thereby causing the computing node to move to the first storage space Read and write data, including:

The memory node maps the first logical memory to a first physical memory of the memory node to implement allocating the remote physical memory to the computing node, thereby causing the computing node to the first physical The memory reads and writes data, and the size of the first physical memory is the XB.

In combination with the second aspect or the first embodiment of the second aspect to the fourth embodiment of the second aspect, in the fifth embodiment of the second aspect,

The method further includes:

The memory node establishes a mapping relationship between the first start address and a second start address, where the second start address refers to an address of the first logical memory start position in the idle logical memory .

With reference to any one of the second embodiment of the second aspect to the fifth embodiment of the second aspect, in the sixth embodiment of the second aspect,

The number of cores of the memory node is greater than the number of processes running the memory node before the first target process is created,

After the first target process is created, the number of processes running on the memory node does not exceed The number of cores of the memory node.

In a third aspect, an embodiment of the present invention provides a remote memory allocation device, where the device includes:

a receiving unit, configured to receive a request sent by the computing node, where the request instructs the computing node to apply for a remote memory of the XB to the memory node, where the X is greater than zero and less than 2 ⁿ , and the n is the CPU of the memory node The data width of the register;

An allocating unit, configured to allocate, by the free kernel memory of the memory node, the first kernel memory of the XB to the computing node;

After the allocating unit allocates the first kernel memory of the XB from the free kernel memory of the memory node to the computing node, and before the computing node reads and writes data to the remote memory,

a mapping unit, configured to map the first kernel memory to a first storage space of the memory node, to implement allocating the remote memory to the computing node, thereby causing the computing node to the first storage Spatially reading and writing data, the size of the first storage space is the XB; the first storage space is physical memory, or the first storage space is virtual memory, or the first storage space is physical Memory and virtual memory.

In conjunction with the third aspect, in a first embodiment of the third aspect,

In conjunction with the first embodiment of the third aspect, in the second embodiment of the third aspect,

The mapping unit is specifically configured to map the first kernel memory to the first physical memory of the memory node, so as to implement allocating the remote physical memory to the computing node, thereby causing the computing node to The first physical memory reads and writes data, and the size of the first physical memory is the XB.

With reference to the third aspect, the first implementation manner of the third aspect, or the second implementation manner of the third aspect, in the third implementation manner of the third aspect, the request further indicates a first start address, where A starting address is an address of the remote memory starting location facing the computing node;

The device also includes:

The establishing unit is configured to establish a mapping relationship between the first starting address and the second starting address, where the second starting address is an address of the first kernel memory starting position in the free core memory.

In a fourth aspect, an embodiment of the present invention provides a remote memory allocation device, where the device includes:

a receiving unit, configured to receive a request sent by the computing node, where the request instructs the computing node to apply for a remote memory of the XB to the memory node, where the X is greater than zero and less than 2 ⁿ , and the n is the memory node a data width of a CPU general purpose register, wherein the memory node and the computing node are connected by a network communication;

An allocating unit, configured to allocate, according to the request, the first logical memory of the XB from the idle logical memory of the first target process, where the first target process refers to running in the memory node, a process having a first interface that interacts with the computing node and capable of allocating logical memory to the computing node through the first interface, the idle logical memory of the first target process is greater than or equal to the XB;

After the allocating unit allocates the first logical memory of the XB from the idle logical memory of the first target process to the computing node, and before the computing node reads and writes data to the remote memory,

a mapping unit, configured to map the first logical memory to a first storage space of the memory node, to implement allocating the remote memory to the computing node, so that the computing node is to the first storage Spatially reading and writing data, the size of the first storage space is the XB; the first storage space is physical memory and virtual memory, or the first storage space is physical memory, or the first storage The space is virtual memory.

In conjunction with the fourth aspect, in a first embodiment of the fourth aspect,

After the receiving unit receives the request sent by the computing node, and the allocating unit allocates the first logical memory of the XB from the idle logical memory of the first target process to the computing node according to the request, The device also includes:

a selection unit for selecting the first target process from among all processes running on the memory node.

With reference to the fourth aspect, in a second implementation manner of the fourth aspect,

Creating a unit for creating the first target process and assigning logic to the first target process RAM.

With reference to the fourth aspect, the first embodiment of the fourth aspect, or the second embodiment of the fourth aspect, in the third embodiment of the fourth aspect,

With the third implementation manner of the fourth aspect, in a fourth implementation manner of the fourth aspect, in the case that the request further indicates that the remote memory is a remote physical memory,

The mapping unit is specifically configured to map the first logical memory to the first physical memory of the memory node, so as to implement allocating the remote physical memory to the computing node, thereby causing the computing node to The first physical memory reads and writes data, and the size of the first physical memory is the XB.

In combination with the fourth aspect or the fourth embodiment of the fourth aspect to the fourth embodiment of the fourth aspect, in the fifth embodiment of the fourth aspect,

The device also includes:

a establishing unit, configured to establish a mapping relationship between the first starting address and a second starting address, where the second starting address refers to a starting position of the first logical memory in the idle logical memory address.

With reference to any one of the second embodiment of the fourth aspect to the fifth embodiment of the fourth aspect, in the sixth embodiment of the fourth aspect,

After the first target process created by the creation unit, the number of processes running on the memory node does not exceed the number of cores of the memory node.

In a fifth aspect, an embodiment of the present invention provides a remote memory allocation system, where the system includes a computing node, a memory node, and a third implementation manner of the first to third aspects of the third aspect or the third aspect. The remote memory allocation device of any one of the embodiments:

The computing node is configured to send the request received by the remote memory allocation device;

The memory node is configured to provide the first storage space;

The computing node is further configured to acquire the first storage space as the remote memory.

In conjunction with the fifth aspect, in a first implementation of the fifth aspect, the remote memory allocation device is integrated in the memory node.

In a sixth aspect, an embodiment of the present invention provides a remote memory allocation system, where the system includes a computing node, a memory node, and a sixth implementation manner of the first to fourth aspects of the fourth aspect or the fourth aspect. The remote memory allocation device of any one of the embodiments:

The memory node is configured to provide the first storage space;

In conjunction with the sixth aspect, in a first embodiment of the sixth aspect,

The remote memory allocation device is integrated in the memory node. It can be seen that, in the remote memory allocation method provided by the embodiment of the present invention, the memory node allocates the first kernel memory from the free kernel memory of the memory node according to the request of the computing node, and then maps the first kernel memory to The first storage space is configured to allocate remote memory for the compute node. As a common knowledge in the art, in the memory node, the kernel memory is the memory allocated by the operating system for the kernel object, and is the logical memory. The size of the kernel memory is 2 ⁿ B, where n is the data width of the general-purpose register of the memory node CPU. That is, the kernel memory is much larger than the physical memory of the memory node. Based on the virtual memory technology, the kernel memory can be mapped not only to physical memory but also to virtual memory (ie, virtual memory expanded from external memory). Therefore, with the technical solution provided by the embodiment of the present invention, when the computing node applies for the remote memory to the memory node, the computing node can apply for a larger remote memory to the memory node without being limited by the upper limit of the free physical memory in the memory node.

DRAWINGS

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments or the prior art description will be briefly described below. Obviously, the drawings in the following description are only some of the present invention. For the embodiments, those skilled in the art can obtain other drawings according to the drawings without any creative work.

FIG. 1 is a schematic diagram of an application scenario of a remote memory allocation method according to an embodiment of the present disclosure;

1b is a schematic flowchart of a remote memory allocation method according to an embodiment of the present invention;

1c is a schematic flowchart of a remote memory allocation method according to an embodiment of the present invention;

2a is a schematic structural diagram of a remote memory allocation device according to an embodiment of the present invention;

2b is a schematic structural diagram of another remote memory allocation device according to an embodiment of the present invention;

2c is a schematic structural diagram of still another remote memory allocation device according to an embodiment of the present invention;

FIG. 2d (1) is a schematic structural diagram of still another remote memory allocation device according to an embodiment of the present invention;

2d (2) is a schematic structural diagram of still another remote memory allocation device according to an embodiment of the present invention;

2 e is a schematic structural diagram of still another remote memory allocation device according to an embodiment of the present invention;

3a is a schematic structural diagram of a remote memory allocation system according to an embodiment of the present invention;

FIG. 3b is a schematic structural diagram of another remote memory allocation system according to an embodiment of the present invention.

detailed description

The technical solutions in the embodiments of the present invention are clearly and completely described in the following with reference to the accompanying drawings in the embodiments of the present invention. It is obvious that the described embodiments are a part of the embodiments of the present invention, but not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative efforts are within the scope of the present invention.

Before explaining the technical solution described in the present invention, two mature prior art technologies are clarified:

1. Implementing a mapping of logical addresses to physical addresses based on a memory paging mechanism. Specifically, based on the paging mechanism, each process can obtain an independent logical address space, thereby accessing the physical address through a logical address, and can provide protection of the logical address space. In order to improve the address translation performance, the general processor includes a Memeory Management Unit (MMU) and a Translation Lookside Buffer (TLB) to optimize memory access performance.

2, virtual memory. It refers to the dynamic expansion of physical memory based on virtual memory technology. Specifically, the expansion of physical memory is realized by using an external storage device (HDD/SDD/NVM, etc.).

As shown in FIG. 1a, a schematic diagram of an application scenario of a remote memory allocation method according to an embodiment of the present invention includes a memory node 11 and a computing node 13, and a high-speed interconnect interface between the memory node 11 and the computing node 13 Connected to implement information interaction between the memory node 11 and the compute node 13.

Embodiment 1

Referring to FIG. 1b, a flowchart of a remote memory allocation method according to an embodiment of the present invention is applied to the application scenario shown in FIG. 1a. Specifically, the following steps are included:

S101. The memory node receives a request sent by a computing node, where the request indicates that the computing node requests the remote memory of the XB from the memory node, where the memory node and the computing node are connected by using a network communication. Where X is greater than zero and less than 2 ⁿ , and n is the data width of the memory node CPU general purpose register;

It should be noted that the memory node in the embodiment of the present invention refers to a node having a physical storage space and a network communication function. The memory node is referred to as a memory node in the embodiment of the present invention for convenience of description, and should not be construed as a limitation. Provisions. Similarly, the computing node described in the embodiment of the present invention refers to a node having a computing capability and a network communication function. The computing node is also referred to as a computing node in the embodiment of the present invention, and should not be construed as a limitation.

In another embodiment of the present invention, the request further indicates that the remote memory is a remote physical memory; or the request further indicates that the remote memory is a remote virtual memory; or the request further indicates The remote memory is remote physical memory and remote virtual memory.

In still another embodiment of the present invention, the request further indicates that the first start address is further indicated, and the first start address refers to an address of the remote memory start location facing the computing node.

S103. The memory node allocates the first kernel memory of the XB from the free core memory of the memory node to the computing node.

It should be noted that the technical solution described in the embodiment of the present invention is directed to the case where the CPU of the memory node operates in a kernel mode. It should be understood that, the memory node allocates the first kernel memory of the XB to the computing node from the free kernel memory of the memory node, and specifically includes: the memory node determining whether the free kernel memory of the memory node is If the free memory is greater than or equal to the remote memory, and the memory node allocates the XB to the computing node from the free core memory. A kernel memory. It should be understood that in the case where the free kernel memory is smaller than the remote memory, the following steps are not continued, that is, the memory node does not allocate the remote memory for the computing node.

It should be noted that, in order to better implement the solution, the operating system of the memory node provides sufficient kernel memory for the kernel of the memory node when setting parameters for the kernel. For example, for a 64-bit CPU, the kernel memory can be configured to 8EB, which provides sufficient kernel memory for allocation to compute nodes.

S105. After the memory node allocates the first kernel memory of the XB from the free kernel memory of the memory node to the computing node, and before the computing node reads and writes data to the remote memory, The method also includes:

The memory node maps the first kernel memory to a first storage space of the memory node to The computing device allocates the remote memory, and the computing node reads and writes data to the first storage space, where the size of the first storage space is the XB; the first storage space is Physical memory, or the first storage space is virtual memory, or the first storage space is physical memory and virtual memory.

It is noted that after the memory node allocates the first kernel memory of the XB from the free kernel memory of the memory node to the computing node, the computing node is allocated a logical memory of size XB because Before the real storage space of the first kernel memory map, it is only a piece of logical memory. Since the logical memory cannot be used to read data, the computing node cannot read data into the first kernel memory. Referring to step S105, after the computing node is allocated the first kernel memory, and before the first kernel memory is mapped to the first storage space, the computing node does not have the first kernel memory. Read and write data. The computing node reads and writes data to the first storage space after the first kernel is memory mapped to the first storage space. It should be understood that the "memory node maps the first kernel memory to the first storage space of the memory node" in step S105 is a mapping of a logical address to a physical address based on a memory paging mechanism.

It should be noted that, in a case that the request further indicates that the remote memory is a remote physical memory, the memory node maps the first kernel memory to a first storage space of the memory node to implement Allocating the remote memory to the computing node, so that the computing node reads and writes data to the first storage space, where the size of the first storage space is the XB, specifically: the memory node will The first core memory is mapped to the first physical memory of the memory node, so as to allocate the remote physical memory to the computing node, so that the computing node reads and writes data to the first physical memory. The size of the first physical memory is the XB. It should be understood that, before the memory node maps the first kernel memory to the first physical memory of the memory node, the method of the embodiment of the present invention further includes: the memory node determining the memory node Whether the free physical memory is greater than or equal to the remote memory, and if the free physical memory is greater than or equal to the remote memory, the memory node allocates the calculated node from the free physical memory. The first physical memory. It should be understood that, in the case that the free physical memory is smaller than the remote memory, the following steps are not continued, that is, the memory node does not allocate the remote memory to the computing node.

It is further noted that the request further indicates that the remote memory is a remote physical memory and a remote virtual memory, and the size of the remote physical memory is KB, and the size of the remote virtual memory is XB. In the case of -KB, the memory node maps the first kernel memory to the first storage space of the memory node to implement allocation of the remote memory to the computing node, thereby causing the computing node to The first storage space reads and writes data, and the size of the first storage space is the XB, specifically: the memory node maps a portion of the first kernel memory with a size of KB to the memory node. a physical memory, the memory node mapping a portion of the first kernel memory having an size of XB-KB to the virtual memory of the memory node, thereby causing the computing node to the physical memory of the KB and the XB- Read and write data in KB's virtual memory. Similarly, before the memory node maps a portion of the first kernel memory with a size of KB to the physical memory of the memory node, the memory node further determines whether the free physical memory of the memory node is greater than or equal to The KB, and in a case where the free physical memory is greater than or equal to the KB, the memory node maps a portion of the first kernel memory having a size of KB to a physical memory of the memory node. It is noted that, in a case that the request further indicates the first start address, the technical solution provided by the embodiment of the present invention further includes: the memory node establishing between the first start address and the second start address Mapping relationship, the second starting address is an address of the first kernel memory starting position in the free core memory. It should be noted that, when the computing node to the distal end of the memory read data A, the data will be a starting position for the A and the data A, the address of the computing node A ₁ is sent to the memory a node, the memory node determining, according to the mapping relationship, an address A ₂ in the free kernel memory of the starting position of the data A, and mapping the A ₂ to the storage space A of the memory node _3, to write the data to the A ₃ in A. The A ₃ may be physical memory or virtual memory, or part of the A ₃ is physical memory, and the other part is virtual memory.

It can be seen that, in the technical solution provided by the embodiment of the present invention, the memory node allocates the first kernel memory from the free kernel memory of the memory node according to the request of the computing node, and then maps the first kernel memory to the first storage space. To achieve allocation of remote memory for compute nodes. As a common knowledge in the field, in the memory node, the kernel memory is the memory allocated by the operating system for the kernel object, the logical memory, the kernel memory is much larger than the physical memory of the memory node, and based on the virtual memory technology, the kernel memory can not only map. To physical memory, you can also map to virtual memory (that is, virtual memory that is extended from external memory). Therefore, when the computing node requests the remote memory from the memory node, the computing node is not limited by the upper limit of the free physical memory in the memory node, and the computing node can apply for a larger memory node. Remote memory.

Embodiment 2

Referring to FIG. 1c, a flowchart of a remote memory allocation method according to an embodiment of the present invention is applied to the application scenario shown in FIG. 1a. Specifically, the following steps are included:

S111. The memory node receives a request sent by the computing node, where the request indicates that the computing node requests the remote memory of the XB from the memory node, where the X is greater than zero and less than 2 ⁿ , and the n is the memory. a data width of a node CPU general purpose register, wherein the memory node and the computing node are connected by a network communication;

In still another embodiment of the present invention, the request further indicates a first start address, where the first start address refers to an address of the remote memory start location facing the computing node.

S113. The memory node allocates, according to the request, the first logical memory of the XB from the idle logical memory of the first target process, where the first target process refers to running in the memory node. a process having a first interface that interacts with the computing node and capable of allocating logical memory to the computing node through the first interface, the idle logical memory of the first target process is greater than or equal to the XB;

It is worth noting that when the program is called into memory by the operating system, the operating system allocates certain resources (such as logical memory, etc.) to the program, and then performs a series of complicated operations to make the program become a process for the system to call. For a process, the relationship between its logical memory and the number of bits of the CPU is usually: the logical memory of the process is 2 ^r B, where r is the number of bits of the CPU. Usually, the logical memory of a process is much larger than physical memory, and a large part of the logical memory of a process is often idle.

It should be noted that the first target process described in the embodiment of the present invention is selected from multiple processes running on the memory node, or the first target process is created by the memory node. That is, after the memory node receives the request sent by the computing node, and the memory node allocates the first logical memory of the XB from the idle logical memory of the first target process to the computing node according to the request, The method further includes: the memory node selecting the first target process from all processes running in the memory node; or the memory node creating the first target process and being the first The target process allocates logical memory. Preferably, the memory node creates the first target process in a case where the first target process does not exist in all processes running in the memory node.

In the case that a plurality of computing nodes request remote memory from the memory node, and the memory node needs to create two or more of the first target processes, it should be noted that the number of cores of the memory node is greater than that of the creation. In the case that the first target process runs the number of processes on the memory node, the method further includes: after the first target process is created, the number of processes running on the memory node does not exceed The number of cores of the memory node. This is to ensure the speed of the process. It should be known that if the number of processes running on the memory node exceeds the number of cores of the memory node, the process runs slower.

S115, after the memory node allocates the first logical memory of the XB from the idle logical memory of the first target process to the computing node, and before the computing node reads and writes data to the remote memory The method further includes:

It is noted that, after the memory node allocates the first logical memory of the XB from the idle logical memory of the first target process to the computing node, the computing node is allocated a logical memory of size XB. Because the virtual memory is only a piece of logical memory before the real storage space of the first logical memory map, since the logical memory cannot be used to read data, the computing node cannot read data to the first logical memory. . Referring to step S115, the computing node is assigned the first After a logical memory, and before the first logical memory is mapped to the first storage space, the computing node does not read or write data to the first logical memory. The computing node reads and writes data to the first storage space after the first logical memory is mapped to the first storage space. It should be understood that the “memory node maps the first logical memory to the first storage space of the memory node” in step S115 is a mapping of a logical address to a physical address based on a memory paging mechanism.

It should be noted that, in a case that the request further indicates that the remote memory is a remote physical memory, the memory node maps the first logical memory to a first storage space of the memory node to implement Allocating the remote memory to the computing node, so that the computing node reads and writes data to the first storage space, specifically, the memory node mapping the first logical memory to the memory node The first physical memory is configured to allocate the remote physical memory to the computing node, so that the computing node reads and writes data to the first physical memory, and the size of the first physical memory is the XB. It should be understood that, before the memory node maps the first logical memory to the first physical memory of the memory node, the method of the embodiment of the present invention further includes: the memory node determining the memory node Whether the free physical memory is greater than or equal to the remote memory, and if the free physical memory is greater than or equal to the remote memory, the memory node allocates the calculated node from the free physical memory. The first physical memory. It should be understood that, in the case that the free physical memory is smaller than the remote memory, the following steps are not continued, that is, the memory node does not allocate the remote memory to the computing node.

It is noted that, in a case that the request further indicates the first start address, the technical solution provided by the embodiment of the present invention further includes: the memory node establishing between the first start address and the second start address a mapping relationship, where the second starting address refers to an address of the first logical memory starting position in the idle logical memory. It should be noted that, when the computing node to the distal end of the memory read data A, the data will be a starting position for the A and the data A, the address of the computing node A ₁ is sent to the memory a node, the memory node determining, according to the mapping relationship, an address A ₂ in the free kernel memory of the starting position of the data A, and mapping the A ₂ to the storage space A of the memory node _3, to write the data to the A ₃ in A. The A ₃ may be physical memory or virtual memory, or part of the A ₃ is physical memory, and the other part is virtual memory.

It can be seen that, in the technical solution provided by the embodiment of the present invention, the memory node is requested according to the computing node. Allocating first logical memory to the computing node from idle logical memory of the first target process, the first target process running in the memory node; mapping the first logical memory to the first of the memory node Storage space to enable allocation of remote memory to the compute node. As a common knowledge in the art, the logical memory of a process is much larger than the physical memory of a memory node. Based on the virtual memory technology, the logical memory of a process can be mapped not only to physical memory but also to virtual memory (ie, extended from external memory). Virtual memory). Therefore, with the technical solution provided by the embodiment of the present invention, when the computing node applies for the remote memory to the memory node, the computing node can apply for a larger remote memory to the memory node without being limited by the upper limit of the free physical memory in the memory node.

Embodiment 3

FIG. 2a is a schematic structural diagram of a remote memory allocation device 200 according to an embodiment of the present invention. Preferably, the remote memory allocation device 200 according to the embodiment of the present invention is integrated into the memory node shown in FIG. 1a. . It should be noted that the remote memory allocation device 200 of the embodiment of the present invention is an execution body of the method described in the first embodiment, and can be used to perform the method described in the first embodiment. Specifically, the remote memory allocation device 200 includes:

The receiving unit 201 is configured to receive a request sent by the computing node, where the request instructs the computing node to apply for a remote memory of the XB to the memory node, where the X is greater than zero and less than 2 ⁿ , and the n is the memory node CPU The data width of the general purpose register;

As an embodiment of the present invention, the request further indicates that the remote memory is a remote physical memory; or the request further indicates that the remote memory is a remote virtual memory; or the request further indicates The remote memory is the remote physical memory and the remote virtual memory.

As another embodiment of the present invention, the request further indicates a first start address, where the first start address refers to an address of the remote memory start location facing the computing node.

The allocating unit 203 is configured to allocate, by the free kernel memory of the memory node, the first kernel memory of the XB to the computing node;

It should be noted that the technical solution described in the embodiment of the present invention is directed to the case where the CPU of the memory node operates in a kernel mode.

a mapping unit 205, configured to map the first kernel memory to the first storage space of the memory node And allocating the remote memory to the computing node, so that the computing node reads and writes data to the first storage space; wherein the size of the first storage space is the XB, The first storage space is physical memory, or the first storage space is virtual memory, or the first storage space is physical memory and virtual memory.

It should be noted that, in the case that the request further indicates that the remote memory is the remote physical memory, the mapping unit 205 is specifically configured to map the first kernel memory to the first physical memory of the memory node. The remote physical memory is allocated to the computing node, so that the computing node reads and writes data to the first physical memory, and the size of the first physical memory is the XB.

In the case where the request further indicates the first start address, referring to the remote memory allocation device 210 described in FIG. 2b, the remote memory allocation device 210 further includes an establishing unit 212, wherein the establishing unit 212 is configured to establish a mapping relationship between the first start address and the second start address, where the second start address refers to an address of the first kernel memory start position in the free core memory.

It should be noted that the content described in this embodiment and the first embodiment can be referred to each other, and the duplicated content is not described again.

It can be seen that, in the remote memory allocation device provided by the embodiment of the present invention, the memory node allocates the first kernel memory from the free kernel memory of the memory node according to the request of the computing node, and then maps the first kernel memory to The first storage space is configured to allocate remote memory for the compute node. As a common knowledge in the field, in the memory node, the kernel memory is the memory allocated by the operating system for the kernel object, the logical memory, the kernel memory is much larger than the physical memory of the memory node, and based on the virtual memory technology, the kernel memory can not only map. To physical memory, you can also map to virtual memory (that is, virtual memory that is extended from external memory). Therefore, the remote memory allocation device provided by the embodiment of the present invention, when the computing node applies for the remote memory to the memory node, is not limited by the upper limit of the free physical memory in the memory node, and the computing node can apply for a greater distance to the memory node. Side memory.

Embodiment 4

2c is a schematic structural diagram of a remote memory allocation device 220 according to an embodiment of the present invention. Preferably, the remote memory allocation device 220 according to the embodiment of the present invention is integrated into the memory node shown in FIG. 1a. . It should be noted that the remote memory allocation device 220 of the embodiment of the present invention is an execution body of the method described in the second embodiment, and can be used to perform the method described in the second embodiment. Specifically, the remote memory allocation device 220 includes:

The receiving unit 221 is configured to receive a request sent by the computing node, where the request instructs the computing node to apply to the memory node for the remote memory of the XB, where the X is greater than zero and less than 2 ⁿ , and the n is the memory a data width of a node CPU general purpose register, wherein the memory node and the computing node are connected by a network communication;

The allocating unit 223 is configured to allocate, according to the request, the first logical memory of the XB from the idle logical memory of the first target process, where the first target process refers to running in the memory node. a process having a first interface that interacts with the computing node and capable of allocating logical memory to the computing node through the first interface, the idle logical memory of the first target process is greater than or equal to the XB;

It should be noted that the first target process described in the embodiment of the present invention is selected from multiple processes running on the memory node, or the first target process is created by the memory node.

Specifically, referring to the remote memory allocation device 230 shown in FIG. 2d (1), the remote memory allocation device 230 further includes a selecting unit 232, after the receiving unit 231 receives the request sent by the computing node, and the allocating unit 233 according to the Requesting, before the first logical memory of the first target process allocates the first logical memory of the XB to the computing node, the selecting unit 232 is configured to select the first target from all processes running in the memory node process.

Alternatively, referring to the remote memory allocation device 240 shown in FIG. 2d (2), the remote memory allocation device 240 further includes a creating unit 242, after the receiving unit 241 receives the request sent by the computing node, and the allocating unit 243 according to the request Before the first logical memory of the first target process allocates the first logical memory of the XB to the computing node, the creating unit 242 is configured to create the first target process and allocate logical memory for the first target process .

It should be noted that, in a case where the number of kernels of the memory node is greater than the number of processes running the memory node before the first target process is created, the creating unit 242 creates the first target process and runs in the The number of processes on the memory node does not exceed the number of cores of the memory node. This is to ensure the speed of the process, it should be known that if the process running on the memory node The number of threads exceeds the number of cores in the memory node, and the process runs slower.

The mapping unit 225 is configured to map the first logical memory to the first storage space of the memory node, so as to implement allocating the remote memory to the computing node, so that the computing node is to the first The first storage space is the physical memory and the virtual memory, or the first storage space is physical memory, or the first The storage space is virtual memory.

It should be noted that, in a case that the request further indicates that the remote memory is the remote physical memory, the mapping unit 225 is specifically configured to map the first logical memory to the first physical memory of the memory node, The remote physical memory is allocated to the computing node, so that the computing node reads and writes data to the first physical memory, and the size of the first physical memory is the XB.

It should be noted that, referring to the remote memory allocation device 250 shown in FIG. 2e, the remote memory allocation device 250 further includes an establishing unit 252. In the case where the request further indicates the first starting address, the establishing unit 252 is configured to use Establishing a mapping relationship between the first start address and a second start address, where the second start address refers to an address of the first logical memory start position in the idle logical memory.

It should be noted that the content described in this embodiment and the second embodiment can be referred to each other, and the duplicated content is not described again.

It can be seen that, in the remote memory allocation device provided by the embodiment of the present invention, the memory node allocates the first logical memory to the computing node from the idle logical memory of the first target process according to the request of the computing node, where the first target process Running in the memory node; mapping the first logical memory to the first storage space of the memory node to implement allocating remote memory to the computing node. As a common knowledge in the art, the logical memory of a process is much larger than the physical memory of a memory node. Based on the virtual memory technology, the logical memory of a process can be mapped not only to physical memory but also to virtual memory (ie, extended from external memory). Virtual memory). Therefore, the remote memory allocation device provided by the embodiment of the present invention, when the computing node applies for the remote memory to the memory node, is not limited by the upper limit of the free physical memory in the memory node, and the computing node can apply for a greater distance to the memory node. Side memory.

Embodiment 5

Referring to the remote memory allocation system shown in FIG. 3a, a computing node 301, a memory node 305, and a remote memory allocation device 303 as described in the third embodiment are included.

Specifically, the computing node 301 is configured to send the request received by the remote memory allocation device 303;

The memory node 305 is configured to provide the first storage space;

The computing node 301 is further configured to acquire the first storage space as the remote memory.

Preferably, the remote memory allocation device 303 is integrated in the memory node 305.

It should be noted that the function of the remote memory allocation device 303 is described in the explanation of the third embodiment, and details are not described herein again.

It can be seen that, based on the beneficial effects of the remote memory allocation device 303 described in the third embodiment, the remote memory allocation system according to the embodiment of the present invention is used, and the computing node 301 is not required to apply for the remote memory to the memory node 305. The memory node 305 can limit the upper limit of the free physical memory, and the compute node 301 can apply for a larger remote memory to the memory node 305.

Embodiment 6

Referring to the remote memory allocation system shown in FIG. 3b, a computing node 311, a memory node 315, and a remote memory allocation device 313 as described in the fourth embodiment are included.

Specifically, the computing node 311 is configured to send the request received by the remote memory allocation device 313.

The memory node 315 is configured to provide the first storage space;

The computing node 311 is further configured to acquire the first storage space as the remote memory.

Preferably, the remote memory allocation device 313 is integrated in the memory node 315.

It should be noted that the function of the remote memory allocation device 313 is described in the explanation of Embodiment 4, and details are not described herein again.

It can be seen that, based on the effect of the remote memory allocation device 313 described in the fourth embodiment, the remote memory allocation system according to the embodiment of the present invention is used. When the computing node 311 requests the remote memory from the memory node 315, the memory is not affected by the memory. The limit of the upper limit of the free physical memory in the node 315, the compute node 311 can apply for a larger remote memory to the memory node 315.

The above embodiments can be referred to each other.

The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of the embodiment.

In addition, each functional unit in the network device provided by each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

The integrated unit, if implemented in the form of a software functional unit and sold or used as a standalone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention, which is essential or contributes to the prior art, or all or part of the technical solution, may be embodied in the form of a software product stored in a storage medium. A number of instructions are included to cause a computer device (which may be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, and the like. .

Those of ordinary skill in the art will appreciate that the elements and algorithm steps of the various examples described in connection with the embodiments disclosed herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the solution. A person skilled in the art can use different methods for implementing the described functions for each particular application, but such implementation should not be considered to be beyond the scope of the present invention.

The various embodiments in the specification are described in a progressive manner, and the same or similar parts between the various embodiments may be referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiment.

The embodiments of the invention described above are not intended to limit the scope of the invention. Any modifications, equivalent substitutions and improvements made within the spirit and scope of the invention are intended to be included within the scope of the invention.

Claims

A remote memory allocation method, comprising:

The memory node receives a request sent by the computing node, the request instructing the computing node to apply for remote memory of the XB to the memory node, where the memory node and the computing node are connected by network communication, the X Greater than zero and less than 2 n , the n is the data width of the memory node CPU general purpose register;

The memory node allocates the first kernel memory of the XB to the computing node from the free kernel memory of the memory node;

After the memory node allocates the first kernel memory of the XB from the free kernel memory of the memory node for the computing node, and before the computing node reads and writes data to the remote memory, the method Also includes:

The memory node maps the first kernel memory to the first storage space of the memory node, so as to allocate the remote memory to the computing node, so that the computing node is to the first storage space Read and write data, the size of the first storage space is the XB; the first storage space is physical memory, or the first storage space is virtual memory, or the first storage space is physical memory And virtual memory.
The method of claim 1 wherein:

The request further indicates that the remote memory is a remote physical memory; or

The request further indicates that the remote memory is a remote virtual memory; or

The request further indicates that the remote memory is remote physical memory and remote virtual memory.
The method of claim 2 wherein:

In the case where the request further indicates that the remote memory is remote physical memory,

The memory node maps the first kernel memory to the first storage space of the memory node, so as to allocate the remote memory to the computing node, so that the computing node is to the first storage space Reading and writing data, the size of the first storage space is the XB, and specifically includes:

The memory node maps the first kernel memory to a first physical memory of the memory node to implement allocating the remote physical memory to the computing node, thereby causing the computing node to the first physical The memory reads and writes data, and the size of the first physical memory is the XB.
The method according to any one of claims 1 to 3, wherein the request further indicates the first a starting address, where the first starting address is an address of the remote memory starting location facing the computing node;

The method further includes:

The memory node establishes a mapping relationship between the first start address and a second start address, where the second start address refers to an address of the first kernel memory start position in the free core memory.
A remote memory allocation method, comprising:

The memory node receives a request sent by the computing node, where the request indicates that the computing node requests the remote memory of the XB from the memory node, where the X is greater than zero and less than 2 n , and the n is the memory node CPU a data width of the general purpose register, wherein the memory node and the computing node are connected by a network communication;

The memory node allocates, according to the request, the first logical memory of the XB from the idle logical memory of the first target process to the computing node, where the first target process refers to running in the memory node, a first interface that interacts with the computing node and can allocate a logical memory to the computing node through the first interface, where the idle logical memory of the first target process is greater than or equal to the XB;

After the memory node allocates the first logical memory of the XB from the idle logical memory of the first target process to the computing node, and before the computing node reads and writes data to the remote memory, The method also includes:

Mapping, by the memory node, the first logical memory to a first storage space of the memory node, to implement allocating the remote memory to the computing node, thereby causing the computing node to move to the first storage space Reading and writing data; wherein the size of the first storage space is the XB, the first storage space is physical memory and virtual memory, or the first storage space is physical memory, or the first The storage space is virtual memory.
The method of claim 5 wherein:

After the memory node receives the request sent by the computing node, and the memory node allocates the first logical memory of the XB from the idle logical memory of the first target process to the computing node according to the request, the method Also includes:

The memory node selects the first target process from among all processes running on the memory node.
The method of claim 5 wherein:

After the memory node receives the request sent by the computing node, and the memory node allocates the first logical memory of the XB from the idle logical memory of the first target process to the computing node according to the request, the method Also includes:

The memory node creates the first target process and allocates logical memory for the first target process.
A method according to any one of claims 5 to 7, wherein:

The request further indicates that the remote memory is a remote physical memory; or

The request further indicates that the remote memory is a remote virtual memory; or

The request further indicates that the remote memory is remote physical memory and remote virtual memory.
The method of claim 8 wherein:

In the case where the request further indicates that the remote memory is remote physical memory,

Mapping, by the memory node, the first logical memory to a first storage space of the memory node, to implement allocating the remote memory to the computing node, thereby causing the computing node to move to the first storage space Read and write data, including:

The memory node maps the first logical memory to a first physical memory of the memory node to implement allocating the remote physical memory to the computing node, thereby causing the computing node to the first physical The memory reads and writes data, and the size of the first physical memory is the XB.
A method according to any one of claims 5 to 9, wherein

The request further indicates a first start address, where the first start address refers to an address of the remote memory start location facing the computing node;

The method further includes:

The memory node establishes a mapping relationship between the first start address and a second start address, where the second start address refers to an address of the first logical memory start position in the idle logical memory .
A method according to any one of claims 7 to 10, characterized in that:

The number of cores of the memory node is greater than the number of processes running the memory node before the first target process is created,

After the first target process is created, the number of processes running on the memory node does not exceed the number of cores of the memory node.
A remote memory allocation device, comprising:

a receiving unit, configured to receive a request sent by the computing node, where the request instructs the computing node to apply for a remote memory of the XB to the memory node, where the X is greater than zero and less than 2 n , and the n is the CPU of the memory node The data width of the register;

An allocating unit, configured to allocate, by the free kernel memory of the memory node, the first kernel memory of the XB to the computing node;

After the allocating unit allocates the first kernel memory of the XB from the free kernel memory of the memory node to the computing node, and before the computing node reads and writes data to the remote memory,

a mapping unit, configured to map the first kernel memory to a first storage space of the memory node, to implement allocating the remote memory to the computing node, thereby causing the computing node to the first storage Spatially reading and writing data, the size of the first storage space is the XB; the first storage space is physical memory, or the first storage space is virtual memory, or the first storage space is physical Memory and virtual memory.
The device of claim 12 wherein:

The request further indicates that the remote memory is a remote physical memory; or

The request further indicates that the remote memory is a remote virtual memory; or

The request further indicates that the remote memory is remote physical memory and remote virtual memory.
The device of claim 13 wherein:

In the case where the request further indicates that the remote memory is remote physical memory,

The mapping unit is specifically configured to map the first kernel memory to the first physical memory of the memory node, so as to implement allocating the remote physical memory to the computing node, thereby causing the computing node to The first physical memory reads and writes data, and the size of the first physical memory is the XB.
The apparatus according to any one of claims 12 to 14, wherein the request further indicates a first start address, and the first start address is that the remote memory start position is facing the computing node. address;

The device also includes:

The establishing unit is configured to establish a mapping relationship between the first starting address and the second starting address, where the second starting address is an address of the first kernel memory starting position in the free core memory.
A remote memory allocation device, comprising:

a receiving unit, configured to receive a request sent by the computing node, where the request instructs the computing node to apply for a remote memory of the XB to the memory node, where the X is greater than zero and less than 2 n , and the n is the memory node a data width of a CPU general purpose register, wherein the memory node and the computing node are connected by a network communication;

An allocating unit, configured to allocate, according to the request, the first logical memory of the XB from the idle logical memory of the first target process, where the first target process refers to running in the memory node, a process having a first interface that interacts with the computing node and capable of allocating logical memory to the computing node through the first interface, the idle logical memory of the first target process is greater than or equal to the XB;

After the allocating unit allocates the first logical memory of the XB from the idle logical memory of the first target process to the computing node, and before the computing node reads and writes data to the remote memory,

a mapping unit, configured to map the first logical memory to a first storage space of the memory node, to implement allocating the remote memory to the computing node, so that the computing node is to the first storage Spatially reading and writing data, the size of the first storage space is the XB; the first storage space is physical memory and virtual memory, or the first storage space is physical memory, or the first storage The space is virtual memory.
The device of claim 16 wherein:

After the receiving unit receives the request sent by the computing node, and the allocating unit allocates the first logical memory of the XB from the idle logical memory of the first target process to the computing node according to the request, The device also includes:

a selection unit for selecting the first target process from among all processes running on the memory node.
The device of claim 16 wherein:

After the receiving unit receives the request sent by the computing node, and the allocating unit allocates the first logical memory of the XB from the idle logical memory of the first target process to the computing node according to the request, The device also includes:

Creating a unit for creating the first target process and allocating logical memory for the first target process.
Apparatus according to any one of claims 16 to 18, wherein:

The request further indicates that the remote memory is a remote physical memory; or

The request further indicates that the remote memory is a remote virtual memory; or

The request further indicates that the remote memory is remote physical memory and remote virtual memory.
The device of claim 19 wherein:

In the case where the request further indicates that the remote memory is remote physical memory,

The mapping unit is specifically configured to map the first logical memory to the first physical memory of the memory node, so as to implement allocating the remote physical memory to the computing node, thereby causing the computing node to The first physical memory reads and writes data, and the size of the first physical memory is the XB.
Apparatus according to any one of claims 16 to 20, wherein

The request further indicates a first start address, where the first start address refers to an address of the remote memory start location facing the computing node;

The device also includes:

a establishing unit, configured to establish a mapping relationship between the first starting address and a second starting address, where the second starting address refers to a starting position of the first logical memory in the idle logical memory address.
Apparatus according to any one of claims 18 to 21, wherein:

The number of cores of the memory node is greater than the number of processes running the memory node before the first target process is created,

After the first target process created by the creation unit, the number of processes running on the memory node does not exceed the number of cores of the memory node.
A remote memory allocation system, comprising: a computing node, a memory node, and the remote memory allocation device according to any one of claims 12 to 15:

The computing node is configured to send the request received by the remote memory allocation device;

The memory node is configured to provide the first storage space;

The computing node is further configured to acquire the first storage space as the remote memory.
The system of claim 23 wherein:

The remote memory allocation device is integrated in the memory node.
A remote memory allocation system, comprising: a computing node, a memory node, and the remote memory allocation device of any one of claims 16 to 22:

The computing node is configured to send the request received by the remote memory allocation device;

The memory node is configured to provide the first storage space;

The computing node is further configured to acquire the first storage space as the remote memory.
The system of claim 25 wherein:

The remote memory allocation device is integrated in the memory node.