WO2022170769A1 - 通信方法、装置及系统 - Google Patents

通信方法、装置及系统 Download PDF

Info

Publication number
WO2022170769A1
WO2022170769A1 PCT/CN2021/120844 CN2021120844W WO2022170769A1 WO 2022170769 A1 WO2022170769 A1 WO 2022170769A1 CN 2021120844 W CN2021120844 W CN 2021120844W WO 2022170769 A1 WO2022170769 A1 WO 2022170769A1
Authority
WO
WIPO (PCT)
Prior art keywords
memory
host
network device
memory pool
pool
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2021/120844
Other languages
English (en)
French (fr)
Inventor
袁辉
夏寅贲
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Huawei Technologies Co Ltd
Original Assignee
Huawei Technologies Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Huawei Technologies Co Ltd filed Critical Huawei Technologies Co Ltd
Priority to EP21925420.8A priority Critical patent/EP4276643B1/en
Publication of WO2022170769A1 publication Critical patent/WO2022170769A1/zh
Priority to US18/447,046 priority patent/US12380018B2/en
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F12/00—Accessing, addressing or allocating within memory systems or architectures
    • G06F12/02—Addressing or allocation; Relocation
    • G06F12/0223—User address space allocation, e.g. contiguous or non contiguous base addressing
    • G06F12/023—Free address space management
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F12/00—Accessing, addressing or allocating within memory systems or architectures
    • G06F12/02—Addressing or allocation; Relocation
    • G06F12/0223—User address space allocation, e.g. contiguous or non contiguous base addressing
    • G06F12/0284—Multiple user address space allocation, e.g. using different base addresses
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F13/00—Interconnection of, or transfer of information or other signals between, memories, input/output devices or central processing units
    • G06F13/14—Handling requests for interconnection or transfer
    • G06F13/20—Handling requests for interconnection or transfer for access to input/output bus
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F12/00—Accessing, addressing or allocating within memory systems or architectures
    • G06F12/02—Addressing or allocation; Relocation
    • G06F12/08—Addressing or allocation; Relocation in hierarchically structured memory systems, e.g. virtual memory systems
    • G06F12/0802—Addressing of a memory level in which the access to the desired data or data block requires associative addressing means, e.g. caches
    • G06F12/0806—Multiuser, multiprocessor or multiprocessing cache systems
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F12/00—Accessing, addressing or allocating within memory systems or architectures
    • G06F12/02—Addressing or allocation; Relocation
    • G06F12/08—Addressing or allocation; Relocation in hierarchically structured memory systems, e.g. virtual memory systems
    • G06F12/0802—Addressing of a memory level in which the access to the desired data or data block requires associative addressing means, e.g. caches
    • G06F12/0877—Cache access modes
    • G06F12/0882—Page mode
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F13/00—Interconnection of, or transfer of information or other signals between, memories, input/output devices or central processing units
    • G06F13/14—Handling requests for interconnection or transfer
    • G06F13/16—Handling requests for interconnection or transfer for access to memory bus
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F13/00—Interconnection of, or transfer of information or other signals between, memories, input/output devices or central processing units
    • G06F13/38—Information transfer, e.g. on bus
    • G06F13/42—Bus transfer protocol, e.g. handshake; Synchronisation
    • G06F13/4282—Bus transfer protocol, e.g. handshake; Synchronisation on a serial bus, e.g. I2C bus, SPI bus
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F15/00—Digital computers in general; Data processing equipment in general
    • G06F15/16—Combinations of two or more digital computers each having at least an arithmetic unit, a program unit and a register, e.g. for a simultaneous processing of several programs
    • G06F15/163—Interprocessor communication
    • G06F15/173—Interprocessor communication using an interconnection network, e.g. matrix, shuffle, pyramid, star, snowflake
    • G06F15/17306—Intercommunication techniques
    • G06F15/17331—Distributed shared memory [DSM], e.g. remote direct memory access [RDMA]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F12/00—Accessing, addressing or allocating within memory systems or architectures
    • G06F12/02—Addressing or allocation; Relocation
    • G06F12/08—Addressing or allocation; Relocation in hierarchically structured memory systems, e.g. virtual memory systems
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F12/00—Accessing, addressing or allocating within memory systems or architectures
    • G06F12/02—Addressing or allocation; Relocation
    • G06F12/08—Addressing or allocation; Relocation in hierarchically structured memory systems, e.g. virtual memory systems
    • G06F12/0802—Addressing of a memory level in which the access to the desired data or data block requires associative addressing means, e.g. caches
    • G06F12/0806—Multiuser, multiprocessor or multiprocessing cache systems
    • G06F12/0813—Multiuser, multiprocessor or multiprocessing cache systems with a network or matrix configuration
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F12/00—Accessing, addressing or allocating within memory systems or architectures
    • G06F12/02—Addressing or allocation; Relocation
    • G06F12/08—Addressing or allocation; Relocation in hierarchically structured memory systems, e.g. virtual memory systems
    • G06F12/10—Address translation
    • G06F12/109—Address translation for multiple virtual address spaces, e.g. segmentation
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F13/00—Interconnection of, or transfer of information or other signals between, memories, input/output devices or central processing units
    • G06F13/14—Handling requests for interconnection or transfer
    • G06F13/20—Handling requests for interconnection or transfer for access to input/output bus
    • G06F13/28—Handling requests for interconnection or transfer for access to input/output bus using burst mode transfer, e.g. direct memory access DMA, cycle steal
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2212/00—Indexing scheme relating to accessing, addressing or allocation within memory systems or architectures
    • G06F2212/10—Providing a specific technical effect
    • G06F2212/1016—Performance improvement
    • G06F2212/1024—Latency reduction
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2212/00—Indexing scheme relating to accessing, addressing or allocation within memory systems or architectures
    • G06F2212/15—Use in a specific computing environment
    • G06F2212/154—Networked environment
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2212/00—Indexing scheme relating to accessing, addressing or allocation within memory systems or architectures
    • G06F2212/25—Using a specific main memory architecture
    • G06F2212/251—Local memory within processor subsystem
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2212/00—Indexing scheme relating to accessing, addressing or allocation within memory systems or architectures
    • G06F2212/25—Using a specific main memory architecture
    • G06F2212/254—Distributed memory
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2212/00—Indexing scheme relating to accessing, addressing or allocation within memory systems or architectures
    • G06F2212/30—Providing cache or TLB in specific location of a processing system
    • G06F2212/306—In system interconnect, e.g. between two buses
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2212/00—Indexing scheme relating to accessing, addressing or allocation within memory systems or architectures
    • G06F2212/65—Details of virtual memory and virtual address translation
    • G06F2212/657—Virtual address space management
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2213/00—Indexing scheme relating to interconnection of, or transfer of information or other signals between, memories, input/output devices or central processing units
    • G06F2213/0026—PCI express

Definitions

  • the present application relates to the field of communication, and more particularly, to a communication method, apparatus and system.
  • a host eg, a server, etc.
  • resources such as computing, memory, and storage are typically included. All applications on the host need to run in memory.
  • the memory in the host is generally configured in advance, which may cause insufficient memory during the running of the application, thereby affecting the running performance of the application.
  • the present application provides a communication method, device and system, which can improve the running performance of an application in a host.
  • a communication method is provided.
  • the first host receives a memory access address sent by the network device, the memory access address points to a storage unit in the first memory pool, the network device is connected to the first host, and the network device is used to exchange and forward the services of the first host , the network device is further configured to manage the first memory pool; when the memory of the first host satisfies a preset condition, the first host accesses the storage unit according to the memory access address.
  • the existing network equipment connected to the host and capable of exchanging and forwarding services of the host usually does not include the memory management function, which means that in the present application, the memory management function needs to be deployed in the network equipment in advance.
  • the management of the first memory pool by the network device includes implementing functions such as address isolation, access control, message distribution, flow control, and access conflict handling.
  • functions such as address isolation, access control, message distribution, flow control, and access conflict handling.
  • the first host can receive a memory access address sent by a network device, wherein the memory access address points to a storage unit in the first memory pool, so that when the memory of the first host meets a preset condition, The first host can access the storage unit in the first memory pool according to the memory access address, so that the memory of the first host can be expanded, and the running performance of the application program in the first host can be improved; and the first memory pool is managed by the network device At the time, the management difficulty and cost of the first memory pool can be reduced.
  • the network device is further connected to at least one second host, the network device is configured to exchange and forward services of the at least one second host, the at least one second host
  • the host provides a second memory pool, and the first memory pool includes the second memory pool.
  • the at least one second host provides a second memory pool, in other words, the second memory pool may be a logical memory pool formed by the memory of one or more second hosts.
  • the first memory pool is managed by the network device, and the first memory pool includes the second memory pool, which means that the second memory pool in this application is also managed by the network device.
  • the first memory pool may include a second memory pool
  • the second memory pool is a logical memory pool formed by the memory of one or more second hosts, so that the memory of the first host satisfies a preset condition
  • the memory of the second host can be accessed according to the memory access address, so that the memory of the first host can be expanded, and the running performance of the application program in the first host can be improved; and the memory utilization rate of the second host can be improved;
  • the device manages the second memory pool of the second host, which can reduce the difficulty and cost of managing the second memory pool.
  • the network device includes a third memory pool, and the first memory pool includes the third memory pool.
  • the existing network device connected to the host and capable of exchanging and forwarding services of the host usually does not include a memory pool, which means that in this implementation, a memory pool needs to be deployed in the network device in advance.
  • the first memory pool may include a third memory pool
  • the third memory pool is a memory pool of a network device, so that when the memory meets a preset condition, the first host can access the memory of the network device according to the memory access address.
  • the third memory pool can expand the memory of the first host and improve the running performance of the applications in the first host; and can shorten the memory compared to accessing the memory of the second host (that is, accessing the memory of the second memory pool).
  • the network device manages the third memory pool, the difficulty and cost of managing the third memory pool can be reduced.
  • the first memory pool may only include the second memory pool, or may only include the third memory pool, or may include both the second memory pool and the third memory pool (that is, the first memory pool includes The logical memory pool of the second memory pool and the third memory pool).
  • managing the first memory pool by the network device can not only expand the memory of the first host, but also reduce the difficulty and cost of managing the memory pool.
  • the method before the first host receives the memory access address sent by the network device, the method further includes: when the memory of the first host satisfies the preset condition, The first host sends a request message to the network device, where the request message is used to request memory in the first memory pool.
  • the preset condition is any one of the following: the memory usage rate of the first host is greater than the first threshold; the remaining memory space of the first host is less than the second Threshold; the remaining memory space of the first host is less than the memory space required by the first host to process services in the future target time period; or, the memory usage policy of the first host is to preferentially use the first memory pool.
  • the above-mentioned first threshold or second threshold may be a specific value or a percentage.
  • the first threshold may be a percentage, for example, may be 80%, 90%, 98%, etc., which is not limited in this application . It should be understood that when the memory usage rate of the first host is greater than the first threshold, it means that the application program in the first host has occupied more memory space.
  • the second threshold when the preset condition is that the remaining memory space of the first host is less than the second threshold, the second threshold may be a specific value, for example, a specific value such as 0G, 5G, 8G, etc.; it may also be a percentage, for example: it may be 0%, 10%, 20%, etc., which are not limited in this application. It should be understood that when the remaining memory space of the first host is less than the second threshold, it means that less memory space remaining in the first host can be used by the application.
  • the preset condition is that the remaining memory space of the first host is less than the memory space required by the first host to process services in the future target time period.
  • the above method further includes: predicting the memory space required for the first host to process services (ie, run the application program) in the future target time period.
  • the present application can predict in advance the memory space required by the first host to process services in the future target time period, and then when the memory of the first host meets the preset conditions (that is, the remaining memory space of the first host is larger than that of the first host in the future). If the memory space required to process the business in the target time period is small), the first memory pool is accessed in advance according to the memory access address, so as to avoid the delay problem caused by the first host running out of memory and then requesting memory from the network device. The running performance of the application in the first host is improved.
  • the first host includes a network card, and the first host communicates with the network device through a remote direct memory access RDMA protocol.
  • the first host includes a bus interface, and the first host communicates with the network device through a high-speed peripheral component interconnecting a PCIe bus or a computing express link CXL.
  • the method further includes: the first host sends a notification message to the network device, the notification message The memory access address is included to cause the network device to release the storage unit.
  • the first memory pool can also be used as a shared memory pool of multiple hosts.
  • a notification message may be sent to the network device through the first host, so that the network device releases the storage unit for use by other hosts.
  • releasing a storage unit includes modifying the storage unit from a used state to an idle state.
  • a communication method receives a request message sent by the first host, where the request message is used to request memory in the first memory pool, the network device is connected to the first host, and the network device is used for exchanging and forwarding services of the first host, The network device is also used for managing the first memory pool; the network device sends a memory access address to the first host, where the memory access address points to a storage unit in the first memory pool.
  • the sending of the memory access address by the network device to the first host specifically refers to that the network device sends the memory access address to the first host according to the request message.
  • the network device is further connected to at least one second host, the network device is configured to exchange and forward services of the at least one second host, the at least one second host
  • the host provides a second memory pool, and the first memory pool includes the second memory pool.
  • the network device includes a third memory pool, and the first memory pool includes the third memory pool.
  • the first host communicates with the network device through a remote direct memory access RDMA protocol.
  • the first host communicates with the network device through a high-speed peripheral component interconnecting a PCIe bus or a computing express link CXL.
  • sending the memory access address by the network device to the first host includes: the network device determining a free storage unit in the first memory pool; the network device sending the memory access address to the first host.
  • the first host sends the memory access address corresponding to the idle storage unit, so that the first host uses the idle storage unit.
  • the method further includes: the network device records the status of each storage unit in the first memory pool, where the status includes idle or used.
  • the status of being in use means that the memory space of the storage unit is occupied
  • the status of being idle means that the memory space of the storage unit is not occupied.
  • the method further includes: the network device receives a notification message sent by the first host, where the notification message includes the memory access address; the network device receives the notification message according to the notification message Free the storage unit.
  • the first memory pool can also be used as a shared memory pool of multiple hosts.
  • the network device may receive a notification message sent by the first host, where the notification message includes a memory access address, and then release a storage unit corresponding to the memory access address according to the notification message for use by other hosts.
  • releasing a storage unit includes modifying the storage unit from a used state to an idle state.
  • a communication device configured to apply to a first host, and the communication device includes: a receiving module configured to receive a memory access address sent by a network device, where the memory access address points to a memory access address in the first memory pool.
  • a storage unit the network device is connected to the first host, the network device is used to exchange and forward services of the first host, and the network device is also used to manage the first memory pool;
  • a processing module is used to When the memory of the host satisfies the preset condition, the storage unit is accessed according to the memory access address.
  • the network device is further connected to at least one second host, the network device is configured to exchange and forward services of the at least one second host, the at least one second host
  • the host provides a second memory pool, and the first memory pool includes the second memory pool.
  • the network device includes a third memory pool, and the first memory pool includes the third memory pool.
  • the communication device further includes: a sending module, configured to send a request message to the network device when the memory of the first host satisfies the preset condition, the The request message is used to request the memory in the first memory pool.
  • the preset condition is any one of the following: the memory usage rate of the first host is greater than the first threshold; the remaining memory space of the first host is less than the second Threshold; the remaining memory space of the first host is less than the memory space required by the first host to process services in the future target time period; or, the memory usage policy of the first host is to preferentially use the first memory pool.
  • the first host includes a network card, and the first host communicates with the network device through a remote direct memory access RDMA protocol.
  • the first host includes a bus interface, and the first host communicates with the network device through a high-speed peripheral component interconnecting a PCIe bus or a computing express link CXL.
  • the sending module when the first host no longer needs to use the storage unit, the sending module is further configured to send a notification message to the network device, where the notification message includes the memory access address to cause the network device to release the storage unit.
  • a communication apparatus configured to apply to a network device, and the communication apparatus includes: a receiving module, configured to receive a request message sent by a first host, where the request message is used to request a memory pool in the first memory pool. memory, the network device is connected to the first host, the network device is used to exchange and forward the services of the first host, and the network device is also used to manage the first memory pool; a sending module is used to send the first host Send a memory access address, where the memory access address points to a storage unit in the first memory pool.
  • the network device is further connected to at least one second host, the network device is configured to exchange and forward services of the at least one second host, the at least one second host
  • the host provides a second memory pool, and the first memory pool includes the second memory pool.
  • the network device includes a third memory pool, and the first memory pool includes the third memory pool.
  • the first host communicates with the network device through a remote direct memory access RDMA protocol.
  • the first host and the network device communicate via a high-speed peripheral component interconnecting a PCIe bus or a computing express link CXL.
  • the communication device further includes: a processing module configured to determine free storage units in the first memory pool; the sending module is further configured to send a message to the first memory pool.
  • the host sends the memory access address corresponding to the idle storage unit, so that the first host uses the idle storage unit.
  • the processing module is further configured to record the status of each storage unit in the first memory pool, where the status includes idle or used.
  • the receiving module is further configured to receive a notification message sent by the first host, where the notification message includes the memory access address; the processing module is further configured to: The storage unit is released according to the notification message.
  • a communication system comprising: a communication device as in the third aspect or any possible implementation manner of the third aspect, and a communication device as in the fourth aspect or any possible implementation manner of the fourth aspect communication device.
  • a communication apparatus including a processor and a memory; the processor executes the instructions in the memory, so that the communication apparatus executes the first aspect or any possible implementation manner of the first aspect and/or, performing the communication method in the second aspect or any possible implementation manner of the second aspect.
  • a computing device comprising: at least one processor and a memory, the at least one processor is coupled to the memory for reading and executing instructions in the memory to execute the first aspect or the communication method in any possible implementation manner of the first aspect; and/or, performing the communication method in the second aspect or any possible implementation manner of the second aspect.
  • a computer program product comprising instructions that, when the computer program product runs on a computer, cause the computer to perform the communication method as in the first aspect or any possible implementation of the first aspect; and /or, perform the communication method in the second aspect or any possible implementation manner of the second aspect.
  • a computer-readable storage medium which is characterized by comprising instructions; the instructions are used to implement the communication method in the first aspect or any possible implementation manner of the first aspect; and/or , to implement the communication method in the second aspect or any possible implementation manner of the second aspect.
  • a tenth aspect provides a chip, the chip includes a processor and a data interface, the processor reads an instruction stored in a memory through the data interface, and executes the first aspect or any possibility of the first aspect and/or, performing the communication method in the second aspect or any possible implementation manner of the second aspect.
  • the chip may further include a memory, in which instructions are stored, the processor is configured to execute the instructions stored in the memory, and when the instructions are executed, the The processor is configured to execute the communication method in the first aspect or any possible implementation manner of the first aspect; and/or, execute the communication method in the second aspect or any possible implementation manner of the second aspect.
  • a chip system in an eleventh aspect, includes at least one processor for supporting functions involved in implementing the above first aspect or some implementations of the first aspect; and/or, implementing the above-mentioned first aspect
  • the functions involved in the second aspect or some implementations of the second aspect such as receiving or processing data and/or information involved in the above methods.
  • the chip system further includes a memory for storing program instructions and data, and the memory is located inside the processor or outside the processor.
  • the chip system may be composed of chips, or may include chips and other discrete devices.
  • Figure 1 is an example diagram of a traditional DCN architecture
  • FIG. 2 is an example diagram of a memory access flow of an application in a server provided by an embodiment of the present application
  • FIG. 3 is an example diagram of a communication method provided by an embodiment of the present application.
  • FIG. 4 is an exemplary diagram of the composition of a first memory pool provided by an embodiment of the present application.
  • FIG. 5 is an example diagram of a DCN architecture provided by an embodiment of the present application.
  • FIG. 6 is an example diagram of a TOR-based memory pooling architecture provided by an embodiment of the present application.
  • FIG. 7 is an example diagram of another TOR-based memory pooling architecture provided by an embodiment of the present application.
  • FIG. 8 is an example diagram of a communication apparatus 800 provided by an embodiment of the present application.
  • FIG. 9 is an example diagram of a communication apparatus 900 provided by an embodiment of the present application.
  • FIG. 10 is an example diagram of a communication system 1000 provided by an embodiment of the present application.
  • FIG. 11 is an exemplary block diagram of a hardware structure of a communication apparatus 1100 provided by an embodiment of the present application.
  • a host eg, a server, etc.
  • resources such as computing, memory, and storage are typically included. All applications running on the host need to be in memory.
  • the memory in the host is generally configured in advance, which may cause insufficient memory during the running of the application, thereby affecting the running performance of the application.
  • DCN data center network
  • FIG 1 is an example diagram of a traditional DCN architecture.
  • traditional data centers are mainly based on a server-centric architecture, in which each server has a fixed number of computations (ie, central processing unit (CPU)) ), memory (memory) and storage (storage, for example, solid state drive (SSD), hard disk drive (HDD), etc.) resources.
  • Servers in the same rack (rack) are not directly interconnected, but are interconnected through corresponding top of rack (TOR) switches.
  • a TOR switch (hereinafter referred to as TOR for short) may also be referred to as an access switch or a leaf switch.
  • the TOR communicates with the TOR through the aggregation switch. Therefore, under this architecture, servers in different racks can communicate through TOR and aggregation switches.
  • Each server in DCN can be regarded as an independent data processing unit. The following describes the memory access process of the application in the server with reference to FIG. 2 .
  • the memory access process applied in the server mainly includes the following steps:
  • the operating system in the server assigns a virtual address, and the virtual address contains page number and address offset information.
  • the memory management unit (MMU) in the CPU converts the virtual address into a physical address according to the address mapping information in the page table, so as to realize the access to the physical memory during application processing.
  • step 2) If in step 2), the allocated virtual address cannot find its corresponding physical address information in the page table (for example, the physical memory is insufficient), the system will generate a page fault (page fault), and the system will exchange ( swap) to obtain address space from server storage.
  • the system will have a simple cold and hot data replacement function, which will put a part of the cold data in the memory into storage, so that the memory can be freed for more applications. .
  • the present application proposes a communication method, so that when the memory of the host meets a preset condition, the host can access the memory in the memory pool managed and controlled by the network device. Therefore, not only the memory expansion of the host can be realized, but also the fast access to the memory can be realized, thereby improving the running performance of the application program in the host; and the management difficulty and cost of the memory pool can be reduced.
  • FIG. 3 is an example diagram of a communication method provided by an embodiment of the present application. As shown in FIG. 3, the method 300 may include S310 and S320, and each step in the method 300 will be described in detail below.
  • the network device sends the memory access address to the first host.
  • the first host receives the memory access address sent by the network device.
  • the memory access address points to a storage unit in the first memory pool
  • the network device is connected to the first host, the network device is used to exchange and forward services of the first host, and the network device is also used to manage the first memory pool.
  • the host (including the first host and the second host hereinafter) involved in the embodiments of the present application may be any computing device, such as any one of a server, a computer, a desktop computer, a virtual machine, or other user equipment, etc. item, which is not limited in this application. It should be understood that the host may communicate with the distributed storage system through a network, and an operating system and other application programs are installed in the host. For ease of description, in the following embodiments, various functions of the host will be described by taking a server as an example, referring to FIG. 5 to FIG. 7 .
  • the network device involved in the embodiments of the present application is connected to the host (may be a direct connection or an indirect connection), and can exchange and forward services of the host.
  • the network device may be any one of an access switch, an intelligent switch, an aggregation switch, or other network device forms with switching functions, which is not limited in this application.
  • TOR will be used as an example for description.
  • the network device and the host in the embodiments of the present application can perform end-to-end data interaction, and the termination of the communication protocol can be implemented at the network device (that is, the processing related to the communication protocol is performed on the received data packet, not only forwarding only).
  • an existing network device connected to a host and capable of exchanging and forwarding services of the host usually does not include a memory management function, which means that in the present application, a memory management function needs to be deployed in the network device in advance.
  • the management of the first memory pool by the network device includes implementing functions such as address isolation, access control, message distribution, flow control, and access conflict handling.
  • functions such as address isolation, access control, message distribution, flow control, and access conflict handling.
  • the first memory pool may be a T-level memory pool or a G-level memory pool, which is not limited in this application.
  • T and G are the units of memory.
  • the network device may also record the status of each storage unit in the first memory pool, where the status includes idle or used. After a storage unit is allocated to a certain host, the network device may set the state of the storage unit to use, and when the storage unit is not allocated, the network device may set the state of the storage unit to idle. For example, the network device may set a status flag for each storage unit, and different values of the status flag may indicate different states of the storage unit.
  • sending the memory access address by the network device to the first host includes: the network device determines an idle storage unit in the first memory pool; and the network device sends the memory access address corresponding to the idle storage unit to the first host, so that the first host Use the free storage unit.
  • the method 300 may further include step S330, where the memory of the first host satisfies the predetermined conditions.
  • the first host sends a request message to the network device, where the request message is used to request memory in the first memory pool.
  • the network device receives the request message sent by the first host.
  • the network device sending the memory access address to the first host actually means that the network device sends the memory access address to the first host according to the request message.
  • Step S330 can also be replaced with the following implementation process: the network device learns the memory usage in the first host through the first host or other memory monitoring devices, and then the network device sends the memory to the first host according to the memory usage in the first host. address. For example, the network device may send the memory access address to the first host when the memory usage rate of the first host is relatively high, or the remaining memory of the first host is small, or the remaining memory of the first host cannot meet later requirements, etc. It should be understood that the present application does not limit the conditions for triggering the network device to send the memory access address to the first host.
  • the above preset conditions include but are not limited to any one of the following: the memory usage rate of the first host is greater than the first threshold; the remaining memory space of the first host is less than the second threshold; the remaining memory space of the first host is less than the first threshold. Memory space required by a host to process services in a future target time period; or, the memory usage policy of the first host is to use the first memory pool preferentially.
  • first threshold and second threshold may be specific numerical values or percentages.
  • the first threshold may be a percentage, for example, may be 80%, 90%, 98%, etc., which is not limited in this application . It should be understood that when the memory usage rate of the first host is greater than the first threshold, it means that the running of the application program in the first host has occupied more memory space.
  • the second threshold when the preset condition is that the remaining memory space of the first host is less than the second threshold, the second threshold may be a specific value, for example, a specific value such as 0G, 5G, 8G, etc.; it may also be a percentage, for example: it may be 0%, 10%, 20%, etc., which are not limited in this application. It should be understood that when the remaining memory space of the first host is less than the second threshold, it means that less memory space remaining in the first host can be used for running the application program.
  • the method 300 may further include: predicting the memory space required by the first host to process services (ie, run an application) in a future target time period. It means that the present application can predict in advance the memory space required by the first host to process services in the future target time period, and then when the memory of the first host meets the preset conditions (that is, the remaining memory space of the first host is larger than that of the first host in the future).
  • the first host accesses the memory of the first memory pool in advance according to the memory access address, so as to avoid the delay caused by the first host running out of memory and then requesting memory from the network device. Therefore, the running performance of the application in the first host can be further improved.
  • the network device may send a memory access address to the first host, wherein the memory access address points to a storage unit in the first memory pool, so that when the memory of the first host meets a preset condition, the A host can access the storage unit in the first memory pool of the network device according to the memory access address, so that the memory of the first host can be expanded, and the running performance of the application program in the first host can be improved; and the network device is used as the first memory pool.
  • the management difficulty and cost of the first memory pool can be reduced.
  • the network device may also be connected with at least one second host (two are shown in FIG. 4 ), and the network device is used for exchanging and forwarding services of the at least one second host.
  • the at least one second host provides a second memory pool
  • the first memory pool includes the second memory pool.
  • the first memory pool may only include the second memory pool.
  • the at least one second host provides a second memory pool, in other words, the second memory pool may be a logical memory pool formed by the memories of one or more second hosts.
  • the memory in the second memory pool changes dynamically with the usage of the memory on each host. Therefore, when there are few applications running on the first host, the remaining memory on the first host may also belong to the second memory pool.
  • the first memory pool is managed by the network device, and the first memory pool includes the second memory pool, which means that the second memory pool in this application is also managed by the network device.
  • the memory on each host may include multiple storage units, which means a logical second memory provided by one or more second hosts
  • a pool can also include multiple storage units.
  • the plurality of storage units on each host may be composed entirely of dynamic random access memory (DRAM), or may be composed entirely of storage class memory (SCM) (for example, non-volatile memory).
  • DRAM dynamic random access memory
  • SCM storage class memory
  • Volatile memory non-volatile memory, NVM
  • phase-change memory phase-change memory
  • PCM phase-change memory
  • Intel persistent memory Japanese pass, AEP, etc.
  • HBM high bandwidth memory
  • HBM high bandwidth memory
  • DRAM dynamic random access memory
  • SCM static random access memory
  • the application does not limit the type and composition of the storage unit in each host, which means that this application is for the logical second memory composed of one or more second hosts.
  • the types and composition modes of the multiple storage units included in the pool are also not limited.
  • each host connected to the network device needs to register memory with the network device in real time or at intervals, that is, to provide the network device with information about its own available memory space.
  • Network equipment collects all information and maintains, manages and distributes it.
  • the first host (which can be any host connected to the network device) needs to obtain memory from the memory pool (that is, the preset condition is met)
  • the first host can apply to the network device, and the network device sends the memory pool information to the first host according to the acquired memory pool information.
  • the host allocates a memory address, and then the first host directly accesses the allocated memory.
  • the access path of the memory is: second host memory-(second host network card)-network device-(first host network card)-first host memory-first host CPU.
  • the first memory pool may include a second memory pool
  • the second memory pool is a logical memory pool formed by the memory of one or more second hosts, so that the memory of the first host satisfies a preset condition
  • the memory of the second host can be accessed according to the memory access address, so that the memory of the first host can be expanded, and the running performance of the application program in the first host can be improved; and the memory utilization rate of the second host can be improved;
  • the device manages the second memory pool of the second host the difficulty and cost of managing the second memory pool can be reduced.
  • the network device may include a third memory pool, and in this case, the first memory pool may include a third memory pool.
  • the first memory pool may only include the third memory pool.
  • the existing network device connected to the host and capable of exchanging and forwarding services of the host generally does not include a memory pool, which means that in this implementation, a memory pool needs to be deployed in the network device in advance.
  • the access path of the memory is: network device memory-(first host network card)-first host memory-first host CPU.
  • the third memory pool provided by the network device may include multiple storage units.
  • the plurality of storage units in the third memory pool may be composed entirely of DRAM, may be composed entirely of SCM (for example, NVM, PCM, AEP, etc.), or may be composed of a mixture of HBM, DRAM, SCM, etc.
  • the application does not limit the type and composition of the storage units in the third memory pool. It should also be understood that in the case of the above mixed composition, the network device also needs to perform hierarchical management on the third memory pool.
  • the plurality of storage units in the third memory pool may be deployed (ie, accessed) in the network device in various ways.
  • multiple storage units can be accessed through a memory interface directly provided by a chip (for example, an application-specific integrated circuit (ASIC) chip) in a network device, and the chip can perform memory on the multiple storage units.
  • ASIC application-specific integrated circuit
  • the device has a built-in field-programmable gate array (FPGA), where the FPGA manages memory and provides a memory interface, which is used to access multiple storage units; it can also be plugged into an FPGA or FPGA through a network device It performs memory management and provides a memory interface, which is used to access multiple storage units.
  • FPGA field-programmable gate array
  • this application does not limit the access method of multiple storage units.
  • the memory management can be implemented in the above manner, or can also be implemented by other newly added memory management modules or processing modules, which is not limited in this application.
  • the first memory pool may include a third memory pool
  • the third memory pool is a memory pool of a network device, so that when the memory meets a preset condition, the first host can access the memory of the network device according to the memory access address.
  • the third memory pool can expand the memory of the first host and improve the running performance of the applications in the first host; and can shorten the memory compared to accessing the memory of the second host (that is, accessing the memory of the second memory pool).
  • the network device manages the third memory pool, the difficulty and cost of managing the third memory pool can be reduced.
  • the first memory pool may include the second memory pool and the third memory pool (that is, the first memory pool is a logical memory pool including the second memory pool and the third memory pool), and The first memory pool is managed by the network device.
  • the first host when the first host accesses the memory in the first memory pool, it may first access the memory in the second memory pool, that is, it may first access the memory of the second host, or it may first access the network.
  • this application does not limit the access order of the memory.
  • the first memory pool only includes the third memory pool in the network device as an example for description.
  • the first memory pool can be used as a shared memory pool of multiple hosts, so that when the memory of any host connected to the network device reaches a preset condition, it can request access to the first memory pool. of memory.
  • the method 300 may further include: the first host sends a notification message to the network device.
  • the notification message includes the memory access address of the storage unit, so that the network device releases the storage unit.
  • the network device receives the notification message sent by the first host; the network device releases the storage unit corresponding to the memory access address according to the notification message. It should be understood that releasing a storage unit includes modifying the storage unit from a used state to an idle state.
  • a notification message may be sent to the network device through the first host, so that the network device releases the corresponding storage unit according to the notification message for used by other hosts.
  • the communication between the first host and the network device may be through a remote direct memory access (Remote Direct Memory Access, RDMA) protocol or the like.
  • the communication can be performed through the simplified RDMA protocol or the standard RDMA protocol.
  • the simplified RDMA protocol includes a communication protocol obtained after functionally deleting or optimizing the existing RDMA protocol, which is not limited in this embodiment. It should be understood that since the standard RDMA protocol is too complicated, in the following embodiments, the preferred communication mode is to simplify the RDMA protocol, as shown in FIG. 6 . It should be understood that when the first host communicates with the network device through the RDMA protocol, it needs to be implemented through a network card on the first host and a communication module on the network device.
  • the first host and the network device may also communicate via a high-speed peripheral component interconnect (Peripheral Component Interconnect Express, PCIe) bus or a Compute Express Link (Compute Express Link, CXL) bus, as shown in FIG. 7 .
  • PCIe peripheral component interconnect Express
  • CXL Compute Express Link
  • the memory of the first memory pool can be accessed directly by means of synchronous memory semantics or direct memory access (DMA), which speeds up the remote access speed.
  • DMA direct memory access
  • the communication between the first host and the network device needs to be implemented by a bus interface on the first host and a communication module on the network device.
  • an engine such as PCIe/CXL and a DMA engine need to be added to the network device.
  • the present application does not limit the application scenarios of the method 300 .
  • the method 300 may be applied to the DCN architecture shown in FIG. 5 .
  • FIG. 5 is an example diagram of a DCN architecture provided by an embodiment of the present application. As shown in Figure 5, this architecture is mainly based on the traditional DCN architecture shown in Figure 1, and the TOR switch (ie the network device) is connected to the memory pool (ie the third memory pool, and in this example, the The first memory pool only includes the third memory pool as an example for description), and a corresponding function module (memory management module) is also deployed in the TOR switch to manage the memory in the memory pool.
  • the TOR switch ie the network device
  • the memory pool ie the third memory pool, and in this example, the The first memory pool only includes the third memory pool as an example for description
  • a corresponding function module memory management module
  • a communication module (not shown) is also deployed to realize the end-to-end data interaction between the TOR switch and the server (ie, the first host), and to realize the termination of the communication protocol at the TOR switch (ie, the receiving The message performs the processing related to the communication protocol), so that the server can access the memory on the TOR switch.
  • the DCN architecture provided by the embodiments of the present application and the application processing flow based on the architecture are exemplarily introduced below with reference to FIG. 6 and FIG. 7 .
  • the memory pool of the TOR switch takes the DRAM pool as an example
  • the storage takes the SSD as an example.
  • FIG. 6 and FIG. 7 are only an example, and do not constitute a limitation to the present application.
  • FIG. 6 is an example diagram of a TOR-based memory pooling architecture provided by an embodiment of the present application.
  • TOR includes a DRAM pool, and TOR can also provide memory management functions. It should be understood that, regarding the access manner of the DRAM and the implementation modules of the memory management function, reference may be made to the above description.
  • the standard/simplified RDMA protocol stack needs to be implemented on TOR. This enables the server to access the memory on TOR through a high-speed network when the memory on the server side is insufficient.
  • the data transmission process can be divided into two types: one is: when the server wants to access the memory on the corresponding TOR, it can be directly implemented by simplifying RDMA.
  • the communication is: TOR memory - server network card - server local memory - server local CPU.
  • the simplified RDMA is a communication protocol obtained by deleting or optimizing the functions of the existing RDMA, which is not limited in this embodiment.
  • the other is: TOR and TOR can communicate through standard RDMA.
  • the server may not include a prefetch module.
  • the application processing flow is as follows:
  • the operating system in the server will allocate a virtual address, and the virtual address contains page number and address offset information.
  • the MMU in the CPU converts the virtual address into a physical address according to the address mapping information in the page table, so as to realize the access to the physical memory during application processing.
  • step 2) the system will preferentially use the local memory (that is, the memory in the server), and when the local memory is insufficient, the system will try to access the memory on the corresponding TOR.
  • the specific process is as follows:
  • the server When the server's local memory meets the preset conditions (for example, the remaining memory is insufficient), the server will request the TOR to use the TOR memory, and the memory management module on the TOR will allocate a part of the memory space to the server after receiving the request, and Sends the memory access address of the allocated memory space to the server.
  • the server When the server's local memory meets the preset conditions (for example, the remaining memory is insufficient), the server will request the TOR to use the TOR memory, and the memory management module on the TOR will allocate a part of the memory space to the server after receiving the request, and Sends the memory access address of the allocated memory space to the server.
  • the server After the server obtains the memory access address, it can access the memory on the TOR by means such as RDMA (for example, the simplified RDMA shown in FIG. 6 ).
  • RDMA for example, the simplified RDMA shown in FIG. 6 .
  • the communication between the server and the corresponding TOR is implemented by the network card on the server and the communication module on the TOR.
  • the system can access the memory on the TOR in another rack through standard RDMA or other methods, or it can exchange the memory from the SSD in the local server. Get the address space without limitation.
  • a cluster refers to a system consisting of servers and TORs.
  • the memory configuration requirements in the server can be reduced through statistical multiplexing of cluster memory, so that when the total amount of cluster memory is the same (reduce the configuration memory on the server side, deploy the reduced memory on the TOR, so that the The total amount of memory remains unchanged) to improve memory utilization and application performance; under the same application performance, the total amount of cluster memory is reduced (statistical reuse of memory on TOR), thereby saving costs.
  • each server is configured with 100G memory
  • each server in the architecture of this application, each server can be configured with 60G memory, and then each TOR can be configured with 80G memory (assuming that one TOR is connected to two servers), so that The total amount of memory in the cluster of the architecture of the present application is the same as the total amount of memory of the cluster of the traditional architecture.
  • the memory configured in the TOR can be used as the shared memory of all servers, so that when the memory in some servers and the total memory of the corresponding TOR are insufficient, the memory on other TORs can be used, so that in the cluster When the total amount of memory remains unchanged, memory utilization and application performance can be improved.
  • each server is configured with 100G of memory. Some servers require 100G to run applications, some require 60G to run, and some require 80G. In short, it is not fixed. Then, in the architecture of this application, each server can be configured with 60G memory, and then configure memory for each TOR (0-80G, also assuming a TOR connection according to the memory occupied by the application running on each server in the architecture of Figure 1). There are 2 servers), so that in the case of the same application performance, the total amount of cluster memory is reduced, thereby saving costs.
  • a prefetching module can be introduced into the server to reduce the delay generated when accessing the memory of TOR through the network in actual operation.
  • a memory access address monitoring module (not shown in the figure) may also be added to the server, which is not limited in this embodiment.
  • the prefetching module and the memory access address monitoring module may exist independently, or may exist in the kernel of the operating system in the server, which is not limited in this embodiment.
  • the application processing flow is as follows:
  • the memory access address monitoring module will analyze the application's memory access in real time, and by tracking the application's memory access, predict the memory space required for the application to run in the future target time period. After obtaining this information, if the system determines that the remaining local memory of the server will soon be unable to meet future needs, the prefetch module can trigger pre-memory access between the server and TOR to obtain the address space needed in the future in advance, thereby preventing the system from page faults in the future.
  • obtaining the address space required in the future in advance may be obtaining the memory access address sent by TOR in advance, and transferring the processed data in the local memory to TOR in advance according to the memory access address, so as to reserve the local memory for the application to run. use.
  • the server does not include a prefetch module
  • the TOR switch obtains the required address space, and the application processing stops at this time. , wait until memory space is available before continuing with the app.
  • the address space required in the future can be obtained in advance by means of prefetching, and the address space required in the future can be obtained in advance and application processing can occur at the same time, which can greatly reduce the memory access delay and The incidence of page faults, and can also improve application performance.
  • the server and the TOR may not communicate through the simplified RDMA method shown in FIG. 6 , but communicate after interconnection through a bus such as PCIe/CXL as shown in FIG. 7 .
  • a bus such as PCIe/CXL
  • the communication between the server and the TOR needs to be implemented by the bus interface on the server and the communication module on the TOR.
  • the specific communication process in this case is: TOR memory - server local memory - server local CPU.
  • the communication between the server and the TOR is realized through PCIe/CXL, and the memory on the TOR can be accessed directly by means of synchronous memory semantics or DMA, which speeds up the remote access speed. It should be understood that in this case, TOR also needs to add PCIe/CXL and other engines and DMA engines.
  • the present application further provides a communication apparatus 800 and a communication apparatus 900 .
  • FIG. 8 is an example diagram of a communication apparatus 800 provided by an embodiment of the present application.
  • the communication device 800 is applied to the above-mentioned first host.
  • the apparatus 800 includes a receiving module 810 and a processing module 820 .
  • the receiving module 810 is configured to receive the memory access address sent by the network device.
  • the memory access address points to a storage unit in the first memory pool, the network device is connected to the first host, the network device is used to exchange and forward services of the first host, and the network device is also used to manage the first memory pool.
  • the processing module 820 is configured to access the storage unit according to the memory access address when the memory of the first host satisfies the preset condition.
  • the network device may also be connected to at least one second host, the network device is used to exchange and forward services of the at least one second host, the at least one second host provides a second memory pool, and the first memory pool includes: the second memory pool.
  • the network device may include a third memory pool, and the first memory pool may include a third memory pool.
  • the communication apparatus 800 may further include: a sending module 830 .
  • the sending module 830 may be configured to send a request message to the network device when the memory of the first host satisfies a preset condition.
  • the request message is used to request memory in the first memory pool.
  • the preset condition can be any one of the following: the memory usage rate of the first host is greater than the first threshold; the remaining memory space of the first host is less than the second threshold; the remaining memory space of the first host is less than The memory space required for processing the service in the future target time period; or, the memory usage policy of the first host is to preferentially use the first memory pool.
  • the first host may include a network card, and the first host and the network device may communicate through the remote direct memory access RDMA protocol.
  • the first host may include a bus interface, and the first host and the network device may communicate via a high-speed peripheral component interconnecting a PCIe bus or a computing express link CXL.
  • the sending module 830 may be further configured to send a notification message to the network device, where the notification message includes the memory access address, so that the network device releases the storage unit.
  • FIG. 9 is an example diagram of a communication apparatus 900 provided by an embodiment of the present application.
  • the communication apparatus 900 is applied to the above-mentioned network equipment.
  • the communication device 900 includes a receiving module 910 and a sending module 920 .
  • the receiving module 910 is configured to receive the request message sent by the first host.
  • the request message is used to request memory in the first memory pool
  • the network device is connected to the first host
  • the network device is used to exchange and forward services of the first host
  • the network device is also used to manage the first memory pool.
  • the sending module 920 is configured to send a memory access address to the first host, where the memory access address points to a storage unit in the first memory pool.
  • the network device may also be connected to at least one second host, the network device is used to exchange and forward services of the at least one second host, the at least one second host provides a second memory pool, and the first memory pool includes: the second memory pool.
  • the network device may include a third memory pool, and the first memory pool may include a third memory pool.
  • the first host and the network device may communicate through the remote direct memory access RDMA protocol.
  • the first host and the network device may communicate via a high-speed peripheral component interconnecting a PCIe bus or a computing express link CXL.
  • the communication apparatus 900 may further include: a processing module 930 .
  • the processing module 930 may be configured to determine an idle storage unit in the first memory pool; the sending module 920 may also be configured to send the memory access address corresponding to the idle storage unit to the first host, so that the first host uses the idle storage unit .
  • processing module 930 may also be recorded as a memory management module, which is not limited.
  • the processing module 930 may be further configured to record the status of each storage unit in the first memory pool.
  • the status includes idle or in use.
  • the receiving module 910 may be further configured to receive a notification message sent by the first host, where the notification message includes a memory access address.
  • the processing module 930 may also be configured to release the storage unit corresponding to the memory access address according to the notification message.
  • FIG. 10 is an example diagram of a communication system 1000 provided by an embodiment of the present application. As shown in FIG. 10 , the communication system 1000 includes a communication device 800 and a communication device 900 .
  • Figure 1100 is an exemplary block diagram of a hardware structure of a communication apparatus 1100 provided by an embodiment of the present application.
  • the communication apparatus 1100 may specifically be a computer device.
  • the communication device 1100 includes a memory 1110 , a processor 1120 , a communication interface 1130 and a bus 1140 .
  • the memory 1110 , the processor 1120 , and the communication interface 1130 are connected to each other through the bus 1140 for communication.
  • the memory 1110 may be a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM).
  • the memory 1110 may store a program, and when the program stored in the memory 1110 is executed by the processor 1120, the processor 1120 is configured to execute each step of the communication method of the embodiment of the present application.
  • the processor 1120 may adopt a general-purpose CPU, a microprocessor, an ASIC, a graphics processing unit (graphics processing unit, GPU), or one or more integrated circuits, for executing related programs, so as to implement the communication method of the method embodiment of the present application.
  • the processor 1120 can also be an integrated circuit chip with signal processing capability.
  • the communication method of the present application may be implemented by an integrated logic circuit of hardware in the processor 1120 or an instruction in the form of software.
  • the above-mentioned processor 1120 may also be a general-purpose processor, a digital signal processor (DSP), an ASIC, an FPGA or other programmable logic devices, discrete gate or transistor logic devices, and discrete hardware components.
  • DSP digital signal processor
  • ASIC application specific integrated circuit
  • FPGA field-programmable gate array
  • a general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
  • the steps of the method disclosed in conjunction with the embodiments of the present application may be directly embodied as executed by a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor.
  • the software modules may be located in random access memory, flash memory, read-only memory, programmable read-only memory or electrically erasable programmable memory, registers and other storage media mature in the art.
  • the storage medium is located in the memory 1110, and the processor 1120 reads the information in the memory 1110, and combines its hardware to complete the functions required to be performed by the modules included in the apparatus of the embodiments of the present application, or to execute the communication methods of the method embodiments of the present application.
  • the communication interface 1130 implements communication between the device 1100 and other devices or a communication network using a transceiving device such as, but not limited to, a transceiver.
  • the bus 1140 may include a pathway for communicating information between the various components of the device 1100 (eg, the memory 1110, the processor 1120, the communication interface 1130).
  • An embodiment of the present application further provides a computing device, including: at least one processor and a memory, wherein the at least one processor is coupled to the memory and configured to read and execute instructions in the memory to execute the present application The communication method of the method embodiment.
  • the embodiments of the present application also provide a computer program product containing instructions, when the computer program product is run on a computer, the computer program product enables the computer to execute the communication method of the method embodiments of the present application.
  • the embodiments of the present application further provide a computer-readable storage medium, including instructions; the instructions are used to implement the communication method of the method embodiments of the present application.
  • An embodiment of the present application further provides a chip, the chip includes a processor and a data interface, the processor reads an instruction stored in a memory through the data interface, and executes the communication method of the method embodiment of the present application.
  • the chip may further include a memory, in which instructions are stored, the processor is configured to execute the instructions stored in the memory, and when the instructions are executed, the The processor is configured to execute the communication method of the method embodiment of the present application.
  • the disclosed system, apparatus and method may be implemented in other manners.
  • the apparatus embodiments described above are only illustrative.
  • the division of the units is only a logical function division. In actual implementation, there may be other division methods.
  • multiple units or components may be combined or Can be integrated into another system, or some features can be ignored, or not implemented.
  • the shown or discussed mutual coupling or direct coupling or communication connection may be through some interfaces, indirect coupling or communication connection of devices or units, and may be in electrical, mechanical or other forms.
  • the units described as separate components may or may not be physically separated, and 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 in this embodiment.
  • each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically alone, or two or more units may be integrated into one unit.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Hardware Design (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Data Exchanges In Wide-Area Networks (AREA)
  • Computer And Data Communications (AREA)

Abstract

本申请提供了一种通信方法、装置及系统，可以应用于数据中心网络DCN架构中。其中，该方法包括：网络设备向第一主机发送内存访问地址，该内存访问地址指向第一内存池中的存储单元，该网络设备与该第一主机连接，该网络设备用于交换和转发该第一主机的业务，该网络设备还用于管理该第一内存池；在该第一主机的内存满足预设条件时，该第一主机根据该内存访问地址访问该存储单元。本申请方案能够提高主机中应用程序的运行性能。

Description

通信方法、装置及系统
本申请要求于2021年02月10日提交国家知识产权局、申请号为202110184060.3、发明名称为“一种内存共享系统和方法”以及2021年06月11日提交中国专利局、申请号为202110656360.7、申请名称为“通信方法、装置及系统”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及通信领域，并且更具体地，涉及一种通信方法、装置及系统。
背景技术
在主机(例如，服务器等)中，通常包括计算、内存(memory)和存储(storage)等资源。主机中所有应用程序都需要运行在内存中。但是，主机中的内存一般都是事先配置好的，这使得在应用程序运行的过程中，可能会出现内存不足的情况，从而影响应用程序的运行性能。
在主机出现内存不足时，通常会通过访问主机中的存储(例如，硬盘)来获得内存空间，但访问存储的时延(可达毫秒级)远大于直接访问内存(100纳秒)的时延，这样会造成应用程序运行性能的大幅下降。
因此，如何提高主机中应用程序的运行性能是亟需解决的技术问题。
发明内容
本申请提供一种通信方法、装置及系统，能够提高主机中应用程序的运行性能。
第一方面，提供了一种通信方法。第一主机接收网络设备发送的内存访问地址，该内存访问地址指向第一内存池中的存储单元，该网络设备与该第一主机连接，该网络设备用于交换和转发该第一主机的业务，该网络设备还用于管理该第一内存池；在该第一主机的内存满足预设条件时，该第一主机根据该内存访问地址访问该存储单元。
应理解，现有与主机连接且能够对主机的业务进行交换和转发的网络设备中通常不包括内存管理功能，也就意味着在本申请中，需要预先在网络设备中部署内存管理功能。
应理解，网络设备对第一内存池的管理，包括实现地址隔离、访问控制、消息分发、流量控制、访问冲突处理等功能，而这些功能正是网络设备所擅长的，因而能够降低第一内存池的管理难度和成本。
在本申请提供的通信方法中，第一主机可以接收网络设备发送的内存访问地址，其中，内存访问地址指向第一内存池中的存储单元，以便在第一主机的内存满足预设条件时，第一主机能够根据内存访问地址访问第一内存池中的存储单元，从而能够扩展第一主机的内存，提高第一主机中应用程序的运行性能；并且，由网络设备做第一内存池的管理时，能够降低第一内存池的管理难度和成本。
结合第一方面，在第一方面的一种实现方式中，该网络设备还与至少一个第二主机连接，该网络设备用于交换和转发该至少一个第二主机的业务，该至少一个第二主机提供第二内存池，该第一内存池包括该第二内存池。
其中，该至少一个第二主机提供第二内存池，换句话来讲，是指：第二内存池可以是一个或多个第二主机的内存形成的逻辑上的内存池。
应理解，第一内存池由网络设备进行管理，第一内存池包括第二内存池，意味着本申请中第二内存池也由网络设备进行管理。
在本申请实施例中，第一内存池可以包括第二内存池，第二内存池是一个或多个第二主机的内存形成的逻辑上的内存池，使得第一主机在内存满足预设条件时，可以根据内存访问地址访问第二主机的内存，从而能够扩展第一主机的内存，提高第一主机中应用程序的运行性能；并且，能够提高第二主机的内存利用率；并且，由网络设备管理第二主机的第二内存池，能够降低第二内存池的管理难度和成本。
结合第一方面，在第一方面的一种实现方式中，该网络设备包括第三内存池，该第一内存池包括该第三内存池。
应理解，现有与主机连接且能够对主机的业务进行交换和转发的网络设备中通常不包括内存池，也就意味着在该实现方式中，还需要预先在网络设备中部署内存池。
在本申请实施例中，第一内存池可以包括第三内存池，第三内存池为网络设备的内存池，使得第一主机在内存满足预设条件时，可以根据内存访问地址访问网络设备的第三内存池，从而能够扩展第一主机的内存，提高第一主机中应用程序的运行性能；且相较于访问第二主机的内存(即访问第二内存池的内存)而言能够缩短内存的访问路径，从而能够降低内存的访问时延；并且，由网络设备做第三内存池的管理时，能够降低第三内存池的管理难度和成本。
可选地，实际操作中，第一内存池可以仅包括第二内存池，或者可以仅包括第三内存池，或者可以同时包括第二内存池和第三内存池(即第一内存池是包括第二内存池和第三内存池的逻辑内存池)。并且，由网络设备管理第一内存池，既可以扩展第一主机的内存，又可以降低内存池的管理难度和成本。
结合第一方面，在第一方面的一种实现方式中，在该第一主机接收网络设备发送的内存访问地址之前，该方法还包括：在该第一主机的内存满足该预设条件时，该第一主机向该网络设备发送请求消息，该请求消息用于请求该第一内存池中的内存。
结合第一方面，在第一方面的一种实现方式中，该预设条件为以下任意一种：该第一主机的内存使用率大于第一阈值；该第一主机的内存剩余空间小于第二阈值；该第一主机的内存剩余空间小于该第一主机在未来目标时间段处理业务所需的内存空间；或者，该第一主机的内存使用策略为优先使用该第一内存池。
可选地，上述第一阈值或第二阈值可以是具体数值也可以是百分比。
示例性地，预设条件为第一主机的内存使用率大于第一阈值时，该第一阈值可以为百分比，例如：可以是80％、90％、98％等，本申请对此不做限定。应理解，第一主机的内存使用率大于第一阈值时，表示第一主机中应用程序已占用较多内存空间。
示例性地，预设条件为第一主机的内存剩余空间小于第二阈值时，该第二阈值可以是具体数值，例如，0G、5G、8G等具体数值；也可以是百分比，例如：可以是0％、10％、 20％等，本申请对此不做限定。应理解，第一主机的内存剩余空间小于第二阈值时，表示第一主机中剩余较少的内存空间可以供应用程序使用。
可选地，若预设条件为该第一主机的内存剩余空间小于该第一主机在未来目标时间段处理业务所需的内存空间。上述方法还包括：预测对第一主机在未来目标时间段处理业务(即运行应用程序)所需的内存空间。
意味着，本申请可以预先预测第一主机在未来目标时间段处理业务所需的内存空间，然后在第一主机的内存满足预设条件时(即第一主机内存剩余空间比第一主机在未来目标时间段处理业务所需的内存空间小时)，就根据内存访问地址预先访问第一内存池，从而能够避免第一主机内存用完再去向网络设备请求内存所带来的时延问题，可以进一步提高第一主机中应用程序的运行性能。
结合第一方面，在第一方面的一种实现方式中，该第一主机包括网卡，该第一主机与该网络设备通过远程直接内存访问RDMA协议通信。
结合第一方面，在第一方面的一种实现方式中，该第一主机包括总线接口，该第一主机与该网络设备通过高速外围组件互联PCIe总线或计算快速链路CXL通信。
结合第一方面，在第一方面的一种实现方式中，当该第一主机不再需要使用该存储单元时，该方法还包括：该第一主机向该网络设备发送通知消息，该通知消息包括该内存访问地址，以使该网络设备释放该存储单元。
可选地，本申请中第一内存池还可以作为多个主机的共享内存池。
在本申请实施例中，在第一主机不再需要使用网络设备中的存储单元时，可以通过第一主机向网络设备发送通知消息，以使网络设备释放该存储单元，以供其他主机使用。
应理解，释放存储单元包括将存储单元从使用状态修改为空闲状态。
第二方面，提供了一种通信方法。网络设备接收第一主机发送的请求消息，该请求消息用于请求第一内存池中的内存，该网络设备与该第一主机连接，该网络设备用于交换和转发该第一主机的业务，该网络设备还用于管理该第一内存池；该网络设备向该第一主机发送内存访问地址，该内存访问地址指向该第一内存池中的存储单元。
应理解，网络设备向第一主机发送内存访问地址，具体是指网络设备根据请求消息向第一主机发送内存访问地址。
结合第二方面，在第二方面的一种实现方式中，该网络设备还与至少一个第二主机连接，该网络设备用于交换和转发该至少一个第二主机的业务，该至少一个第二主机提供第二内存池，该第一内存池包括该第二内存池。
结合第二方面，在第二方面的一种实现方式中，该网络设备包括第三内存池，该第一内存池包括该第三内存池。
结合第二方面，在第二方面的一种实现方式中，该第一主机与该网络设备通过远程直接内存访问RDMA协议通信。
结合第二方面，在第二方面的一种实现方式中，该第一主机与该网络设备通过高速外围组件互联PCIe总线或计算快速链路CXL通信。
结合第二方面，在第二方面的一种实现方式中，该网络设备向该第一主机发送内存访问地址包括：该网络设备确定该第一内存池中的空闲存储单元；该网络设备向该第一主机发送该空闲存储单元对应的内存访问地址，以使该第一主机使用该空闲存储单元。
结合第二方面，在第二方面的一种实现方式中，该方法还包括：该网络设备记录该第一内存池中的各存储单元的状态，该状态包括空闲或使用。
应理解，状态为使用是指存储单元的内存空间被占用，状态为空闲是指存储单元的内存空间没有被占用。
结合第二方面，在第二方面的一种实现方式中，该方法还包括：该网络设备接收该第一主机发送的通知消息，该通知消息包括该内存访问地址；该网络设备根据该通知消息释放该存储单元。
可选地，本申请中第一内存池还可以作为多个主机的共享内存池。
在本申请实施例中，网络设备可以接收第一主机发送的通知消息，该通知消息包括内存访问地址，然后根据通知消息释放内存访问地址对应的存储单元，以供其他主机使用。
应理解，释放存储单元包括将存储单元从使用状态修改为空闲状态。
第三方面，提供了一种通信装置，该通信装置应用于第一主机，该通信装置包括：接收模块，用于接收网络设备发送的内存访问地址，该内存访问地址指向第一内存池中的存储单元，该网络设备与该第一主机连接，该网络设备用于交换和转发该第一主机的业务，该网络设备还用于管理该第一内存池；处理模块，用于在该第一主机的内存满足预设条件时，根据该内存访问地址访问该存储单元。
结合第三方面，在第三方面的一种实现方式中，该网络设备还与至少一个第二主机连接，该网络设备用于交换和转发该至少一个第二主机的业务，该至少一个第二主机提供第二内存池，该第一内存池包括该第二内存池。
结合第三方面，在第三方面的一种实现方式中，该网络设备包括第三内存池，该第一内存池包括该第三内存池。
结合第三方面，在第三方面的一种实现方式中，该通信装置还包括：发送模块，用于在该第一主机的内存满足该预设条件时，向该网络设备发送请求消息，该请求消息用于请求该第一内存池中的内存。
结合第三方面，在第三方面的一种实现方式中，该预设条件为以下任意一种：该第一主机的内存使用率大于第一阈值；该第一主机的内存剩余空间小于第二阈值；该第一主机的内存剩余空间小于该第一主机在未来目标时间段处理业务所需的内存空间；或者，该第一主机的内存使用策略为优先使用该第一内存池。
结合第三方面，在第三方面的一种实现方式中，该第一主机包括网卡，该第一主机与该网络设备通过远程直接内存访问RDMA协议通信。
结合第三方面，在第三方面的一种实现方式中，该第一主机包括总线接口，该第一主机与该网络设备通过高速外围组件互联PCIe总线或计算快速链路CXL通信。
结合第三方面，在第三方面的一种实现方式中，当该第一主机不再需要使用该存储单元时，该发送模块还用于，向该网络设备发送通知消息，该通知消息包括该内存访问地址，以使该网络设备释放该存储单元。
第四方面，提供了一种通信装置，该通信装置应用于网络设备，该通信装置包括：接收模块，用于接收第一主机发送的请求消息，该请求消息用于请求第一内存池中的内存，该网络设备与该第一主机连接，该网络设备用于交换和转发该第一主机的业务，该网络设备还用于管理该第一内存池；发送模块，用于向该第一主机发送内存访问地址，该内存访 问地址指向该第一内存池中的存储单元。
结合第四方面，在第四方面的一种实现方式中，该网络设备还与至少一个第二主机连接，该网络设备用于交换和转发该至少一个第二主机的业务，该至少一个第二主机提供第二内存池，该第一内存池包括该第二内存池。
结合第四方面，在第四方面的一种实现方式中，该网络设备包括第三内存池，该第一内存池包括该第三内存池。
结合第四方面，在第四方面的一种实现方式中，该第一主机与该网络设备通过远程直接内存访问RDMA协议通信。
结合第四方面，在第四方面的一种实现方式中，该第一主机与该网络设备通过高速外围组件互联PCIe总线或计算快速链路CXL通信。
结合第四方面，在第四方面的一种实现方式中，该通信装置还包括：处理模块，用于确定该第一内存池中的空闲存储单元；该发送模块还用于，向该第一主机发送该空闲存储单元对应的内存访问地址，以使该第一主机使用该空闲存储单元。
结合第四方面，在第四方面的一种实现方式中，该处理模块还用于，记录该第一内存池中的各存储单元的状态，该状态包括空闲或使用。
结合第四方面，在第四方面的一种实现方式中，该接收模块还用于，接收该第一主机发送的通知消息，该通知消息包括所述内存访问地址；该处理模块还用于，根据该通知消息释放该存储单元。
第五方面，提供了一种通信系统，包括：如第三方面或者第三方面的任一可能的实现方式中的通信装置以及如第四方面或者第四方面的任一可能的实现方式中的通信装置。
第六方面，提供了一种通信装置，包括处理器和存储器；所述处理器运行所述存储器中的指令，使得所述通信装置执行如第一方面或者第一方面的任一可能的实现方式中的通信方法；和/或，执行如第二方面或者第二方面的任一可能的实现方式中的通信方法。
第七方面，提供了一种计算设备，包括：至少一个处理器和存储器，所述至少一个处理器与所述存储器耦合，用于读取并执行所述存储器中的指令，以执行如第一方面或者第一方面的任一可能的实现方式中的通信方法；和/或，执行如第二方面或者第二方面的任一可能的实现方式中的通信方法。
第八方面，提供了一种包含指令的计算机程序产品，当该计算机程序产品在计算机上运行时，使得计算机执行如第一方面或者第一方面的任一可能的实现方式中的通信方法；和/或，执行如第二方面或者第二方面的任一可能的实现方式中的通信方法。。
第九方面，提供了一种计算机可读存储介质，其特征在于，包括指令；所述指令用于实现如第一方面或者第一方面的任一可能的实现方式中的通信方法；和/或，实现如第二方面或者第二方面的任一可能的实现方式中的通信方法。
第十方面，提供了一种芯片，所述芯片包括处理器与数据接口，所述处理器通过所述数据接口读取存储器上存储的指令，执行如第一方面或者第一方面的任一可能的实现方式中的通信方法；和/或，执行如第二方面或者第二方面的任一可能的实现方式中的通信方法。
可选地，作为一种实现方式，所述芯片还可以包括存储器，所述存储器中存储有指令，所述处理器用于执行所述存储器上存储的指令，当所述指令被执行时，所述处理器用于执 行如第一方面或者第一方面的任一可能的实现方式中的通信方法；和/或，执行如第二方面或者第二方面的任一可能的实现方式中的通信方法。
第十一方面，提供了一种芯片系统，该芯片系统包括至少一个处理器，用于支持实现上述第一方面或第一方面的某些实现中所涉及的功能；和/或，实现上述第二方面或第二方面的某些实现中所涉及的功能，例如接收或处理上述方法中所涉及的数据和/或信息。
在一种可能的设计中，所述芯片系统还包括存储器，所述存储器，用于保存程序指令和数据，存储器位于处理器之内或处理器之外。该芯片系统，可以由芯片构成，也可以包含芯片和其他分立器件。
本申请第二至第十一方面的有益效果可以参考对第一方面及其各实施方式中的有益效果的描述。
附图说明
图1是一种传统DCN架构示例图；
图2是本申请实施例提供的一种应用在服务器内的内存访问流程示例图；
图3是本申请实施例提供的一种通信方法示例图；
图4是本申请实施例提供的一种第一内存池的构成示例图；
图5是本申请实施例提供的一种DCN架构示例图；
图6是本申请实施例提供的一种基于TOR的内存池化架构示例图；
图7是本申请实施例提供的另一种基于TOR的内存池化架构示例图；
图8是本申请实施例提供的通信装置800的示例图；
图9是本申请实施例提供的通信装置900的示例图；
图10是本申请实施例提供的通信系统1000的示例图；
图11是本申请实施例提供的通信装置1100的硬件结构示例性框图。
具体实施方式
下文将先对本申请所涉及的背景技术进行详细介绍。
上述提到，在主机(例如，服务器等)中，通常包括计算、内存和存储等资源。主机中所有应用程序的运行都需要在内存中进行。但是，主机中的内存一般都是事先配置好的，这使得在应用程序运行的过程中，可能会出现内存不足的情况，从而影响应用程序的运行性能。
在主机出现内存不足时，可以通过访问主机中的存储(例如，硬盘)来获得内存空间。下面结合图1和图2，以传统数据中心网络(data center network，DCN)架构中的服务器为例对服务器出现内存不足时的内存访问流程进行介绍。
图1是一种传统DCN架构示例图。如图1所示，传统数据中心主要基于以服务器为中心(server-centric)的架构，在这种架构中，每个服务器内有固定数量的计算(即中央处理器(central processing unit，CPU))、内存(memory)和存储(storage，例如，固态硬盘(solid state drive，SSD)、机械硬盘(hard disk drive，HDD)等)资源。同一个机架(rack)内的服务器之间不直接互连，而是通过对应的架顶(top of rack，TOR)交换机实现互联。TOR交换机(以下简称为TOR)也可以称为接入交换机或者叶(leaf)交换机。 TOR与TOR之间通过汇聚交换机通信。因而，在此架构下，不同机架内的服务器之间可通过TOR和汇聚交换机通信。
在传统DCN架构中，当服务器内的内存被用完时，其无法访问其他服务器的内存。
DCN内的每个服务器都可以看作是一个独立的数据处理单元。下面结合附图2，介绍应用在服务器内的内存访问流程。
如图2所示，在传统DCN架构中，应用在服务器内的内存访问流程主要包括如下步骤：
1)对于每一个应用，服务器中的操作系统会分配虚拟地址，虚拟地址包含页号和地址偏移信息。
2)CPU内的内存管理单元(memory management unit，MMU)根据页表(page table)内的地址映射信息，将虚拟地址转换成物理地址，从而实现应用处理时对物理内存的访问。
3)如果在步骤2)中，分配的虚拟地址在页表中找不到其对应的物理地址信息(例如物理内存不足)，系统会产生缺页(page fault)，此时系统会通过交换(swap)的方式，从服务器存储处获得地址空间。
4)在物理内存被使用的过程中，为了优化访问流程，系统会自带简易的冷热数据替换功能，将内存中的一部分冷数据放到存储中，从而使内存空闲出来给更多应用使用。
可见，在服务器出现内存不足时，系统会从服务器自身的存储获得地址空间，从而保证应用程序的正常运行。但是，通常访问存储的时延(可达毫秒级)远大于直接访问内存(100纳秒)的时延，这样会造成应用程序运行性能的大幅下降。
基于此，本申请提出一种通信方法，使得在主机的内存满足预设条件时，主机可以访问网络设备管控的内存池中的内存。从而不仅能够实现主机内存扩展，还可以实现内存的快速访问，进而能够提高主机中应用程序的运行性能；且能够降低内存池的管理难度和成本。
下面结合附图对本申请方案进行详细介绍。
图3是本申请实施例提供的一种通信方法示例图。如图3所示，方法300可以包括S310和S320，下面对方法300中的各个步骤进行详细描述。
S310，网络设备向第一主机发送内存访问地址。
相应地，第一主机接收网络设备发送的内存访问地址。其中，内存访问地址指向第一内存池中的存储单元，网络设备与第一主机连接，网络设备用于交换和转发第一主机的业务，网络设备还用于管理第一内存池。
应理解，本申请实施例所涉及的主机(包括第一主机和下文中的第二主机)可以为任何计算设备，如服务器、计算机、台式计算机、虚拟机或其他用户设备等等中的任意一项，本申请对此不做限定。应理解，主机可以通过网络与分布式存储系统通信，且主机中安装有操作系统以及其他应用程序。为便于描述，在下文实施例中，将以服务器为例描述主机的各项功能，参见图5至图7部分。
应理解，本申请实施例所涉及的网络设备与主机连接(可以是直接连接，也可以是间接连接)，且能够交换和转发主机的业务。可选地，网络设备可以是接入交换机、智能交换机、汇聚交换机或其他具有交换功能的网络设备形态等中的任意一项，本申请对此不做限定。为便于描述，在下文实施例中，将以TOR为例进行描述。
应理解，本申请实施例中的网络设备和主机可以进行端到端的数据交互，且在网络设备处可以实现通信协议的终结(即对接收的数据报文执行该通信协议相关的处理，而不仅仅是转发)。
还应理解，现有与主机连接且能够对主机的业务进行交换和转发的网络设备中通常不包括内存管理功能，也就意味着在本申请中，需要预先在网络设备中部署内存管理功能。
应理解，网络设备对第一内存池的管理，包括实现地址隔离、访问控制、消息分发、流量控制、访问冲突处理等功能，而这些功能正是网络设备所擅长的，因而能够降低第一内存池的管理难度和成本。
可选地，第一内存池可以是T级内存池，也可以是G级内存池，本申请对此不做限定。其中，T和G为内存的单位。
可选地，网络设备还可以记录第一内存池中的各存储单元的状态，其中，状态包括空闲或使用。当一个存储单元被分配给某个主机后，网络设备可以将该存储单元的状态设置为使用，当该存储单元没有被分配时，网络设备可以将该存储单元的状态设置为空闲。例如，网络设备可以为每个存储单元设置一个状态标志位，该状态标志位的不同取值可以表示该存储单元的不同状态。
可选地，网络设备向第一主机发送内存访问地址包括：网络设备确定第一内存池中的空闲存储单元；网络设备向第一主机发送空闲存储单元对应的内存访问地址，以使第一主机使用该空闲存储单元。
可选地，在网络设备向第一主机发送内存访问地址(相应地，在第一主机接收网络设备发送的内存访问地址)之前，方法300还可以包括步骤S330，在第一主机的内存满足预设条件时，第一主机向网络设备发送请求消息，该请求消息用于请求第一内存池中的内存。相应地，网络设备接收第一主机发送的请求消息。这种情况下，网络设备向第一主机发送内存访问地址，实际是指网络设备根据请求消息向第一主机发送内存访问地址。
步骤S330还可以替换为以下实现过程：网络设备通过第一主机或其他内存监测设备得知第一主机内的内存使用情况，然后网络设备根据第一主机内的内存使用情况向第一主机发送内存访问地址。例如，网络设备可以在第一主机内存使用率较大、或第一主机剩余内存较少或第一主机剩余内存无法满足后期需求等的情况下，向第一主机发送内存访问地址。应理解，本申请不限定触发网络设备向第一主机发送内存访问地址的条件。
可选地，上述预设条件包括但不限于以下任意一种：第一主机的内存使用率大于第一阈值；第一主机的内存剩余空间小于第二阈值；第一主机的内存剩余空间小于第一主机在未来目标时间段处理业务所需的内存空间；或者，第一主机的内存使用策略为优先使用第一内存池。
可选地，上述第一阈值和第二阈值可以是具体数值也可以是百分比。
示例性地，预设条件为第一主机的内存使用率大于第一阈值时，该第一阈值可以为百分比，例如：可以是80％、90％、98％等，本申请对此不做限定。应理解，第一主机的内存使用率大于第一阈值时，表示第一主机中应用程序运行已占用较多内存空间。
示例性地，预设条件为第一主机的内存剩余空间小于第二阈值时，该第二阈值可以是具体数值，例如，0G、5G、8G等具体数值；也可以是百分比，例如：可以是0％、10％、20％等，本申请对此不做限定。应理解，第一主机的内存剩余空间小于第二阈值时，表示 第一主机中剩余较少的内存空间可以供应用程序的运行使用。
可选地，若预设条件为该第一主机的内存剩余空间小于该第一主机在未来目标时间段处理业务所需的内存空间。方法300还可以包括：预测第一主机在未来目标时间段处理业务(即运行应用程序)所需的内存空间。意味着，本申请可以预先预测第一主机在未来目标时间段处理业务所需的内存空间，然后在第一主机的内存满足预设条件时(即第一主机内存剩余空间比第一主机在未来目标时间段处理业务所需的内存空间小时)，第一主机就根据内存访问地址预先访问第一内存池的内存，以避免第一主机内存用完再去向网络设备请求内存所带来的时延问题，从而能够进一步提高第一主机中应用程序的运行性能。
S320，在第一主机的内存满足上述预设条件时，第一主机根据内存访问地址访问存储单元。
应理解，预设条件的介绍可参见上文，此处不再赘述。
在本申请提供的通信方法中，网络设备可以向第一主机发送内存访问地址，其中，内存访问地址指向第一内存池中的存储单元，以便在第一主机的内存满足预设条件时，第一主机能够根据该内存访问地址访问网络设备第一内存池中的存储单元，从而能够扩展第一主机的内存，提高第一主机中应用程序的运行性能；并且，由网络设备做第一内存池的管理时，能够降低第一内存池的管理难度和成本。
下面结合图4详细介绍本申请实施例所涉及的第一内存池的构成。如图4所示：
在第一种可选方式中，网络设备还可以与至少一个第二主机(图4中示出了2个)连接，网络设备用于交换和转发该至少一个第二主机的业务。其中，该至少一个第二主机提供第二内存池，该第一内存池包括该第二内存池。可选地，第一内存池可以仅包括第二内存池。
其中，该至少一个第二主机提供第二内存池，换句话来讲，第二内存池可以是一个或多个第二主机的内存形成的逻辑上的内存池。可选地，第二内存池中的内存是随着各个主机上的内存的使用情况动态变化的。因此，在第一主机上运行的应用程序较少时，第一主机上的剩余内存也可以属于该第二内存池。
应理解，第一内存池由网络设备进行管理，第一内存池包括第二内存池，意味着本申请中第二内存池也由网络设备进行管理。
应理解，在本申请中，每个主机(包括第一主机和第二主机)上的内存都可以包括多个存储单元，意味着由一个或多个第二主机提供的逻辑上的第二内存池也可以包括多个存储单元。可选地，每个主机上的多个存储单元可以全由动态随机存取存储器(dynamic random access memory，DRAM)组成，也可以全由储存级内存(storage class memory，SCM)(例如，非易失性内存(non-volatile memory，NVM)、相变化内存(phase-change memory，PCM)、因特尔持久化内存(apache pass，AEP)等)组成，还可以由高带宽内存(high bandwidth memory，HBM)、DRAM、SCM等混合组成，本申请对每个主机中存储单元的类型以及组成方式不做限定，意味着本申请对于由一个或多个第二主机构成的逻辑上的第二内存池所包括的多个存储单元的类型以及组成方式也不做限定。
应理解，在这种可选方式中，网络设备连接的各个主机需要实时或间隔向网络设备注册内存，即将自身可用内存空间信息提供给网络设备。网络设备收集所有信息，并进行维护、管理和分配。当第一主机(可以是网络设备连接的任意一个主机)需要从内存池获取 内存(即达到预设条件)时，第一主机可向网络设备申请，网络设备根据掌握的内存池信息给第一主机分配内存地址，然后第一主机直接去访问分配到的内存。
在这种情况下，内存的访问路径为：第二主机内存-(第二主机网卡)-网络设备-(第一主机网卡)-第一主机内存-第一主机CPU。
在本申请实施例中，第一内存池可以包括第二内存池，第二内存池是一个或多个第二主机的内存形成的逻辑上的内存池，使得第一主机在内存满足预设条件时，可以根据内存访问地址访问第二主机的内存，从而能够扩展第一主机的内存，提高第一主机中应用程序的运行性能；并且，能够提高第二主机的内存利用率；此外，由网络设备做第二主机的第二内存池的管理时，能够降低第二内存池的管理难度和成本。
在第二种可选方式中，网络设备可以包括第三内存池，这种情况下，第一内存池可以包括第三内存池。可选地，第一内存池可以仅包括第三内存池。
应理解，现有与主机连接且能够交换和转发主机的业务的网络设备中通常不包括内存池，也就意味着在该实现方式中，还需要预先在网络设备中部署内存池。
在这种情况下，内存的访问路径为：网络设备内存-(第一主机网卡)-第一主机内存-第一主机CPU。
应理解，网络设备所提供的第三内存池可以包括多个存储单元。可选地，第三内存池中的多个存储单元可以全部由DRAM组成，也可以全由SCM(例如，NVM、PCM、AEP等)组成，还可以由HBM、DRAM、SCM等混合组成，本申请对第三内存池中存储单元的类型以及组成方式不做限定。还应理解，在上述混合组成的情况下，网络设备还需要对第三内存池做层次化管理。
可选地，第三内存池中的多个存储单元可以通过多种方式部署(即接入)在网络设备中。例如，可以通过网络设备中的芯片(例如，特定应用集成电路(application-specific integrated circuit，ASIC)芯片)直接提供的内存接口接入多个存储单元，并由该芯片对多个存储单元进行内存管理；也可以在网络设备中引入其他新的芯片(含多个CPU核)提供交换芯片功能，并且该芯片可以支持内存管理和接口，该接口用于接入多个存储单元；还可以通过网络设备内置现场可编程逻辑门阵列(field-programmable gate array，FPGA)，其中，FPGA做内存管理，并且提供内存接口，该接口用于接入多个存储单元；还可以通过网络设备外挂FPGA，FPGA做内存管理，并且提供内存接口，该接口用于接入多个存储单元，总之，本申请不限定多个存储单元的接入方式。还应理解，内存管理可以通过上述方式实现，也可以通过其他新增的内存管理模块或处理模块来实现，本申请对此不做限定。
在本申请实施例中，第一内存池可以包括第三内存池，第三内存池为网络设备的内存池，使得第一主机在内存满足预设条件时，可以根据内存访问地址访问网络设备的第三内存池，从而能够扩展第一主机的内存，提高第一主机中应用程序的运行性能；且相较于访问第二主机的内存(即访问第二内存池的内存)而言能够缩短内存的访问路径，从而能够降低内存的访问时延；此外，由网络设备做第三内存池的管理时，能够降低第三内存池的管理难度和成本。
在第三种可选方式中，第一内存池可以包括上述第二内存池和上述第三内存池(即第一内存池是包括第二内存池和第三内存池的逻辑内存池)，并由网络设备管理第一内存池。
应理解，在该可选方式中，第一主机在访问第一内存池中的内存时，可以先访问第二内存池中的内存，即可以先访问第二主机的内存，也可以先访问网络设备中的第三内存池中的内存，本申请不限定内存的访问顺序。
另外，为便于描述，在下文实施例中(图5至图7部分)，均以第一内存池仅包括网络设备中的第三内存池为例进行描述。
可选地，本申请中第一内存池可以作为多个主机的共享内存池，以使得任意一个与网络设备相连接的主机的内存在达到预设条件时，都可以请求访问第一内存池中的内存。
可选地，当第一主机不再需要使用存储单元时，方法300还可以包括：第一主机向网络设备发送通知消息。该通知消息包括该存储单元的内存访问地址，以使网络设备释放该存储单元。相应地，网络设备接收第一主机发送的通知消息；网络设备根据通知消息释放该内存访问地址对应的存储单元。应理解，释放存储单元包括将存储单元从使用状态修改为空闲状态。
在本申请实施例中，在第一主机不再需要使用网络设备中的存储单元时，可以通过第一主机向网络设备发送通知消息，以使网络设备根据通知消息释放对应的存储单元，以供其他主机使用。
应理解，第一主机与网络设备之间可以采用高速的通信方式。
示例性地，第一主机与网络设备之间可以通过远程直接内存访问(Remote Direct Memory Access，RDMA)协议等通信。具体可以通过简化RDMA协议或标准RDMA协议通信。其中，简化RDMA协议包括对现有RDMA协议在功能上进行删减或优化之后所得到的一种通信协议，本实施例对此不做限定。应理解，由于标准的RDMA协议过于复杂，因而在下文实施例中，优选的通信方式为简化RDMA协议，可参见图6。应理解，第一主机与网络设备之间通过RDMA协议通信时，需要通过第一主机上的网卡和网络设备上的通信模块来实现。
示例性地，第一主机与网络设备还可以通过高速外围组件互联(Peripheral Component Interconnect Express，PCIe)总线或计算快速链路(Compute Express Link，CXL)等总线方式通信，可参见图7。在这种情况下，第一内存池的内存可以直接通过同步内存语义或直接内存访问(direct memory access，DMA)的方式被访问，加快了远程访问速度。应理解，第一主机和网络设备之间的通信需要由第一主机上的总线接口和网络设备上的通信模块实现。还应理解，在这种情况下，网络设备上需要增加PCIe/CXL等引擎和DMA引擎。
应理解，本申请对方法300的应用场景不做限定。可选地，作为一个示例，方法300可以应用于图5所示的DCN架构中。
图5是本申请实施例提供的一种DCN架构示例图。如图5所示，该架构主要是在图1所示传统DCN架构的基础上，在TOR交换机(即网络设备)中接入了内存池(即第三内存池，且在该示例中，以第一内存池仅包括第三内存池为例进行描述)，并且TOR交换机中也部署了相应的功能模块(内存管理模块)以管理内存池中的内存。除此之外，还部署了通信模块(未示出)以便实现TOR交换机与服务器(即第一主机)之间的端到端数据交互，并在TOR交换机处实现通信协议的终结(即对接收的报文执行该通信协议相关的处理)，从而使得服务器可以访问TOR交换机上的内存。
进一步地，下面结合图6和图7对本申请实施例所提供的DCN架构以及基于该架构 的应用处理流程进行示例性介绍。在该示例中，TOR交换机的内存池以DRAM池为例，存储以SSD为例。应理解，图6和图7仅为一种示例，不构成对本申请的限定。
图6是本申请实施例提供的一种基于TOR的内存池化架构示例图。如图6所示，TOR内包括DRAM池，并且TOR还可以提供内存管理功能。应理解，关于DRAM的接入方式和内存管理功能的实现模块可参见上文描述。
在该示例中，需要在TOR上实现标准/简化RDMA协议栈。使得当服务器侧内存不足时，服务器可以通过高速网络访问TOR上的内存。在实际操作中，如图6所示，数据传输流程可分为两种：一种是：当服务器要访问对应TOR上的内存时，可以直接通过简化RDMA的方式实现，这种情况下的通信流程为：TOR内存-服务器网卡-服务器本地内存-服务器本地CPU。其中，简化RDMA为对现有RDMA在功能上进行删减或优化之后所得到的一种通信协议，本实施例对此不做限定。另一种是：TOR与TOR之间可通过标准RDMA通信。
作为一种实现方式，在图6所示架构中，服务器可以不包括预取模块。
基于服务器不包括预取模块的架构，应用处理流程如下：
1)对于每一个应用，服务器内的操作系统会分配虚拟地址，虚拟地址包含页号和地址偏移信息。
2)CPU内的MMU根据页表内的地址映射信息，将虚拟地址转换成物理地址，从而实现应用处理时对物理内存的访问。
3)在步骤2)中，系统会优先使用本地内存(即服务器内的内存)，当本地内存不足时，系统会尝试访问对应TOR上的内存。作为一个示例，具体流程如下：
a)首先，在系统配置时，服务器内操作系统的内核模块和TOR上的内存管理模块会建立联系，使服务器能感知TOR侧内存的存在。
b)当服务器本地内存满足预设条件(例如，剩余内存不足)时，服务器将向TOR请求使用TOR的内存，TOR上的内存管理模块在接收到请求后将分配一部分内存空间给该服务器，并向服务器发送所分配内存空间的内存访问地址。
c)服务器获得内存访问地址后，可以通过RDMA等方式(例如，图6所示的简化RDMA)访问TOR上的内存。
在此步骤中，服务器和对应TOR的通信由服务器上的网卡和TOR上的通信模块实现。
4)当本地内存和TOR侧内存均不足的情况下，系统可以通过标准RDMA等方式访问另外机架(rack)内的TOR上的内存，也可以通过交换的方式，从本地服务器内的SSD处获得地址空间，不做限定。
应理解，通常在DCN架构中，每个服务器内所运行的应用不同，使得DCN架构内不同服务器对应的CPU和内存的需求不是固定的，且该比例波动范围很大，难以与固定的资源配比相匹配。那么对于图1所示的传统DCN架构而言，由于服务器无法突破服务器的界限去访问其他服务器的内存。使得当有些服务器内CPU被用完(例如有2个CPU内核，该2个CPU内核均被使用)时，剩余的内存资源将被浪费；类似地，当有些内存资源被用完时，剩余的CPU(例如有4个CPU内核，其中3个CPU内核未被使用)将被浪费，这就会导致整个DCN集群内的CPU和内存整体利用率很低，从而造成资源浪费。
而对于本申请所提供的DCN架构而言，由于在TOR中提供了内存池，在一方面，服 务器的内存能够实现动态扩展，使得服务器在本地内存不足情况下，可以使用TOR上的内存，从而使得应用程序能够保持较高性能，并且可以提高集群资源(例如CPU资源)的利用率。这里的集群是指由服务器和TOR组成的系统。另一方面，可以通过集群内存的统计复用，减少服务器内的内存的配置需求，使得在集群内存总量相同的情况(减少服务器侧配置内存，将减少的内存部署在TOR上，使得集群的内存总量不变)，提高内存利用率以及应用性能；在应用性能一样的情况下，集群内存总量降低(对TOR上内存进行统计复用)，从而节约成本。例如，在图1架构中给每个服务器配置100G内存，在本申请架构中可以给每个服务器配置60G内存，再给每个TOR内配置80G内存(假设一个TOR连接有2个服务器)，使得本申请架构集群内存总量与传统架构集群内存总量相同。由于本申请实架构中，TOR内配置的内存可以作为所有服务器的共享内存，使得有些服务器中的内存以及其所对应的TOR的内存总和不足时，可以使用其他TOR上的内存，从而使得在集群内存总量不变的情况下，能够提高内存利用率以及应用性能。再例如，在图1架构中给每个服务器配置100G内存，其中有些服务器中的应用运行需要占用100G，有些服务器中的应用运行需要占用60G，有些需要80G，总之是不固定的。那么在本申请架构中可以给每个服务器配置60G内存，再按照图1架构中每个服务器上的应用运行需占用的内存，给每个TOR去配置内存(0-80G，同样假设一个TOR连接有2个服务器)，从而使得在应用性能一样的情况下，集群内存总量降低，从而节约成本。
作为另一种实现方式，可以在服务器中引入预取模块，以降低实际操作中通过网络访问TOR的内存时所产生的时延。应理解，这种情况下，服务器中除了增加预取模块外，还可以增加内存访问地址监测模块(图中未示出)，本实施例对此不做限定。应理解，该预取模块和内存访问地址监测模块可以单独存在，也可以存在于服务器内操作系统的内核中，本实施例对此不做限定。
基于服务器包括预取模块的架构，应用处理流程如下：
在应用处理过程中，内存访问地址监测模块会实时分析应用的内存访问情况，通过追踪应用的内存访问情况，预测未来目标时间段内应用运行所需要的内存空间。得到这个信息后如果系统判定服务器本地剩余内存即将无法满足未来的需求，预取模块可以触发服务器和TOR之间的预先内存访问，提前获得未来需要的地址空间，从而避免系统在未来产生缺页。
应理解，提前获得未来需要的地址空间可以是提前获取TOR发送的内存访问地址，并且提前根据内存访问地址将本地内存中已处理的数据传输至TOR，以将本地内存预留出来，供应用运行使用。
应理解，在前一种实现方式中(服务器不包括预取模块)的应用处理过程中，内存不够了(即产生了缺页)才会去TOR交换机获得需要的地址空间，此时应用处理停止，等待获得内存空间后再继续处理该应用。而在后一种实现方式中，通过预取的方式，提前获得未来需要的地址空间，并且，提前获得未来需要的地址空间和应用处理可以是同时发生的，从而能够大幅降低内存访问时延以及缺页的发生率，并且还能够提高应用性能。
在又一种实现方式中，服务器和TOR也可以不通过图6所示简化RDMA的方式通信，而是通过图7所示的PCIe/CXL等总线的方式互联后通信。应理解，在采用标准总线互联后通信时，服务器和TOR之间的通信需要由服务器上的总线接口和TOR上的通信模块实 现。这种情况下的具体通信流程为：TOR内存-服务器本地内存-服务器本地CPU。
在本实现方式中，通过PCIe/CXL实现服务器和TOR的通信，TOR上的内存可以直接通过同步内存语义或DMA的方式被访问，加快了远程访问速度。应理解，在这种情况下，TOR还需要增加PCIe/CXL等引擎和DMA引擎。
为了实现本申请上述各实施例中的方法，本申请还提供了通信装置800和通信装置900。
图8是本申请实施例提供的通信装置800的示例图。该通信装置800应用于上述第一主机。如图8所示，该装置800包括接收模块810和处理模块820。
其中，接收模块810用于接收网络设备发送的内存访问地址。其中，内存访问地址指向第一内存池中的存储单元，网络设备与第一主机连接，网络设备用于对第一主机的业务进行交换和转发，网络设备还用于管理第一内存池。
处理模块820用于在第一主机的内存满足预设条件时，根据内存访问地址访问存储单元。
可选地，网络设备还可以与至少一个第二主机连接，网络设备用于交换和转发该至少一个第二主机的业务，该至少一个第二主机提供第二内存池，该第一内存池包括所该第二内存池。
可选地，网络设备可以包括第三内存池，第一内存池可以包括第三内存池。
可选地，通信装置800还可以包括：发送模块830。发送模块830可以用于在第一主机的内存满足预设条件时，向网络设备发送请求消息。其中，请求消息用于请求第一内存池中的内存。
可选地，预设条件可以为以下任意一种：第一主机的内存使用率大于第一阈值；第一主机的内存剩余空间小于第二阈值；第一主机的内存剩余空间小于第一主机在未来目标时间段处理业务所需的内存空间；或者，第一主机的内存使用策略为优先使用第一内存池。
可选地，第一主机可以包括网卡，第一主机与网络设备可以通过远程直接内存访问RDMA协议通信。
可选地，第一主机可以包括总线接口，第一主机与网络设备可以通过高速外围组件互联PCIe总线或计算快速链路CXL通信。
可选地，当第一主机不再需要使用该存储单元时，发送模块830还可以用于，向网络设备发送通知消息，通知消息包括内存访问地址，以使网络设备释放该存储单元。
图9是本申请实施例提供的通信装置900的示例图。该通信装置900应用于上述网络设备。如图9所示，该通信装置900包括接收模块910和发送模块920。
其中，接收模块910，用于接收第一主机发送的请求消息。其中，请求消息用于请求第一内存池中的内存，网络设备与第一主机连接，网络设备用于对第一主机的业务进行交换和转发，网络设备还用于管理第一内存池。
发送模块920，用于向第一主机发送内存访问地址，内存访问地址指向第一内存池中的存储单元。
可选地，网络设备还可以与至少一个第二主机连接，网络设备用于交换和转发该至少一个第二主机的业务，该至少一个第二主机提供第二内存池，该第一内存池包括所该第二内存池。
可选地，网络设备可以包括第三内存池，第一内存池可以包括第三内存池。
可选地，第一主机与网络设备可以通过远程直接内存访问RDMA协议通信。
可选地，第一主机与网络设备可以通过高速外围组件互联PCIe总线或计算快速链路CXL通信。
可选地，通信装置900还可以包括：处理模块930。处理模块930可以用于确定第一内存池中的空闲存储单元；发送模块920还可以用于，向第一主机发送所述空闲存储单元对应的内存访问地址，以使第一主机使用空闲存储单元。
应理解，处理模块930还可以记为内存管理模块，不做限定。
可选地，处理模块930还可以用于，记录第一内存池中的各存储单元的状态。其中，状态包括空闲或使用。
可选地，接收模块910还可以用于，接收第一主机发送的通知消息，通知消息包括内存访问地址。处理模块930还可以用于，根据通知消息释放该内存访问地址对应的存储单元。
图10是本申请实施例提供的通信系统1000的示例图。如图10所示，该通信系统1000包括通信装置800和通信装置900。
图1100是本申请实施例提供的通信装置1100的硬件结构示例性框图。可选地，该通信装置1100具体可以是一种计算机设备。该通信装置1100包括存储器1110、处理器1120、通信接口1130以及总线1140。其中，存储器1110、处理器1120、通信接口1130通过总线1140实现彼此之间的通信连接。
存储器1110可以是只读存储器(read-only memory，ROM)，静态存储设备，动态存储设备或者随机存取存储器(random access memory，RAM)。存储器1110可以存储程序，当存储器1110中存储的程序被处理器1120执行时，处理器1120用于执行本申请实施例的通信方法的各个步骤。
处理器1120可以采用通用CPU，微处理器，ASIC，图形处理器(graphics processing unit，GPU)或者一个或多个集成电路，用于执行相关程序，以实现本申请方法实施例的通信方法。
处理器1120还可以是一种集成电路芯片，具有信号处理能力。在实现过程中，本申请的通信方法可以通过处理器1120中的硬件的集成逻辑电路或者软件形式的指令完成。
上述处理器1120还可以是通用处理器、数字信号处理器(digital signal processor，DSP)、ASIC、FPGA或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件。可以实现或者执行本申请实施例中的公开的各方法、步骤及逻辑框图。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。结合本申请实施例所公开的方法的步骤可以直接体现为硬件译码处理器执行完成，或者用译码处理器中的硬件及软件模块组合执行完成。软件模块可以位于随机存储器，闪存、只读存储器，可编程只读存储器或者电可擦写可编程存储器、寄存器等本领域成熟的存储介质中。该存储介质位于存储器1110，处理器1120读取存储器1110中的信息，结合其硬件完成本申请实施例的装置中包括的模块所需执行的功能，或者执行本申请方法实施例的通信方法。
通信接口1130使用例如但不限于收发器一类的收发装置，来实现装置1100与其他设备或通信网络之间的通信。
总线1140可包括在装置1100各个部件(例如，存储器1110、处理器1120、通信接口1130)之间传送信息的通路。
本申请实施例还提供了一种计算设备，包括：至少一个处理器和存储器，所述至少一个处理器与所述存储器耦合，用于读取并执行所述存储器中的指令，以执行本申请方法实施例的通信方法。
本申请实施例还提供了一种包含指令的计算机程序产品，当该计算机程序产品在计算机上运行时，使得计算机执行本申请方法实施例的通信方法。
本申请实施例还提供了一种计算机可读存储介质，包括指令；所述指令用于实现本申请方法实施例的通信方法。
本申请实施例还提供了一种芯片，所述芯片包括处理器与数据接口，所述处理器通过所述数据接口读取存储器上存储的指令，执行本申请方法实施例的通信方法。
可选地，作为一种实现方式，所述芯片还可以包括存储器，所述存储器中存储有指令，所述处理器用于执行所述存储器上存储的指令，当所述指令被执行时，所述处理器用于执行本申请方法实施例的通信方法。
本领域普通技术人员可以意识到，结合本文中所公开的实施例描述的各示例的单元及算法步骤，能够以电子硬件、或者计算机软件和电子硬件的结合来实现。这些功能究竟以硬件还是软件方式来执行，取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用来使用不同方法来实现所描述的功能，但是这种实现不应认为超出本申请的范围。
所属领域的技术人员可以清楚地了解到，为描述的方便和简洁，上述描述的系统、装置和单元的具体工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。
在本申请所提供的几个实施例中，应该理解到，所揭露的系统、装置和方法，可以通过其它的方式实现。例如，以上所描述的装置实施例仅仅是示意性的，例如，所述单元的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，例如多个单元或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。另一点，所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口，装置或单元的间接耦合或通信连接，可以是电性，机械或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外，在本申请各个实施例中的各功能单元可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中。
以上所述，仅为本申请的具体实施方式，但本申请的保护范围并不局限于此，任何熟悉本技术领域的技术人员在本申请揭露的技术范围内，可轻易想到变化或替换，都应涵盖在本申请的保护范围之内。因此，本申请的保护范围应以所述权利要求的保护范围为准。

Claims (35)

  1. 一种通信方法，其特征在于，包括：
    第一主机接收网络设备发送的内存访问地址，所述内存访问地址指向第一内存池中的存储单元，所述网络设备与所述第一主机连接，所述网络设备用于交换和转发所述第一主机的业务，所述网络设备还用于管理所述第一内存池；
    在所述第一主机的内存满足预设条件时，所述第一主机根据所述内存访问地址访问所述存储单元。
  2. 根据权利要求1所述的通信方法，其特征在于，所述网络设备还与至少一个第二主机连接，所述网络设备用于交换和转发所述至少一个第二主机的业务，所述至少一个第二主机提供第二内存池，所述第一内存池包括所述第二内存池。
  3. 根据权利要求1或2所述的通信方法，其特征在于，所述网络设备包括第三内存池，所述第一内存池包括所述第三内存池。
  4. 根据权利要求1至3中任一项所述的通信方法，其特征在于，在所述第一主机接收网络设备发送的内存访问地址之前，所述方法还包括：
    在所述第一主机的内存满足所述预设条件时，所述第一主机向所述网络设备发送请求消息，所述请求消息用于请求所述第一内存池中的内存。
  5. 根据权利要求1至4中任一项所述的通信方法，其特征在于，所述预设条件为以下任意一种：
    所述第一主机的内存使用率大于第一阈值；
    所述第一主机的内存剩余空间小于第二阈值；
    所述第一主机的内存剩余空间小于所述第一主机在未来目标时间段处理业务所需的内存空间；或者，
    所述第一主机的内存使用策略为优先使用所述第一内存池。
  6. 根据权利要求1至5中任一项所述的通信方法，其特征在于，所述第一主机包括网卡，所述第一主机与所述网络设备通过远程直接内存访问RDMA协议通信。
  7. 根据权利要求1至5中任一项所述的通信方法，其特征在于，所述第一主机包括总线接口，所述第一主机与所述网络设备通过高速外围组件互联PCIe总线或计算快速链路CXL通信。
  8. 根据权利要求1至7中任一项所述的通信方法，其特征在于，当所述第一主机不再需要使用所述存储单元时，所述方法还包括：
    所述第一主机向所述网络设备发送通知消息，所述通知消息包括所述内存访问地址，以使所述网络设备释放所述存储单元。
  9. 一种通信方法，其特征在于，包括：
    网络设备接收第一主机发送的请求消息，所述请求消息用于请求第一内存池中的内存，所述网络设备与所述第一主机连接，所述网络设备用于交换和转发所述第一主机的业务，所述网络设备还用于管理所述第一内存池；
    所述网络设备向所述第一主机发送内存访问地址，所述内存访问地址指向所述第一内 存池中的存储单元。
  10. 根据权利要求9所述的通信方法，其特征在于，所述网络设备还与至少一个第二主机连接，所述网络设备用于交换和转发所述至少一个第二主机的业务，所述至少一个第二主机提供第二内存池，所述第一内存池包括所述第二内存池。
  11. 根据权利要求9或10所述的通信方法，其特征在于，所述网络设备包括第三内存池，所述第一内存池包括所述第三内存池。
  12. 根据权利要求9至11中任一项所述的通信方法，其特征在于，所述第一主机与所述网络设备通过远程直接内存访问RDMA协议通信。
  13. 根据权利要求9至11中任一项所述的通信方法，其特征在于，所述第一主机与所述网络设备通过高速外围组件互联PCIe总线或计算快速链路CXL通信。
  14. 根据权利要求9至13中任一项所述的通信方法，其特征在于，所述网络设备向所述第一主机发送内存访问地址包括：
    所述网络设备确定所述第一内存池中的空闲存储单元；
    所述网络设备向所述第一主机发送所述空闲存储单元对应的内存访问地址，以使所述第一主机使用所述空闲存储单元。
  15. 根据权利要求14所述的通信方法，其特征在于，所述方法还包括：
    所述网络设备记录所述第一内存池中的各存储单元的状态，所述状态包括空闲或使用。
  16. 根据权利要求14所述的通信方法，其特征在于，所述方法还包括：
    所述网络设备接收所述第一主机发送的通知消息，所述通知消息包括所述内存访问地址；
    所述网络设备根据所述通知消息释放所述内存访问地址对应的存储单元。
  17. 一种通信装置，其特征在于，所述通信装置应用于第一主机，所述通信装置包括：
    接收模块，用于接收网络设备发送的内存访问地址，所述内存访问地址指向第一内存池中的存储单元，所述网络设备与所述第一主机连接，所述网络设备用于交换和转发所述第一主机的业务，所述网络设备还用于管理所述第一内存池；
    处理模块，用于在所述第一主机的内存满足预设条件时，根据所述内存访问地址访问所述存储单元。
  18. 根据权利要求17所述的通信装置，其特征在于，所述网络设备还与至少一个第二主机连接，所述网络设备用于交换和转发所述至少一个第二主机的业务，所述至少一个第二主机提供第二内存池，所述第一内存池包括所述第二内存池。
  19. 根据权利要求17或18所述的通信装置，其特征在于，所述网络设备包括第三内存池，所述第一内存池包括所述第三内存池。
  20. 根据权利要求17至19中任一项所述的通信装置，其特征在于，所述通信装置还包括：
    发送模块，用于在所述第一主机的内存满足所述预设条件时，向所述网络设备发送请求消息，所述请求消息用于请求所述第一内存池中的内存。
  21. 根据权利要求17至20中任一项所述的通信装置，其特征在于，所述预设条件为以下任意一种：
    所述第一主机的内存使用率大于第一阈值；
    所述第一主机的内存剩余空间小于第二阈值；
    所述第一主机的内存剩余空间小于所述第一主机在未来目标时间段处理业务所需的内存空间；或者，
    所述第一主机的内存使用策略为优先使用所述第一内存池。
  22. 根据权利要求17至21中任一项所述的通信装置，其特征在于，所述第一主机包括网卡，所述第一主机与所述网络设备通过远程直接内存访问RDMA协议通信。
  23. 根据权利要求17至21中任一项所述的通信装置，其特征在于，所述第一主机包括总线接口，所述第一主机与所述网络设备通过高速外围组件互联PCIe总线或计算快速链路CXL通信。
  24. 根据权利要求20所述的通信装置，其特征在于，当所述第一主机不再需要使用所述存储单元时，所述发送模块还用于，
    向所述网络设备发送通知消息，所述通知消息包括所述内存访问地址，以使所述网络设备释放所述存储单元。
  25. 一种通信装置，其特征在于，所述通信装置应用于网络设备，所述通信装置包括：
    接收模块，用于接收第一主机发送的请求消息，所述请求消息用于请求第一内存池中的内存，所述网络设备与所述第一主机连接，所述网络设备用于交换和转发所述第一主机的业务，所述网络设备还用于管理所述第一内存池；
    发送模块，用于向所述第一主机发送内存访问地址，所述内存访问地址指向所述第一内存池中的存储单元。
  26. 根据权利要求25所述的通信装置，其特征在于，所述网络设备还与至少一个第二主机连接，所述网络设备用于交换和转发所述至少一个第二主机的业务，所述至少一个第二主机提供第二内存池，所述第一内存池包括所述第二内存池。
  27. 根据权利要求25或26所述的通信装置，其特征在于，所述网络设备包括第三内存池，所述第一内存池包括所述第三内存池。
  28. 根据权利要求25至27中任一项所述的通信装置，其特征在于，所述第一主机与所述网络设备通过远程直接内存访问RDMA协议通信。
  29. 根据权利要求25至27中任一项所述的通信装置，其特征在于，所述第一主机与所述网络设备通过高速外围组件互联PCIe总线或计算快速链路CXL通信。
  30. 根据权利要求25至29中任一项所述的通信装置，其特征在于，所述通信装置还包括：
    处理模块，用于确定所述第一内存池中的空闲存储单元；
    所述发送模块还用于，向所述第一主机发送所述空闲存储单元对应的内存访问地址，以使所述第一主机使用所述空闲存储单元。
  31. 根据权利要求30所述的通信装置，其特征在于，所述处理模块还用于，
    记录所述第一内存池中的各存储单元的状态，所述状态包括空闲或使用。
  32. 根据权利要求30所述的通信装置，其特征在于，所述接收模块还用于，
    接收所述第一主机发送的通知消息，所述通知消息包括所述内存访问地址；
    所述处理模块还用于，根据所述通知消息释放所述内存访问地址对应的存储单元。
  33. 一种通信系统，其特征在于，包括：如权利要求17至24中任一项所述的通信装置以及如权利要求25至32中任一项所述的通信装置。
  34. 一种通信装置，其特征在于，包括处理器和存储器；所述处理器运行所述存储器中的指令，使得所述通信装置执行如权利要求1至8中任一项所述的通信方法；和/或，执行如权利要求9至16中任一项所述的通信方法。
  35. 一种计算机可读存储介质，其特征在于，包括指令；所述指令用于实现如权利要求1至8中任一项所述的通信方法；和/或，实现如权利要求9至16中任一项所述的通信方法。
PCT/CN2021/120844 2021-02-10 2021-09-27 通信方法、装置及系统 Ceased WO2022170769A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
EP21925420.8A EP4276643B1 (en) 2021-02-10 2021-09-27 Communication method, apparatus, and system
US18/447,046 US12380018B2 (en) 2021-02-10 2023-08-09 Method for host access to network device-managed memory pool

Applications Claiming Priority (4)

Application Number Priority Date Filing Date Title
CN202110184060 2021-02-10
CN202110184060.3 2021-02-10
CN202110656360.7A CN114911725A (zh) 2021-02-10 2021-06-11 通信方法、装置及系统
CN202110656360.7 2021-06-11

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US18/447,046 Continuation US12380018B2 (en) 2021-02-10 2023-08-09 Method for host access to network device-managed memory pool

Publications (1)

Publication Number Publication Date
WO2022170769A1 true WO2022170769A1 (zh) 2022-08-18

Family

ID=82761468

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/120844 Ceased WO2022170769A1 (zh) 2021-02-10 2021-09-27 通信方法、装置及系统

Country Status (4)

Country Link
US (1) US12380018B2 (zh)
EP (1) EP4276643B1 (zh)
CN (1) CN114911725A (zh)
WO (1) WO2022170769A1 (zh)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117667300A (zh) * 2022-08-31 2024-03-08 阿里巴巴（中国）有限公司 计算系统及相关方法
CN118244972A (zh) * 2022-12-24 2024-06-25 华为技术有限公司 一种数据存储方法、装置及系统
CN119496792A (zh) * 2023-08-21 2025-02-21 超聚变数字技术有限公司 数据传输方法、及计算节点
CN119402566B (zh) * 2024-12-30 2025-04-29 苏州元脑智能科技有限公司 一种内存管理系统、方法、程序产品及存储介质

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104239222A (zh) * 2013-06-20 2014-12-24 华为技术有限公司 一种内存访问方法、设备和系统
US20150026780A1 (en) * 2012-03-07 2015-01-22 Ntt Docomo, Inc. Host providing system and communication control method
US20170134225A1 (en) * 2015-11-05 2017-05-11 Accelstor, Inc. Network apparatus for temporarily accessing network setting and method using thereof
CN106776048A (zh) * 2017-01-24 2017-05-31 郑州云海信息技术有限公司 一种实时虚拟机内存调度方法及装置
CN108023914A (zh) * 2016-11-03 2018-05-11 阿里巴巴集团控股有限公司 一种内存数据共享系统、内存数据的写入以及读取方法
CN110008140A (zh) * 2019-03-11 2019-07-12 深圳市广和通无线股份有限公司 内存管理方法、装置、计算机设备和存储介质
CN112291086A (zh) * 2020-10-16 2021-01-29 苏州浪潮智能科技有限公司 一种交换机的内存扩容方法、系统及装置

Family Cites Families (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2001002965A2 (en) * 1999-06-30 2001-01-11 Broadcom Corporation Memory management unit for a network switch
US8429652B2 (en) * 2009-06-22 2013-04-23 Citrix Systems, Inc. Systems and methods for spillover in a multi-core system
US8537613B2 (en) * 2011-03-31 2013-09-17 Sandisk Technologies Inc. Multi-layer memory system
US9128843B2 (en) * 2012-10-11 2015-09-08 Industrial Technology Research Institute Method and computer system for memory management on virtual machine system
US9692820B2 (en) * 2013-04-06 2017-06-27 Citrix Systems, Inc. Systems and methods for cluster parameter limit
CN107135189B (zh) * 2016-02-26 2020-02-14 华为技术有限公司 一种报文发送方法及物理机
US20190044809A1 (en) * 2017-08-30 2019-02-07 Intel Corporation Technologies for managing a flexible host interface of a network interface controller
US11636014B2 (en) * 2017-10-31 2023-04-25 SK Hynix Inc. Memory system and data processing system including the same
US12135876B2 (en) * 2018-02-05 2024-11-05 Micron Technology, Inc. Memory systems having controllers embedded in packages of integrated circuit memory
US11792307B2 (en) * 2018-03-28 2023-10-17 Apple Inc. Methods and apparatus for single entity buffer pool management
KR102769757B1 (ko) * 2018-12-21 2025-02-20 에스케이하이닉스 주식회사 메모리 시스템 및 메모리 시스템의 동작방법
US20200322287A1 (en) * 2020-06-18 2020-10-08 Intel Corporation Switch-managed resource allocation and software execution
US11841793B2 (en) * 2021-01-27 2023-12-12 Rambus Inc. Switch-based free memory tracking in data center environments
US12511071B2 (en) * 2022-03-31 2025-12-30 Intel Corporation Advanced interleaving techniques for fabric based pooling architectures
US12498974B2 (en) * 2022-03-31 2025-12-16 Intel Corporation Adaptive collaborative memory with the assistance of programmable networking devices

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150026780A1 (en) * 2012-03-07 2015-01-22 Ntt Docomo, Inc. Host providing system and communication control method
CN104239222A (zh) * 2013-06-20 2014-12-24 华为技术有限公司 一种内存访问方法、设备和系统
US20170134225A1 (en) * 2015-11-05 2017-05-11 Accelstor, Inc. Network apparatus for temporarily accessing network setting and method using thereof
CN108023914A (zh) * 2016-11-03 2018-05-11 阿里巴巴集团控股有限公司 一种内存数据共享系统、内存数据的写入以及读取方法
CN106776048A (zh) * 2017-01-24 2017-05-31 郑州云海信息技术有限公司 一种实时虚拟机内存调度方法及装置
CN110008140A (zh) * 2019-03-11 2019-07-12 深圳市广和通无线股份有限公司 内存管理方法、装置、计算机设备和存储介质
CN112291086A (zh) * 2020-10-16 2021-01-29 苏州浪潮智能科技有限公司 一种交换机的内存扩容方法、系统及装置

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP4276643A4

Also Published As

Publication number Publication date
EP4276643A1 (en) 2023-11-15
EP4276643A4 (en) 2024-07-10
CN114911725A (zh) 2022-08-16
EP4276643B1 (en) 2025-04-23
US12380018B2 (en) 2025-08-05
US20230385190A1 (en) 2023-11-30

Similar Documents

Publication Publication Date Title
JP7752489B2 (ja) メモリリソースを管理するためのシステム及び方法
US11841814B2 (en) System with cache-coherent memory and server-linking switch
US11748278B2 (en) Multi-protocol support for transactions
US11507426B2 (en) Resource pool management method and apparatus, resource pool control unit, and communications device
EP4276643B1 (en) Communication method, apparatus, and system
EP4439312A1 (en) Data storage method and system, storage access configuration method and related device
US20150261698A1 (en) Memory system, memory module, memory module access method, and computer system
WO2022271239A1 (en) Queue scaling based, at least, in part, on processing load
AU2015402888B2 (en) Computer device and method for reading/writing data by computer device
WO2024179298A1 (zh) 跨机柜服务器内存池化方法、装置、设备、服务器及介质
CN116126742A (zh) 内存访问方法、装置、服务器及存储介质
CN118779280B (zh) 降低总线负载的方法、cxl模组、处理系统和处理器芯片
TW202020674A (zh) 數據處理系統
CN113722110B (zh) 计算机系统、内存访问方法及设备
CN118349399A (zh) 一种数据迁移方法、控制器及扩展存储箱
CN119718707B (zh) 缓存管理方法、装置及计算机可读存储介质
HK40062664A (zh) 计算机系统、内存访问方法及设备
HK40062664B (zh) 计算机系统、内存访问方法及设备
WO2026066110A1 (zh) 存储节点、存储阵列和数据访问方法
CN117499511A (zh) 报文处理方法、芯片和计算机设备

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21925420

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 202337051609

Country of ref document: IN

ENP Entry into the national phase

Ref document number: 2021925420

Country of ref document: EP

Effective date: 20230807

NENP Non-entry into the national phase

Ref country code: DE

WWG Wipo information: grant in national office

Ref document number: 2021925420

Country of ref document: EP