Instant migration method and system

By using system single chip and CXL 3.0 shared memory technology in the edge computing platform for seamless state migration, the problem of too long downtime during virtual machine migration is solved, and seamless instant migration and users that meet the 5G communication latency requirements are achieved. Improved experience.

CN119088285APending Publication Date: 2024-12-06WISTRON CORP
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202311010437.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2023-06-06
Filing Date
2023-08-11
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

The existing edge computing platform has a long downtime during the virtual machine instant migration process, which cannot meet the ultra-reliable low-latency communication requirements of the fifth-generation mobile communication technology, resulting in interruption of user equipment services.

Method used

The system single chip and CXL 3.0 shared memory technology are adopted to perform state migration by sharing memory between the source and target ends, network replication and virtual machine resource allocation are avoided, and seamless real-time migration is achieved.

Benefits of technology

It realizes seamless instant migration, meets the ultra-reliable low-latency communication requirements of the fifth generation of mobile communication technology, reduces latency and improves user experience.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119088285A_ABST
    Figure CN119088285A_ABST
Patent Text Reader

Abstract

An immediate migration method for a shared memory of a system includes receiving a state of a first system-on-chip, and storing the state for reading. The first system-on-chip is used for writing the state of the first system-on-chip into the shared memory, and a second system-on-chip is used for reading the state from the shared memory. The shared memory at least consists of at least one part of a first memory in the first system-on-chip, at least one part of a second memory in the second system-on-chip or at least one part of a third memory in the system.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to an instant migration method and a system thereof, and in particular to an instant migration method and a system thereof capable of improving user experience. Background Art

[0002] Existing edge computing platforms (such as VMWare) use virtual machine (VM) hypervisors to facilitate edge clouds. When the network performs live migration of a VM from the source to the target, the hypervisor first freezes / pauses the VM and saves the VM state on the source, then copies the VM state from the source to the target via the network, and then resumes the VM with the VM state on the target. This entire process takes a few milliseconds.

[0003] In addition, for virtual machine provisioning, the hypervisor first allocates virtual machine resources (such as a CPU, memory, or a peripheral component interconnect express (PCIe) card), then loads the virtual machine image file from the virtual machine storage device to the virtual machine, and then starts the virtual machine from the virtual machine image file. This entire process takes several minutes.

[0004] However, the end-to-end latency of ultra-reliable low-latency communication of the fifth-generation mobile communication technology is only allowed to be 1 millisecond. Any delay exceeding 1 millisecond will cause service interruption of the user equipment (UE). Summary of the invention

[0005] Therefore, the present invention mainly provides a real-time migration method and system thereof to improve the deficiencies of the prior art.

[0006] An embodiment of the present invention discloses a real-time migration method for a shared memory of a system, comprising receiving a state of a first system single chip, wherein the first system single chip is used to write the state of the first system single chip into the shared memory; and storing the state for reading, wherein a second system single chip is used to read the state from the shared memory, and the shared memory is composed of at least a portion of a first memory in the first system single chip, at least a portion of a second memory in the second system single chip, or at least a portion of a third memory in the system.

[0007] An embodiment of the present invention discloses a system, comprising a first system single chip, used to write a state of the first system single chip into a shared memory; and a second system single chip, coupled to the first system single chip, used to read the state from the shared memory, wherein the shared memory is composed of at least a portion of a first memory in the first system single chip, at least a portion of a second memory in the second system single chip, or at least a portion of a third memory in the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figures 1 to 3 They are schematic diagrams of a system according to a first embodiment of the present invention;

[0009] Figure 4 It is a schematic diagram of a system-on-chip according to an embodiment of the present invention;

[0010] Figures 5 and 6 They are schematic diagrams of a system according to a first embodiment of the present invention;

[0011] Figure 7 Schematic diagram of a live migration method according to an embodiment of the present invention.

[0012] Explanation of symbols

[0013] 10~60: System

[0014] 10MM1~10MM3: Memory

[0015] 10PU1,10PU2: Processor

[0016] 10SoC1,10SoC2:Device

[0017] 20R1~20Ri: Cabinet

[0018] 20R1SoC11~20RiSoCmn,30SoC11~30SoC1n,30SoC21~30SoC2j,40SoC:System on Chip

[0019] 20R1SVR1~20RiSVRm,30SVR1,30SVR2: Server

[0020] 30HV1,30HV2: System-on-Chip Hypervisor

[0021] 30HW1, 30HW2: Hardware infrastructure

[0022] 30OS11~30OS1n,30OS21~30OS2j: Operating System

[0023] 40Cm, 40cV: CPU core

[0024] 40thdM, 40thdV: Execution thread

[0025] 50C: Consumer

[0026] 50ctrl: arbitration controller

[0027] 50P:Producer

[0028] 10SMM, 50SMM1, 50SMM2: shared memory

[0029] 50SoCpl: Computing Pool

[0030] 70: Live Migration Method

[0031] S700~S706: Steps DETAILED DESCRIPTION

[0032] Figure 1 Schematic diagram of a system 10 according to an embodiment of the present invention. The system 10 may include devices 10SoC1, 10SoC2 and a memory 10MM3 coupled to each other. The devices 10SoC1 and 10SoC2 may include processors 10PU1, 10PU2 and memories 10MM1 and 10MM2, respectively. In one embodiment, the devices 10SoC1, 10SoC2 and the memory 10MM3 may be respectively set in different (or the same) servers, chassis or racks, but are not limited thereto. In one embodiment, the device 10SoC1 and the memory 10MM3 may be separately set in the same server.

[0033] In one embodiment, devices 10SoC1 and 10SoC2 can be implemented using system on chip (SoC) respectively. In the present invention, live migration refers to the process of transferring the running system on chip (operation or state) between different physical devices without disconnecting the user end or application. The memory, storage device or network connection (content or state) of the system on chip is transferred from the source end (such as device 10SoC1) to the target end (such as device 10SoC2). The time between suspending a system on chip (operation) at the source end and resuming the system on chip (operation) at the target end is called downtime. When the downtime of the system on chip in the live migration process is short enough that the user cannot perceive that the service has been interrupted, it is the seamless live migration that the present invention aims to achieve (for example, less than or much less than 1 millisecond).

[0034] In one embodiment, at least one of the memories 10MM1-10MM3, part of the memory units of the memories 10MM1-10MM3 (e.g., part of the memory 10MM1), or a combination thereof may be used as a shared memory 10SMM of the devices 10SoC1 and 10SoC2. In one embodiment, using Compute Express Link (CXL) 3.0, the state of the SoC may be saved in a CXL shared memory (between all CXL nodes (e.g., CPUs or devices) in different servers or racks). In one embodiment, after a SoC at the source end (e.g., device 10SoC1) stores its state in the shared memory 10SMM (e.g., CXL shared memory), another SoC at the target end (e.g., device 10SoC2) may immediately restore the state from the shared memory 10SMM, thereby, the source end and the target end no longer have state copies (because the state is stored in the shared memory 10SMM without the need to copy the state and without the need to copy the state over the network). The state of the device 10SoC1 can be transmitted through, for example, a Peripheral Component Interconnect Express (PCIe) or CXL link without using a network to transmit. Accordingly, seamless instant migration can be achieved and the requirements of Ultra-Reliable and Low Latency Communications (URLLC) of the fifth generation mobile communication technology (5G) can be met.

[0035] In one embodiment, since the devices 10SoC1 and 10SoC2 can be implemented by a system on a chip, the devices 10SoC1 and 10SoC2 themselves include processors 10PU1 and 10PU2 (e.g., central processing units), memories 10MM1 and 10MM2, or storage devices. In other words, unlike the need to allocate resources of a virtual central processing unit, virtual memory, or virtual storage device when restoring a virtual machine on the target end, after the device 10SoC2 (or 10SoC1) reads the state from the shared memory 10SMM, it does not need to perform virtual machine provisioning, which can save the time spent on allocating (computing, storage, or network) resources, thereby achieving seamless instant migration and meeting the requirements of ultra-reliable low-latency communication of 5G.

[0036] In one embodiment, live migration may involve migrating a user plane function (UPF) from a (busy) server to another (idle) server. If live migration takes too long, it will affect the user experience (for example, a self-driving car cannot react in time and may cause a car accident). System 10 can ensure that the user end (such as a self-driving car, a user device or Figure 6 camera shown) to a server (e.g. Figure 6 The artificial intelligence (AI) server shown has a latency of less than 1 millisecond.

[0037] Figure 2 Schematic diagram of a system 20 according to an embodiment of the present invention. The system 20 may include cabinets 20R1-20Ri, the cabinets 20R1-20Ri may include servers 20R1SVR1-20RiSVRm, respectively, the servers 20R1SVR1-20RiSVRm may include system single chips 20R1SoC11-20RiSoCmn, respectively, wherein i, m, and n are positive integers. In one embodiment, the structures of different cabinets (or servers) (for example, the number of servers or system single chips) may be different.

[0038] Similar to the system 10, part, whole or combination of the memory of the system single chip 20R1SoC11-20RiSoCmn or the server 20R1SVR1-20RiSVRm can form a shared memory. The server 20R1SVR1-20RiSVRm can use ( Figure 2 The CXL links (represented by double arrow dotted lines) are interconnected with each other. In other words, the system single chip can use the memory channel (or memory bus) (e.g., between the processor 10PU1 and the memory 10MM1) or the PCIe bus or CXL link (e.g., between the devices 10SoC1 and 10SoC2) to read and write the shared memory. For example, the system single chip 20R1SoC11 at the source end can store its state (e.g., user plane function) in the shared memory, and the system single chip 20R2SoC11 at the target end can use the CXL link to directly read the state from the shared memory to restore the state. Accordingly, seamless real-time migration of the system single chip across servers or even across cabinets can be achieved.

[0039] Figure 3Schematic diagram of a system 30 according to an embodiment of the present invention. The system 30 may include servers 30SVR1 and 30SVR2. The servers 30SVR1 and 30SVR2 may include hardware infrastructure 30HW1 and 30HW2, system-on-chip hypervisors 30HV1 and 30HV2, and operating systems 30OS11 to 30OS1n and 30OS21 to 30OS2j corresponding to system-on-chips 30SoC11 to 30SoC1n and 30SoC21 to 30SoC2j, respectively, where j is a positive integer. The servers 30SVR1 and 30SVR2 may utilize ( Figure 3 The CXL links (represented by double arrow dotted lines) are interconnected with each other.

[0040] To ensure safety, Figure 3 As shown, the SoC hypervisor 30HV1 (or 30HV2) is a first type SoC hypervisor, that is, the SoC hypervisor 30HV1 is directly installed on the hardware infrastructure 30HW1 and can directly access or control hardware resources. There is no additional operating system layer between the SoC hypervisor 30HV1 and the hardware infrastructure 30HW1 and is not easily attacked. In other words, the server (such as 30SVR1) does not have an operating system installed to improve security.

[0041] Please note that the operating systems 30OS11-30OS1n run on the system single chips 30SoC11-30SoC1n, while the system single chip hypervisor 30HV1 manages the hardware system single chips 30SoC11-30SoC1n instead of the operating systems 30OS11-30OS1n, so the system single chip hypervisor 30HV1 is different from the hypervisor that manages the virtual machines. The system single chip hypervisor 30HV1 or 30HV2 can be implemented using a combination of software, firmware or hardware.

[0042] In one embodiment, the operating system (e.g., 30OS11) may be an embedded Linux running in a system-on-chip (e.g., 30SoC11). In one embodiment, a plurality of system-on-chips (e.g., 30SoC11-30SoC14) may be disposed on an interface card (inserted into a server (e.g., 30SVR1)). In one embodiment, the hardware infrastructure (e.g., 30HW1) may include a CPU or a memory (e.g., a dynamic random access memory (DRAM)) of the system-on-chip (e.g., 30SoC11) or the server (e.g., 30SVR1).

[0043] In one embodiment, the technology of transferring the memory state of the system single chip from the source end to the target end can be divided into pre-copy memory migration and post-copy memory migration. If a memory page of the source end changes repeatedly during the migration period (i.e., becomes dirty), the post-copy memory migration uses the CXL shared memory to read / transmit the memory page only once, while the pre-copy memory migration reads / transmits the memory page multiple times corresponding to the occurrence of the change. The pre-copy memory migration retains the latest state of the system single chip of the source end during the migration period, and the state of the system single chip of the source end (e.g., 20R1SoC11) under the post-copy memory migration may be different from the state of the system single chip of the target end (e.g., 20R2SoC11) during the migration period. If the real-time migration of the target end fails, the pre-copy memory migration can successfully restore the system single chip, while the post-copy memory migration cannot restore the system single chip. Therefore, the present invention can adopt the pre-copy memory migration in one embodiment.

[0044] In one embodiment, the pre-copy memory migration can be divided into a pre-copy phase and a stop-and-copy phase. In the pre-copy phase, the SoC hypervisor (e.g., 30HV1) ensures that all memory pages of the SoC (e.g., 30SoC11) on the source side are stored in the CXL shared memory (e.g., copying these memory pages from the source side to the CXL shared memory), while the SoC is still running on the source side. If some memory pages are changed during the pre-copy phase, the changed memory pages will be copied again and again in multiple iterations (and stored in an area of ​​the CXL shared memory). When the number of remaining changed memory pages becomes small enough to perform the stop-and-copy phase, the pre-copy phase ends. On the other hand, if the speed at which the SoC changes memory pages is faster than the speed of iterative transmission to the target side (e.g., the speed at which the target side reads memory pages sequentially), the pre-copy phase will end after reaching a predetermined time limit or a maximum number of iterations, and enter the stop-and-copy phase. During the pause copy phase, the SoC at the source end is paused, and the remaining changed memory pages are transmitted to the target end (e.g., the target end reads the remaining changed memory pages), so that the SoC can be restored at the target end (e.g., SoC 30SoC2j). The downtime caused by the pause copy phase is approximately a few milliseconds to a few seconds, which is related to the number of changed memory pages transmitted during the downtime. If the SoC changes a large number of memory pages during the pre-copy phase, there may be a longer downtime.

[0045] exist Figure 3Before the live migration, the operating system (e.g., 30OS11) has been installed on the system-on-chip (e.g., 30SoC11). When the system-on-chip (e.g., 30SoC11) is used, there is no need to deploy, and there is no need to allocate resources such as a central processing unit, memory, or PCIe card (e.g., a graphics processor or a network interface controller (NIC)) to the system-on-chip. Therefore, unlike the existing virtual machine live migration, the present invention can meet the needs of immediate system-on-chip deployment and save the time spent on allocating (computing, storage, or network) resources.

[0046] In one embodiment, with CXL 3.0, a system-on-chip (e.g., 30SoC11) can be booted from a CXL shared memory between all CXL nodes across servers or across racks. The CXL shared memory may include a kernel (e.g., a Linux kernel) and a root filesystem with installed applications. During the booting process of a system-on-chip (e.g., 30SoC11), first, the boot loader of the system-on-chip may load / boot the kernel stored in the CXL shared memory, then the kernel mounts the root filesystem to the root directory (denoted as " / "), and then executes the initialization scripts accordingly according to the running level. In one embodiment, the booting of the system-on-chip (e.g., 30SoC11) may not involve resource allocation.

[0047] In terms of security, the isolation of the SoC is higher than that of the virtual machine. The SoC hypervisor (e.g., 30HV1) can prohibit the copying or pasting of files from a SoC (e.g., 30SoC11) to a host, a server (e.g., 30SVR1, 30SVR2) or another SoC (e.g., 30SoC12, 30SoC2j). The firewall for managing the network and PCIe is implemented in the SoC (e.g., 30SoC11), so even if a hacker breaks into the server (e.g., 30SVR1), the hacker must break into the SoC (e.g., 30SoC11) to allow the firewall to allow the copying or pasting of files. In other words, the virtual machine is essentially based on software to simulate physical resources, but hardware-based security is better than software-based security (e.g., a software-based virtual machine can be hacked by tampering with the data in the memory). Therefore, the present invention can improve security by directly running an operating system (e.g., 30OS11) or an application on the SoC (e.g., making the virtual machine absent from the SoC).

[0048] In one embodiment, a system-on-chip (e.g., 20R1SoC11 or 30SoC11) may include multiple cores, such as 8 ARM cores or 50 RISC-V cores. The ARM core may be used for data processing, such as Internet Protocol Security (IPsec), virtual switch (e.g., OpenvSwitch), or other data processing or acceleration or simulation of a hardware communication protocol engine. The RISC-V core may be used for signal processing, such as signal processing or acceleration such as baseband processing (satellite or 5G). In one embodiment, the number of RISC-V cores may be greater than the number of ARM cores. In the present invention, the ARM core or RISC-V core of the system-on-chip (e.g., 30SoC11) may be used to simulate the arithmetic logic unit (ALU) of a graphics processor. Similar to the data and signal hardware acceleration that a graphics processor can perform, the RISC-V core can also perform hardware acceleration.

[0049] Figure 4 It is a schematic diagram of a system on a chip 40SoC according to an embodiment of the present invention. The system on a chip 40SoC may include multiple CPU cores 40cM and 40cV. If the number of CPU cores 40cM belonging to the ARM core is 8, and the virtual ALU (virtual ALU, vALU) 40thdM used for data processing may be greater than 8 (for example, 16), many-to-many scheduling / mapping may be performed. If the number of CPU cores 40cV belonging to the RISC-V core is 50, and the virtual ALU 40thdV used for signal processing may be greater than 50 (for example, 100), many-to-many scheduling / mapping may be performed. Among them, a thread may represent a virtual ALU. In other words, virtual ALUs may be divided into two categories, and each category of virtual ALUs may perform many-to-many scheduling. Furthermore, the CPU cores 40cM and 40cV of the system-on-chip can be used to simulate the arithmetic logic unit of the graphics processor, thereby performing artificial intelligence acceleration.

[0050] Accordingly, the system-on-chip (e.g., 30SoC11) can realize data and signal hardware acceleration, and may not include a graphics processor (i.e., the graphics processor is vacant from the system-on-chip), thereby reducing costs, increasing design flexibility (without being limited by the size of the graphics processor), and (by increasing or decreasing the size of the graphics processor) Figure 5 The computing pool 50SoCpl shown in the figure can improve efficiency.

[0051] Figure 5 Schematic diagram of a system 50 according to an embodiment of the present invention. The system 50 may include a producer 50P, an arbitration controller 50ctrl, shared memories 50SMM1 and 50SMM2, a computing pool 50SoCpl, and a consumer 50C.

[0052] In one embodiment, the shared memories 50SMM1 and 50SMM2 may be, for example, CXL shared memories, which may be part, the whole, or a combination of the memories of the system single chip or the server, and may be implemented by the shared memories (e.g., 10SMM) of the systems 10 to 30. The shared memories 50SMM1 and 50SMM2 may be different or the same shared memory.

[0053] In one embodiment, the producer 50P may be a system-on-chip for producing a thread, a data source or a computing pool 50SoCpl, and the consumer 50C may be a system-on-chip for consuming a thread or a computing pool 50SoCpl. In one embodiment, the arbitration controller 50ctrl may be used to lock / unlock the shared memories 50SMM1 and 50SMM2. In one embodiment, the producer 50P, the arbitration controller 50ctrl or the consumer 50C may be implemented using a combination of software, firmware or hardware.

[0054] In one embodiment, the producer 50P and the computing pool 50SoCpl can access the shared memory 50SMM1 via the arbitration controller 50ctrl to ensure the atomic operation of the input data. In other words, the producer 50P can produce data and write the data to the shared memory 50SMM1, and the computing pool 50SoCpl can read the data from the shared memory 50SMM1. In order to avoid conflicts between the write operation of the producer 50P and the read operation of the computing pool 50SoCpl, the arbitration controller 50ctrl will arbitrate (for example, lock a certain memory or a certain storage unit and decide the operation), and prohibit the producer 50P (for example, to the storage unit) from performing a write operation when the computing pool 50SoCpl (for example, to a certain storage unit) performs a read operation, and prohibit the computing pool 50SoCpl from performing a read operation when the producer 50P performs a write operation. Similarly, the consumer 50C and the computing pool 50SoCpl can access the shared memory 50SMM2 via the arbitration controller 50ctrl to ensure the atomic operation of the output data. In order to avoid conflicts between the write operation of the computing pool 50SoCpl and the read operation of the consumer 50C, the arbitration controller 50ctrl will arbitrate and prohibit the consumer 50C from performing a read operation when the computing pool 50SoCpl is performing a write operation, and prohibit the computing pool 50SoCpl from performing a write operation when the consumer 50C is performing a read operation. In other words, once the arbitration controller 50ctrl is locked, only one of the producer 50P, the computing pool 50SoCpl or the consumer 50C can operate.

[0055] In one embodiment, the computing pool 50SoCpl is a system-on-chip pool composed / pooled by (the computing resources of) multiple system-on-chips (of the same or different servers, chassis or cabinets), which can be composed of, for example, central processing units of multiple system-on-chips.

[0056] In one embodiment, the computing pool 50SoCpl may have a performance index. According to the performance index, if the performance of the computing pool 50SoCpl or its system-on-chip is lower or higher than expected, the performance can be fine-tuned in advance by adding / removing at least one system-on-chip from the computing pool 50SoCpl. For example, please refer to Figure 2 and Figure 5 , the computing pool 50SoCpl may contain at a time point t1 Figure 2 The system single chip 20R1SoC11~20RiSoC1n shown in the first row. If (using the algorithm) it is determined that the performance at a time point t2 is insufficient (for example, the performance index is lower than a preset performance index threshold, or a performance index vector including at least one performance index (for example η h (t)) minus its corresponding performance indicator threshold vector, at least one element of a difference vector is less than 0), then Figure 2The system single chips 20R1SoC21-20R2SoC2n shown in the second row are added to the computing pool 50SoCpl. If (using the algorithm) it is determined that the performance at time point t2 is still insufficient, then Figure 2 The SoCs 20R3SoC21-20RiSoC2n or all SoCs 20R3SoC21-20RiSoCmn shown in the second row are added to the computing pool 50SoCpl. Thereafter, if (using an algorithm) it is determined that the performance is excessive at a time point t3 (for example, the performance index is higher than a preset performance index threshold, or all elements of the difference vector obtained by subtracting the corresponding performance index threshold vector from the performance index vector are greater than or equal to 0), then the SoCs 20R3SoC21-20RiSoC2n or all SoCs 20R3SoC21-20RiSoCmn shown in the second row are added to the computing pool 50SoCpl. Figure 2 The second row shows the system-on-chip 20R1SoC21~20RiSoC2n. In this way, Figure 2 The computing pool 50SoCpl composed of the system single chip 20R1SoC11~20RiSoC1n shown in the first row can execute a calculation / task, and Figure 2 The SoCs 20R1SoC21-20RiSoC2n shown in the second row can form / join another computing pool to perform another computing / task. In other words, the present invention can change the composition of the computing pool 50SoCpl in response to the producer 50P or the consumer 50C.

[0057] In one embodiment, the performance indicators of the computing pool 50SoCpl may be predicted and analyzed using a continuous time structural equation model (CTSEM) to predict the performance at a certain point in time (e.g., t) rather than the performance of a time period of a discrete time model. Since CTSEM may involve stochastic differential equations, which are continuous time operations, and it extracts the causal relationship in time to make accurate predictions, it is possible to better monitor the performance or predict whether the performance indicator at the next point in time (e.g., t) is lower or higher than its corresponding performance indicator threshold.

[0058] For example, in a continuous-time structural equation model, the performance indicators (as time causal variables) constitute a performance indicator vector η h (t) may be a function of time. As time goes by, the performance indicator vector η including at least one performance indicator h (t) can satisfy (Equation 1) or dη h (t) = (Aη h (t)+ξ h +Bz h +M∑ u x h,u δ(tu))dt+GdWh (t) (Equation 2). The matrix A can be used to qualitatively describe the performance index vector η using the self-effects on the diagonal and the cross-effects on the off-diagonal. h (t) time relationship. Matrix I is the identity matrix. Random vector ξ h The performance index vector η can be determined h (t) and can satisfy ξ h ~N(κ,φ ξ ), the vector κ can represent the continuous time intercept, and the matrix φ ξ It can be covariance (e.g., covariance between SoCs). The matrix B can represent the (fixed) time-independent prediction vector z h For the performance index vector η h (t), and the number of rows may not be equal to the number of columns. The time-independent prediction vector z h It is usually different for different SoCs, but can be a constant for different SoCs within the time range of the evaluation. The time-dependent prediction vector x h,u It can be observed at time u and only affects the performance indicator vector η at time u h (t), x h,u The pulse pair performance index vector η composed of δ(tu) h The influence of (t) can be expressed as a matrix M. The vector W h (s) can be an independent random walk in continuous time (for example, a Wiener process), dW h (s) can be a random error term. The lower triangular matrix G represents the performance index vector η h (t) changes. Satisfy Q = GG T The matrix Q represents the variance-covariance matrix of the diffusion process in continuous time. The matrices A, B, M, G, and the vector ξ can be obtained by fitting equation 1 or 2 to the data of the system-on-chip of the computing pool 50SoCpl (such as existing performance indicators). h 、z h 、x h,u , W h (s), time point u, or the value of the initial time point t0.

[0059] The continuous-time structural equation model can provide full information maximum likelihood estimation for continuous-time multivariate autoregressive models for time series or panel data. In one embodiment, the modeled dynamic random process is one process (i.e., time series and a single device) rather than multiple processes (i.e., panel data and multiple devices). Accordingly, the performance indicator vector η h(t) may be equal to 1, and there is only one performance indicator vector η1(t), and the performance indicator vector η1(t) may include the performance indicator of at least one system-on-chip of a computing pool (e.g., 50SoCpl), and a system-on-chip has at least one performance indicator. In another embodiment, the performance indicator vector η h (t) can correspond to different computing pools respectively, so the n computing pools will correspond to the performance indicator vectors η1(t)~η n (t), where h can be 1 to n and h is a positive integer.

[0060] Figure 6 Schematic diagram of a system 60 according to an embodiment of the present invention. System 60 may be implemented by systems 10 to 50. Live migration may involve restoring the state of a system-on-chip (e.g., a user plane function of a core network or other network element) on another system-on-chip, and allowing normal operation on the latter system-on-chip. Network elements may be started on a host (e.g., an x86 host or server) as software, a virtual machine, or a container, but system 60 does not use a virtual machine, but allows network elements to directly use the resources of the system-on-chip. System 60 can ensure that the latency from the camera to the artificial intelligence server is less than 1 millisecond.

[0061] Figure 7 Schematic diagram of a real-time migration method 70 according to an embodiment of the present invention. The real-time migration method 70 may be used in a shared memory of a system (eg, 10-60), and may include the following steps:

[0062] Step S700: Start.

[0063] Step S702: Receive a state of a first SoC (eg, 20R1SoC11), wherein the first SoC is used to write the state of the first SoC into a shared memory.

[0064] Step S704: Store the state of the first system single chip for reading, wherein a second system single chip (e.g., 20R2SoC11) reads the state of the first system single chip from a shared memory, and the shared memory is composed of at least a portion of a first memory in the first system single chip, at least a portion of a second memory in the second system single chip, or at least a portion of a third memory in the system.

[0065] Step S706: End.

[0066] The embodiments (e.g., systems 10-60) may be combined or implemented in various ways without conflict. In one embodiment, the state of the system-on-chip may include the content or state of the CPU, memory, storage device, or network connection. Any of the system-on-chip 20R1SoC11-20RiSoCmn, 30SoC11-30SoC1n, 30SoC21-30SoC2j, 40SoC may be implemented with device 10SoC1 or 10SoC2.

[0067] In summary, the present invention replaces virtual machines with system-on-chips, and uses CXL 3.0 to perform seamless system-on-chip instant migration across servers or cabinets, thereby avoiding interruptions and improving user experience on the edge computing platform of the edge cloud. The present invention can achieve instant system-on-chip deployment, so it can reduce latency and improve response on the edge computing platform of the edge cloud. The present invention has better system-on-chip isolation, so the edge computing platform of the edge cloud can provide hardware-based security that is superior to software-based security. The system-on-chip of the present invention can perform hardware acceleration (data or signal) for the host (or server), thereby saving a large number of expensive central processing unit cores on the host. The system-on-chip of the present invention can be pooled, and data or signals can be processed based on a producer-consumer model.

[0068] The above descriptions are only preferred embodiments of the present invention. All equivalent changes and modifications made according to the claims of the present invention should fall within the scope of the present invention.

Claims

1. A real-time migration method for a shared memory of a system, comprising: receiving a status of the first system-on-chip, in, The first system on a chip is used for writing the state of the first system on a chip into the shared memory; as well as The state is stored for reading, wherein the second system single chip is used to read the state from the shared memory, and the shared memory is composed of at least a portion of the first memory within the first system single chip, at least a portion of the second memory within the second system single chip, or at least a portion of the third memory within the system.

2. The instant migration method according to claim 1, in, The first system single chip and the second system single chip are arranged in different servers, chassis or cabinets.

3. The instant migration method according to claim 1, in, The first SoC or the second SoC does not utilize a network to copy or transmit the state.

4. The instant migration method according to claim 1, in, The first system on a chip or the second system on a chip has no virtual machine installed, and before the second system on a chip recovers the state, it is not necessary to configure a virtual machine or allocate virtual machine resources on the second system on a chip.

5. The instant migration method according to claim 1, in, The second SoC reads the state from the shared memory to restore the state in the second SoC, and the second SoC operates according to the state.

6. The instant migration method according to claim 1, in, The first system on a chip utilizes a plurality of central processing unit cores of the first system on a chip to simulate an arithmetic logic unit of a graphics processor, and the first system on a chip does not include a graphics processor.

7. The instant migration method according to claim 1, in, The arbitration controller is used for determining whether to allow the first system on chip or the second system on chip to access the shared memory, and is used for locking or unlocking the shared memory to prevent a read operation and a write operation from occurring simultaneously.

8. The instant migration method according to claim 1, in, The computing pool of the system includes a plurality of processors of a plurality of system single chips, and the plurality of system single chips include the first system single chip or the second system single chip.

9. The instant migration method according to claim 8, in, The system predicts at least one performance indicator of each of the multiple system single chips in the computing pool according to the continuous-time structural equation model, and determines whether the at least one performance indicator is lower than or higher than at least one performance indicator threshold to determine whether to increase or decrease the number of the multiple system single chips in the computing pool.

10. The instant migration method according to claim 9, wherein the performance index vector η formed by the at least one performance index of the plurality of SoCs in the computing pool h (t) Satisfaction A represents the qualitative matrix, t0 represents the initial time point, I represents the identity matrix, ξ h represents a random vector, B represents a transformation matrix, z h represents the time-independent prediction vector, M represents the coefficient matrix, x h,u represents the time-related prediction vector, u represents the pulse time point, W h (s) represents the random walk vector, and G represents the lower triangular matrix.

11. A system comprising: A first system on a chip, used to write the state of the first system on a chip into a shared memory; and A second system on a chip, coupled to the first system on a chip, is used to read the state from the shared memory, in, The shared memory is composed of at least a portion of a first memory in the first system on a chip, at least a portion of a second memory in the second system on a chip, or at least a portion of a third memory in the system.

12. The system according to claim 11, in, The first system single chip and the second system single chip are arranged in different servers, chassis or cabinets.

13. The system of claim 11, in, The first SoC or the second SoC does not utilize a network to copy or transmit the state.

14. The system of claim 11, in, The first system on a chip or the second system on a chip has no virtual machine installed, and before the second system on a chip recovers the state, it is not necessary to configure a virtual machine or allocate virtual machine resources on the second system on a chip.

15. The system of claim 11, in, The second SoC reads the state from the shared memory to restore the state in the second SoC, and the second SoC operates according to the state.

16. The system of claim 11, in, The first system on a chip utilizes a plurality of central processing unit cores of the first system on a chip to simulate an arithmetic logic unit of a graphics processor, and the first system on a chip does not include a graphics processor.

17. The system of claim 11, in, The arbitration controller is used for determining whether to allow the first system on chip or the second system on chip to access the shared memory, and is used for locking or unlocking the shared memory to prevent a read operation and a write operation from occurring simultaneously.

18. The system of claim 11, in, The computing pool of the system includes a plurality of processors of a plurality of system single chips, and the plurality of system single chips include the first system single chip or the second system single chip.

19. The system of claim 18, in, The system predicts at least one performance indicator of each of the multiple system single chips in the computing pool according to the continuous-time structural equation model, and determines whether the at least one performance indicator is lower than or higher than at least one performance indicator threshold to determine whether to increase or decrease the number of the multiple system single chips in the computing pool.

20. The system of claim 19, wherein the at least one performance indicator of the plurality of SoCs of the computing pool constitutes a performance indicator vector n h (t) Satisfaction A represents the qualitative matrix, t0 represents the initial time point, I represents the identity matrix, ξ h represents a random vector, B represents a transformation matrix, z h represents the time-independent prediction vector, M represents the coefficient matrix, x h,u represents the time-related prediction vector, u represents the pulse time point, W h (s) represents the random walk vector, and G represents the lower triangular matrix.