Container real-time migration method and system and medium

Abstract

The invention discloses a container real-time migration method and system and a medium. The method comprises the steps that a deduplication method based on content definition partitioning is introduced into container real-time migration, highly redundant memory pages are effectively detected and removed, and an optimal data reduction algorithm is dynamically selected to improve migration performance by means of an adaptive data reduction technology. Experimental results show that the total migration time can be reduced by 75%, and stable performance can be kept under different working load and bandwidth conditions.

Description

Technical Field

[0001] This application relates to the field of data migration technology, and in particular to methods, systems and media for real-time container migration. Background Technology

[0002] Containerization is a method of packaging software code, along with its dependencies, libraries, and configuration files, into a lightweight runtime environment called a container.

[0003] Due to their lightweight and high-density characteristics, containerized applications are widely deployed in cloud computing platforms. Live container migration is crucial for application reallocation in scenarios such as load balancing and fault recovery. However, existing pre-copy-based migration schemes neglect memory page redundancy during migration, while reducing data redundancy during migration can improve network transmission performance.

[0004] In addition, network transmission performance can change dynamically and is affected by real-time transmission bandwidth. It is necessary to adapt to different network bandwidths to maintain robust performance for real-time migration of containers for various workloads. Summary of the Invention

[0005] The main technical problem addressed in this application is to provide a real-time container migration method, system, and medium, which solves the problem that existing pre-copy-based migration schemes neglect the impact of memory page redundancy and dynamic changes in network bandwidth on data migration performance.

[0006] To solve the above technical problems, a technical solution adopted by this application is to provide a method for real-time container migration, including the steps: First step, perform the nth iteration data check on the container of the source node, capture the original nth dirty page and dump it to the disk, where 1 ≤ n ≤ N, and N represents the total number of required iterations and the corresponding number of dirty pages; Second step, according to the current conditions for switching between the main algorithm and the secondary algorithm, use the main algorithm to perform reduction processing on the nth dirty page, correspondingly obtaining the reduced nth main reduced dirty page, and the nth main processing parameters corresponding to the processing by the main algorithm; use the secondary algorithm to perform reduction processing on the nth dirty page, correspondingly obtaining the reduced nth secondary reduced dirty page, and the nth secondary processing parameters corresponding to the processing by the secondary algorithm; Third step, correspondingly select the nth main reduced dirty page, transmit it to the target node through the network, and collect real-time network bandwidth information; Fourth step, calculate the nth main migration index using the nth main processing parameters and the network bandwidth information, calculate the nth secondary migration index using the nth secondary processing parameters and the network bandwidth information, and compare the nth main migration index and the nth secondary migration index, which serves as the condition for switching between the main algorithm and the secondary algorithm in subsequent iterative processing; Fifth step, in the target node, restore the received nth main reduced dirty page to the nth dirty page; if the current n < N, update it to n + 1, and iterate back to the first step to the fourth step until the N original dirty pages in the container of the source node are transmitted to the target node and restored, and after the last transmission and reception, restore the container in the target node.

[0007] In some embodiments, in the second step, both the main algorithm and the secondary algorithm include a pre-copy algorithm, a data compression algorithm, or a data deduplication algorithm, and the currently used main algorithm is different from the secondary algorithm.

[0008] In some embodiments, in the second step, if the main algorithm or the secondary algorithm is the pre-copy algorithm, the corresponding nth main processing parameter or nth secondary processing parameter includes: the time to dump the dirty page to the disk , the time to archive the data directory , the data transmission time representing, the size of the nth dirty page .

[0009] In some embodiments, in the second step, if the main algorithm or the secondary algorithm is the data compression algorithm or the data deduplication algorithm, the corresponding nth main processing parameter or nth secondary processing parameter includes: the time to dump the dirty page to the disk , the time to archive the data directory , the data transmission time , the size of the nth dirty page , the application time corresponding to the data compression algorithm or the data deduplication algorithm The throughput corresponding to the data compression algorithm or the data deduplication algorithm Data reduction rate .

[0010] In some embodiments, network bandwidth is The nth primary migration metric or the nth secondary migration metric includes RAI:

[0011] in, , , Among them, ignoring and The impact.

[0012] In some embodiments, the conditions for switching between the main algorithm and the secondary algorithm include: when the RAI value of the secondary algorithm satisfies: Furthermore, if the RAI value of the secondary algorithm is lower than the RAI value of the primary algorithm for two consecutive iterations, a switch is triggered.

[0013] In some embodiments, the conditions for switching between the main algorithm and the sub-algorithm further include: if both the main algorithm and the sub-algorithm are the data compression algorithm or the data deduplication algorithm, and the RAI value of the main algorithm and the RAI value of the sub-algorithm are both greater than 1, and both RAI values ​​are greater than 1 in three consecutive iterations, then the main algorithm triggers a switch to the pre-copy algorithm.

[0014] In some embodiments, the main algorithm first processes the original dirty pages, calculates the throughput and reduction rate corresponding to the main algorithm processing, and generates reduced dirty pages. Subsequently, the dirty pages are transmitted to the target node. During the transmission process, the secondary algorithm processes the same original dirty pages again to obtain the throughput and reduction rate corresponding to the secondary algorithm processing.

[0015] This application also discloses a real-time container migration system, comprising: a checkpointer, a data reducer, a migration performance detector, and a data migrater at the source node, and a restorer at the target node; the checkpointer is used to perform iterative checkpointing operations on the container during the migration process, capturing the original dirty pages and dumping them to disk; the data reducer processes the original dirty pages using a data reduction algorithm and collects corresponding performance metrics during the processing, including selecting two algorithms: a main algorithm and a secondary algorithm; using the main algorithm to obtain the reduced dirty pages and their corresponding reduction throughput and reduction rate, and then using the secondary algorithm to calculate their corresponding throughput and reduction rate; the data migrater collects real-time network bandwidth information while transmitting the dirty pages reduced by the main algorithm; the migration performance detector calculates the RAI of each algorithm based on the performance metrics of the main algorithm, the performance metrics of the secondary algorithm, and the current network bandwidth, and dynamically decides whether to switch between the main algorithm and the secondary candidate algorithm based on the RAI values ​​of the two algorithms; the restorer restores the received reduced data into the original dirty pages in each iteration, and restores the container only in the last iteration.

[0016] This application also discloses a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the preceding methods.

[0017] The beneficial effects of this application are as follows: This application discloses a method, system, and medium for real-time container migration. The method includes introducing a content-defined chunking-based deduplication method into real-time container migration to effectively detect and remove highly redundant memory pages, and utilizing adaptive data reduction technology to dynamically select the optimal data reduction algorithm to improve migration performance. Experimental results show that this application can reduce the total migration time by up to 75% while maintaining stable performance under different workloads and bandwidth conditions. Attached Figure Description

[0018] Figure 1 This is a flowchart of one embodiment of the container live migration method of this application; Figure 2 This is a schematic diagram of an embodiment of the container live migration system of this application. Detailed Implementation

[0019] To facilitate understanding of this application, a more detailed description is provided below with reference to the accompanying drawings and specific embodiments. Preferred embodiments of this application are shown in the drawings. However, this application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure of this application.

[0020] It should be noted that, unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The term "and / or" as used in this specification includes any and all combinations of one or more of the associated listed items.

[0021] Virtualization has become a key cloud computing technology in modern data center design and operation to improve resource utilization. It decouples deployed applications from physical servers, enabling administrators to migrate or reallocate applications to cope with workload peaks and / or resource shortages.

[0022] Unlike traditional virtual machine (VM)-based virtualization, containerization is a lightweight and portable form of virtualization technology used to divide applications into microservices that can be deployed, updated, and scaled independently. Typical container technologies include Docker, Podman, and LXC.

[0023] Live container migration is the process of moving applications between different physical machines / clouds. It minimizes disruption to running services by enabling dynamic workload reallocation. Live container migration plays a crucial role in the management and maintenance of container cloud platforms in terms of load balancing, fault tolerance, and high availability.

[0024] To perform live migration, the computing platform first freezes the container on the source node and transfers the container's state (including memory, processes, file system, and network connectivity) to the target node. Subsequently, the platform restores the state and unfreezes the container on the target node. Finally, a rapid cleanup process is performed on the source node.

[0025] There are two main approaches to real-time migration: pre-copy and post-copy algorithms. The difference lies in the order of state transfer. In the pre-copy algorithm, memory is transferred to the target node iteratively until any remaining dirty memory can be transferred within a sufficiently short stop-and-copy phase, thus avoiding long container downtime. In the post-copy algorithm, the container state is immediately transferred to the target node, and execution control is switched immediately. The source node's memory is transferred in the background or acquired as needed while the target node container is running.

[0026] However, existing live container migration solutions primarily rely on User-space Checkpoint and Resume (CRIU) technology and pre-copy strategies to migrate the container's runtime state, root file system, and management context. These live container migration methods aim to minimize downtime (the period when application services are completely unavailable) while reducing the total migration time from start to finish by minimizing data transfer volume.

[0027] Live container migration is very similar to live virtual machine migration, especially when using a pre-copy approach. Each iteration only transfers memory pages modified since the last iteration—the original dirty pages. Most current optimization methods focus on dirty page prediction and data reduction algorithms to minimize data transfer scale. However, they haven't explored deduplication or workload-aware and network bandwidth-aware adaptive data reduction to optimize live container migration. Therefore, these approaches still face two technical challenges: (1) How to identify and eliminate redundancy in memory pages during transmission? (2) How to match data reduction algorithms with diverse workloads and unstable network bandwidth? To address the two technical challenges mentioned above, this application presents a novel live container migration solution designed to improve migration performance under varying workloads and network bandwidths.

[0028] First, this application introduces a content-defined chunking (CDC)-based deduplication algorithm in live container migration to efficiently detect and eliminate memory redundancy with a high reduction rate. Second, to adapt to different workloads and network bandwidths, this application calculates the Reduction Adaptation Index (RAI) during the migration process. This index is obtained by monitoring network bandwidth, reduction throughput, and reduction rate, and the RAI is used to dynamically select the optimal data reduction scheme for the next migration iteration.

[0029] This application utilizes the Reduction Adaptive Index (RAI) to support workload-aware and bandwidth-aware adaptive data reduction, improving performance under complex workloads and network conditions. Furthermore, experimental verification was conducted using an extended Podman framework. Experimental results show that the proposed live container migration solution can reduce total migration time by up to 75%, maintaining robust performance across various workloads under different network bandwidths while preserving considerable downtime.

[0030] In order to reduce the amount of data transmitted during real-time migration, this application focuses on two typical data reduction algorithms: data compression algorithms (hereinafter referred to as compression) and data deduplication algorithms (hereinafter referred to as deduplication).

[0031] (1) Compression is a typical data reduction algorithm that encodes information using fewer bits than the original representation. Lossless compression is a reversible process that uses statistical redundancy to reduce space overhead without losing any information. This application selects the lightweight lossless algorithm LZ4 to optimize real-time container migration. The LZ4 algorithm uses a dictionary-based sliding window technique to achieve extremely high-speed compression and decompression. It also provides configurable parameters that allow users to balance compression ratio and throughput, with a default priority to ensure high throughput.

[0032] (2) Deduplication can minimize storage or transmission overhead by identifying and eliminating redundant data blocks, retaining only unique data blocks. It has been widely used in backup storage, live virtual machine migration, and cloud storage systems. The basic process is to first divide the input data stream into multiple data blocks and generate a fingerprint for each data block using a cryptographic hash function. Then, a data block fingerprint index is constructed, and duplicate data is detected through query operations to determine whether the data block has been stored or transmitted.

[0033] Based on these two data reduction algorithms, the technical solution of this application was obtained through a series of experiments, thereby verifying the data redundancy phenomenon in real-time container migration and the necessity of adopting adaptive data reduction under different workloads and network bandwidth conditions.

[0034] The following is a detailed explanation.

[0035] refer to Figure 1 This embodiment provides a real-time container migration method, including the following steps: S101: First, perform the nth iteration data check on the container of the source node, capture the original nth dirty page and dump it to the disk, 1≤n≤N, where N represents the total number of iterations required and the corresponding number of dirty pages; S102: Second step, according to the current switching conditions of the main algorithm and the sub-algorithm, use the main algorithm to reduce the nth dirty page, and obtain the reduced nth main reduced dirty page, and the nth main processing parameters corresponding to the processing of the main algorithm; use the sub-algorithm to reduce the nth dirty page, and obtain the reduced nth secondary reduced dirty page, and the nth secondary processing parameters corresponding to the processing of the sub-algorithm. S103: The third step is to select the nth primary page to reduce the dirty page, transmit it to the target node via the network, and collect real-time network bandwidth information. S104: Fourth step, calculate the nth primary migration index using the nth primary processing parameters and the network bandwidth information, calculate the nth secondary migration index using the nth secondary processing parameters and the network bandwidth information, and compare the nth primary migration index and the nth secondary migration index as the conditions for switching between the primary algorithm and the secondary algorithm in subsequent iterative processing; S105: The fifth step is to restore the nth primary reduced dirty page received this time into the nth dirty page in the target node; if the current n < N, update it to n + 1, and iteratively return to the first to fourth steps until the N original dirty pages in the container of the source node are transferred to the target node and restored, and after the last transfer and reception, restore the container in the target node.

[0036] Through the above steps, the present application provides a novel real-time container migration method, aiming to improve the migration performance under diverse workloads and different network bandwidth conditions. The present application introduces deduplication into container migration and can dynamically select between different data reduction algorithms based on workload characteristics and network bandwidth. Experimental results show that the present application can significantly shorten the total migration time under various workloads and bandwidths. In addition, the present application uses RAI to dynamically select the optimal data reduction algorithm according to different workloads and network bandwidths at runtime, rather than relying on a fixed algorithm, which has good flexibility.

[0037] Furthermore, in the second step S102, there is a relative relationship between the primary algorithm and the secondary algorithm. That is, the data reduction algorithm used in the current iteration is the primary algorithm, and the reduced dirty page processed by this primary algorithm will be transferred to the target node. And the secondary algorithm also processes to obtain the corresponding reduced dirty page, but it is not transferred to the target node, but is used to obtain the processing parameters of the secondary algorithm, including throughput and reduction rate.

[0038] However, when in this iteration, according to "comparing the nth primary migration metric and the nth secondary migration metric" in the fourth step S104, if the current secondary algorithm is selected to replace the current primary algorithm, then in the second step S102 of the next iteration, the current secondary algorithm will be used as the primary algorithm, and the current primary algorithm will be switched to the secondary algorithm, and the status of these two algorithms is exchanged. Correspondingly, both the primary algorithm and the secondary algorithm include a pre-copy algorithm (abbreviated as the Origin scheme), a data compression algorithm (abbreviated as the Compression scheme), or a data deduplication algorithm (abbreviated as the Deduplication scheme), and the currently used primary algorithm and secondary algorithm are different.

[0039] The following further explains the nth primary processing parameters corresponding to the processing of the primary algorithm and the nth secondary processing parameters corresponding to the processing of the secondary algorithm. For reference, first analyze the composition of the single-iteration migration time under the original pre-copy scheme and data reduction schemes (including compression algorithm and deduplication algorithm). In the original pre-copy scheme without reduction, the total migration time can be expressed as the first calculation formula: , where represents the time to dump the dirty page to the disk, Indicates the time of the archived data catalog. Indicates data transmission time. Indicates the size of the dirty page. This indicates network bandwidth. Additionally... and The cost is very small and can be ignored.

[0040] In comparison, migration time after adopting data reduction techniques This can be expressed as the second calculation formula: , in, Indicates the time taken to apply the data reduction algorithm. This represents the throughput of the data reduction algorithm. This indicates the data reduction rate.

[0041] Total migration time For reference, the reduction adaptive exponent RAI is further defined as follows: and The ratio corresponds to the calculation of the nth primary migration index and the nth secondary migration index in step S104: , when This demonstrates that data reduction algorithms can improve migration performance. This application provides a method for quantitatively comparing migration performance among different data reduction algorithms, where a smaller RAI value indicates better performance.

[0042] Correspondingly, the main algorithm and sub-algorithm in this application include the aforementioned: compression algorithms (such as LZ4), deduplication algorithms, and also the aforementioned pre-copying method. This application selects the main algorithm and sub-algorithm for the next iteration based on the RAI value of different algorithms, i.e., decides whether to switch.

[0043] Preferably, in step S104, the conditions for switching between the main algorithm and the secondary algorithm include: triggering the switch when the RAI value of the secondary algorithm is lower than that of the main algorithm twice consecutively. Specifically, if the RAI values ​​of both algorithms are consistently greater than 1, it indicates that under the current workload and bandwidth conditions, data reduction cannot improve performance. In this case, this application will revert to the original pre-copy method.

[0044] Furthermore, the conditions for switching between the main algorithm and the sub-algorithm include: This application sets a threshold: when the RAI value of three consecutive iterations is greater than 1, a rollback to the pre-copy method is triggered.

[0045] Table 1 further illustrates the implementation process of the adaptive data reduction algorithm adopted in this application.

[0046]

[0047] This application includes at least two reduction algorithms: LZ4 compression and a CDC-based deduplication algorithm. The chunking algorithm within the deduplication process directly impacts RAI. There are two main types of chunking methods: Fixed-Size Chunking (FSC) and Content-Defined Chunking (CDC). FSC divides the data stream into fixed-size chunks, while CDC generates variable-size chunks based on the inherent characteristics of the data. However, FSC suffers from boundary drift; even a small change in the input data can cause boundary shifts in all subsequent chunks, significantly reducing deduplication efficiency. This problem also degrades the performance of FSC deduplication in real-time container migration.

[0048] In step S102, if both the main algorithm and the secondary algorithm are executed simultaneously on the original dirty pages, their respective processing parameters, including throughput and reduction rate, can be obtained, but this introduces additional system overhead. Therefore, this application proposes a pipelined mechanism: first, the main algorithm processes the original dirty pages, calculates the throughput and reduction rate corresponding to its processing, and generates reduced dirty pages. These dirty pages are then transmitted to the target node. During transmission, the secondary algorithm processes the same original dirty pages again, obtaining the throughput and reduction rate corresponding to its processing. This pipelined design limits system overhead to the level of a single algorithm, avoiding the cumulative overhead caused by running two algorithms simultaneously. Therefore, this application employs a pipelined data reduction mechanism to reduce overall system overhead, resulting in lower overall system overhead.

[0049] Based on the same inventive concept, this application also includes a container live migration system, such as... Figure 2 As shown, the source node includes: a checkpointer, a data reducer, a migration performance detector, and a data migrater, while the target node includes a restorer.

[0050] The checkpointer performs iterative checkpointing operations on the container during migration, capturing the original dirty pages and dumping them to disk. The data reducer processes the original dirty pages using a data reduction algorithm and collects corresponding performance metrics during the processing, including selecting two algorithms: a main algorithm and a secondary algorithm. The main algorithm is used to obtain the reduced dirty pages and their corresponding reduction throughput and reduction rate, and the secondary algorithm is used to calculate their corresponding throughput and reduction rate. The data migrater collects real-time network bandwidth information while transmitting the dirty pages reduced by the main algorithm. The migration performance detector calculates the RAI (Recovery Area Index) for each algorithm based on the performance metrics of the main algorithm, the secondary algorithm, and the current network bandwidth, and dynamically decides whether to switch between the main algorithm and the secondary candidate algorithm based on the RAI values ​​of these two algorithms. The restorer restores the received reduced data to the original dirty pages in each iteration, and only restores the container in the last iteration. For detailed explanations, please refer to the aforementioned... Figure 1 The explanation will not be repeated here.

[0051] The following experiments verify the practical application effect of this application. Currently, there are various containerization platforms, such as Docker, Podman, LXC, and OpenVZ. It is worth noting that Podman containers adopt a daemon-free architecture and a rootless execution mode, thereby reducing resource overhead and minimizing the risk of containers escaping to the host machine and gaining higher privileges. It is currently the default container engine for Fedora and RHEL distributions.

[0052] To support live container migration, several tools have been developed, including BLCR, DMTCP, and CRIU. Notably, CRIU avoids kernel modifications by implementing functionality entirely in user space and is the designated live migration tool for the Podman container platform.

[0053] The experimental prototype implementation of this application contains approximately 1600 lines of Go code (v1.24.1). Podman is extended as a checkpointer by enhancing its CLI support for the `--image-path` and `--parent-path` options and modifying its core library to correctly pass the necessary parameters to the CRIU, thus supporting pre-copy migration. The data reducer currently supports two algorithms: LZ4 compression and CDC-based deduplication. Specifically, a third-party LZ4 compression library is used, configured in fast mode, and executed using four concurrent Go goroutines for parallelism.

[0054] For deduplication, this application implements the FastCDC algorithm with an average block size of 8KB. Each data block uses xxhash to generate a unique fingerprint for redundancy detection, and in scenarios with tens of gigabytes of memory space on each node, xxhash is sufficient to ensure an extremely low probability of hash collisions. The data migrater uses gRPC to establish direct communication between the source and destination nodes.

[0055] Based on the same inventive concept, this application also provides a computer-readable storage medium storing program code, which can be called by a processor to execute the methods described in the above method embodiments.

[0056] Computer-readable storage media can be electronic storage devices such as flash memory, EEPROM (Electrically Erasable Programmable Read-Only Memory), EPROM, hard disk, or ROM. Optionally, computer-readable storage media include nontransitory computer-readable media. A computer-readable storage medium has storage space for program code that performs any of the method steps described above. This program code can be read from or written to one or more computer program products. The program code can be compressed in an appropriate form.

[0057] Therefore, this application discloses a method, system, and medium for real-time container migration. The method includes introducing a content-defined chunking-based deduplication approach into real-time container migration to effectively detect and remove highly redundant memory pages, and utilizing adaptive data reduction technology to dynamically select the optimal data reduction algorithm to improve migration performance. Experimental results show that this application can reduce the total migration time by up to 75% while maintaining stable performance under different workloads and bandwidth conditions.

[0058] The above are merely embodiments of this application and do not limit the scope of this patent application. Any equivalent structural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of this application.

Claims

1. A method for real-time container migration, characterized in that, Including the steps: In the first step, perform the nth iteration data check on the container of the source node, capture the original nth dirty page and dump it to the disk, where 1 ≤ n ≤ N, and N represents the total number of required iterations and the corresponding number of dirty pages; In the second step, according to the current conditions for switching between the main algorithm and the secondary algorithm, use the main algorithm to perform reduction processing on the nth dirty page, correspondingly obtaining the reduced nth main reduced dirty page, and the nth main processing parameters corresponding to the processing by the main algorithm; Use the secondary algorithm to perform reduction processing on the nth dirty page, correspondingly obtaining the reduced nth secondary reduced dirty page, and the nth secondary processing parameters corresponding to the processing by the secondary algorithm; In the third step, correspondingly select the nth main reduced dirty page, transmit it to the target node through the network, and collect real-time network bandwidth information; In the fourth step, calculate the nth main migration index using the nth main processing parameters and the network bandwidth information, calculate the nth secondary migration index using the nth secondary processing parameters and the network bandwidth information, and compare the nth main migration index and the nth secondary migration index, which serves as the condition for switching between the main algorithm and the secondary algorithm in subsequent iterative processing; In the fifth step, in the target node, restore the nth main reduced dirty page received this time to the nth dirty page; If the current n < N, update it to n + 1, and iterate back to the first step to the fourth step until the N original dirty pages in the container of the source node are transmitted to the target node and restored, and after the last transmission and reception, restore the container in the target node.

2. The container real-time migration method according to claim 1, characterized in that, In the second step, both the main algorithm and the secondary algorithm include a pre-copy algorithm, a data compression algorithm, or a data deduplication algorithm, and the currently used main algorithm is different from the secondary algorithm.

3. The container real-time migration method according to claim 2, characterized in that, In the second step, if the main algorithm or the sub-algorithm is the pre-copy algorithm, the corresponding nth main processing parameter or nth sub-processing parameter includes: the time to dump dirty pages to disk. The time of the archived data catalog Data transmission time This indicates the size of the nth dirty page. .

4. The container real-time migration method according to claim 3, characterized in that, In the second step, if the main algorithm or the sub-algorithm is the data compression algorithm or the data deduplication algorithm, the corresponding nth main processing parameter or nth sub-processing parameter includes: the time to dump dirty pages to disk. The time of the archived data catalog Data transmission time The size of the nth dirty page The application time corresponding to the data compression algorithm or the data deduplication algorithm The throughput corresponding to the data compression algorithm or the data deduplication algorithm Data reduction rate .

5. The container real-time migration method according to claim 4, characterized in that, Network bandwidth is The nth primary migration metric or the nth secondary migration metric includes the metric RAI: in, , ， Among them, ignoring and The impact.

6. The container real-time migration method according to claim 5, characterized in that, The conditions for switching between the main algorithm and the secondary algorithm include: when the RAI value of the secondary algorithm satisfies: Furthermore, if the RAI value of the secondary algorithm is lower than the RAI value of the primary algorithm for two consecutive iterations, a switch is triggered.

7. The container real-time migration method according to claim 6, characterized in that, The conditions for switching between the main algorithm and the secondary algorithm also include: if both the main algorithm and the secondary algorithm are the data compression algorithm or the data deduplication algorithm, and the RAI value of the main algorithm and the RAI value of the secondary algorithm are both greater than 1, and these two RAI values are both greater than 1 in three consecutive iterations, the main algorithm triggers a switch to the pre-copy algorithm.

8. The container real-time migration method according to claim 2, characterized in that, First, the main algorithm processes the original dirty page, calculates the throughput and reduction rate corresponding to the processing by the main algorithm, and generates a reduced dirty page, and then the dirty page is transmitted to the target node; During the transmission process, the secondary algorithm processes the same original dirty page to obtain the throughput and reduction rate corresponding to the processing by the secondary algorithm.

9. A container live migration system, characterized in that, Including: In the source node, it includes a checkpointor, a data reducer, a migration performance detector, a data migrator, and in the target node, it includes a restorer; The checkpointor is used to perform iterative checkpoint operations on the container during migration, capture the original dirty page and dump it to the disk; The data reducer processes the original dirty page through a data reduction algorithm and collects the corresponding performance metrics during the processing, including selecting two algorithms: the main algorithm and the secondary algorithm; using the main algorithm to obtain the reduced dirty page and its corresponding reduced throughput and reduction rate, and then using the secondary algorithm to calculate its corresponding throughput and reduction rate; The data migrator collects real-time network bandwidth information while transmitting the dirty page reduced by the main algorithm; The migration performance detector calculates the RAI of each algorithm based on the performance metrics of the main algorithm, the performance metrics of the sub-algorithms, and the current network bandwidth, and dynamically decides whether to switch between the main algorithm and the sub-candidate algorithm based on the RAI values ​​of the two algorithms. The restorer recovers the received shrunken data into the original dirty pages in each iteration, and only recovers the container in the last iteration.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 8.

Images