Realization method and system for server-oriented non-perceptual computing scene elastic control plane

Abstract

The invention provides an elastic control plane implementation method and system for a server-oriented non-perceptual computing scene, and the method comprises the steps: proposing a functional server non-perceptual computing control plane, and achieving the automatic telescoping of the control plane through splitting the control plane into stateless functions; a functional control plane is realized based on collaborative design of a stateless function and a dynamic partition strategy in non-perceptual computation of a server, and then a control plane capacity expansion and contraction strategy and a control component resource view allocation strategy based on convergence hash are designed. According to the invention, the technical problem of low control plane expansibility in a server non-perception computing scene due to the fact that the control plane is poor in elasticity and cannot meet delay and resource requirements under different loads is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of cloud computing, and in particular to a method and system for implementing an elastic control plane for server-aware computing scenarios. Background Art

[0002] In recent years, server-aware computing has become an increasingly popular cloud computing development paradigm due to its advantages such as high abstraction of resources and programming, on-demand usage billing, and dynamic expansion. In a server-aware computing system, the control plane is responsible for maintaining the state of the entire cluster and selecting specific functions to execute user requests. For example, the existing invention patent application document "A Separate Memory Management Method for Server-Aware Computing" with publication number CN117519953A, the existing method includes: applied to a server-aware computing platform, including: a platform user sends a request, and the request is placed in a task waiting queue for scheduling; the task scheduler determines the execution order of each task and the available local memory size according to the current cluster resource information and the characteristic information of each task in the task waiting queue, and allocates a working node to each task; the characteristic information represents a function of the execution time of the task changing with the local memory; the working node executes the assigned task, and when the memory used by the task exceeds the corresponding available local memory size, the remote memory pool is accessed to execute the task; the remote memory pool is a machine cluster for providing memory resources; the working node monitor monitors the task execution of each working node. Like the aforementioned non-perceptual computing platform in the prior art, the current mainstream server non-perceptual computing platform mostly uses a centralized control plane, and each control plane component needs to communicate frequently to complete state synchronization. In this process, due to the limited computing resources of the centralized components, there is a throughput upper limit, which cannot meet the throughput requirements under high load. At the same time, multiple control plane components rely on central storage for state synchronization, and frequent database reading and writing will cause non-negligible communication overhead. For example, under the current platform, the central scheduler can only schedule less than 200 function instances per second. The communication delay caused by centralized communication can account for 90% of the total delay of a single request service. This causes the total delay of the request service to exceed 100ms, which violates the service level objective (SLO). Therefore, how to solve the problem of low scalability of the control plane caused by the communication and computing bottlenecks of the current platform control plane is an important challenge in the field of server non-perceptual computing.

[0003] Existing work has proposed an optimization scheme on the centralized control plane, using direct connection between components and distributed control components to reduce the control plane overhead. For example, the existing invention patent application document "Server-aware computing scheduling method for resource-decoupled data centers" with publication number CN116302404A, the existing method includes: determining the corresponding task type according to the received task request RPC; determining the distribution ratio corresponding to the task type according to the task type; and distributing the tasks in the task type to the computing node or storage node for execution with the corresponding probability according to the distribution ratio. This technology reduces the control plane overhead from two aspects, thereby improving the control plane throughput. First, it uses direct transmission between components to avoid expensive database read and write operations, thereby reducing communication overhead. Second, it uses multiple distributed components to replace the original centralized control component to process requests in parallel to improve the throughput of the control component. Despite this, this technology still has some limitations: First, there is bottleneck transfer, and there is still a throughput bottleneck when facing burst loads. In the real server-aware computing production environment, high bursts of load are inevitable. Second, it is still a static control plane, which requires platform developers to estimate the load size and statically configure the control plane resources. When facing fluctuating loads, there will still be problems of idle resources and insufficient resources. Therefore, although the current work has made improvements to a certain extent, it still fails to completely solve the expansion problem of the control plane under server-unaware computing.

[0004] The existing invention patent application document "A server-unaware computing adaptive resource scheduling method, system and computer device" with publication number CN116126488A, the existing method includes: obtaining workload log data of the server-unaware computing function, obtaining characteristic indicator analysis of the workload log data, determining the threshold range θ of the function according to the characteristic indicator, determining the type of function, and selecting a corresponding scheduling strategy according to the threshold range θ of the function. The server-unaware computing adaptive resource scheduling method, system and computer device provided by this application analyze the workload in the actual generation environment, and set thresholds through a multi-target threshold mechanism according to multiple characteristic indicators, so that it can automatically select its appropriate resource scheduling mechanism under different threshold conditions, thereby realizing fine-grained resource management of server-unaware computing.

[0005] The existing public document "Function Scheduling Based on Spatiotemporal Characteristics in Server-Aware Computing Scenarios" analyzes the function scheduling problem in server-aware computing scenarios and locates three factors that affect the function completion time, namely queuing time, startup time, and execution time. Based on this analysis, a mathematical model is proposed to formally model the function scheduling problem in server-aware computing scenarios. Secondly, a server-aware computing scheduling algorithm FuncSched based on function spatiotemporal characteristics is proposed. The algorithm considers the function execution time and function startup time in the time dimension, and the function resource occupancy in the spatial dimension.

[0006] Existing public literature fully considers workload-related characteristic indicators when designing scheduling strategies, striving to maximize resource utilization and minimize function execution time. However, the control plane architecture in the aforementioned existing literature ignores the problem of high workload volatility. When faced with high concurrent requests that exceed the estimated load, the existing control plane architecture can easily become a bottleneck in the system, causing request timeouts and the system to fail to achieve the expected service quality goals. When the existing scheduling strategy handles extremely fluctuating loads, the performance of the control plane architecture will significantly decrease, thereby affecting the reliability and response speed of the overall system. This is because the traditional control plane architecture did not fully consider its scalability when it was first designed. As the number of requests surges, the number of concurrent tasks that the system needs to handle also increases accordingly, and the control plane architecture with fixed resource configuration is difficult to quickly adapt to such changes.

[0007] In summary, the existing technology has the technical problem that the control plane has poor elasticity and cannot meet the delay and resource requirements under different loads, resulting in low scalability of the control plane in the server-unaware computing scenario. Summary of the invention

[0008] The technical problem to be solved by the present invention is: how to solve the technical problem in the prior art that the control plane has poor elasticity and cannot meet the delay and resource requirements under different loads, resulting in low scalability of the control plane in the server-unaware computing scenario.

[0009] The present invention adopts the following technical solutions to solve the above technical problems: A method for realizing an elastic control plane for server-unaware computing scenarios includes:

[0010] S1. Functionalize the server-aware computing control plane, that is, split the server-aware computing control plane into stateless control functions to perform automatic scaling actions of the server-aware computing control plane. As shown in Table 1, the functions of the control plane are split into three independent modules: the distributor, the scheduler, and the cluster manager, to ensure that they can independently scale up and down to achieve corresponding functions.

[0011]

[0012] In the bottom functional control plane, the stateless control function is used to perform user interaction operations and execute the request distribution operation of the server non-aware computing control plane to perform instance scheduling, wherein the server non-aware computing control plane includes: a top control plane and a bottom functional control plane;

[0013] S2, based on the stateless functions and dynamic partitioning strategy in the server-unaware computing control plane, collaboratively design a functional control plane;

[0014] S3. Design the control plane expansion and contraction strategy. Based on the converged hash, design the control component resource view allocation strategy to handle different demands when the load fluctuates, balance the decision speed and decision quality, and divide the resource view for the control instances corresponding to different stateless control functions.

[0015] The present invention proposes an elastic control plane technology for server-unaware computing scenarios, which supports automatic expansion and contraction of the control plane to meet the throughput needs of users. The present invention realizes dynamic expansion and contraction of the control plane by splitting the control plane into stateless functions, thereby improving the elasticity of the control plane, meeting the delay and resource requirements under different loads, and solving the problem of low scalability of the control plane in server-unaware computing scenarios.

[0016] In a more specific technical solution, in S1, the top-level control plane includes: a gateway, a top-level controller, and a global state storage.

[0017] In a more specific technical solution, the gateway is used as the front-end API endpoint, the gateway is used to accept external function requests, the execution function of the external function request is determined by calling the request distribution function, and the external function request is forwarded to the execution function.

[0018] In a more specific technical solution, a top-level controller is used to coordinate the calling relationship between control function instances, and resource views are divided correspondingly for instances of the underlying functional control plane.

[0019] Compared with the existing work, the present invention proposes a functionalized server-aware computing control plane, which realizes automatic scaling of the control plane by splitting the control plane into stateless functions.

[0020] The present invention utilizes the top-level controller to coordinate the calling relationship between control function instances and divides resource views for the bottom-level functionalized control plane instances to reduce the probability of decision conflicts.

[0021] In more specific technical solutions, S3 includes:

[0022] S31, using a preset scaling engine to monitor the load of each stateless control function, when the load is greater than a preset load threshold, using the scaling engine to trigger the scaling process, calling a preset scheduling function to select a deployment location for the control instance of the stateless control function;

[0023] S32, performing a cold start operation;

[0024] S33, performing an adding operation on the functionalized control plane to adjust the throughput parameters of the functionalized control plane;

[0025] S34. When a decision conflict or failure occurs, the resource view is dynamically divided into different control instances to avoid instance decision conflicts of no less than two stateless control functions.

[0026] The present invention realizes a functionalized control plane based on the collaborative design of stateless functions and dynamic partitioning strategies in server-unaware computing, and further designs a control plane expansion and contraction strategy and a control component resource view allocation strategy based on convergent hashing.

[0027] In a more specific technical solution, S31 also includes:

[0028] S311, when the functionalized control plane is expanded, obtaining and using request latency as an expansion and contraction indicator to perceive load tilt information of the functionalized control plane;

[0029] S312: when the request delay of the stateless control function is greater than or equal to the preset delay threshold, use the scaling engine to offload the redundant tasks to the idle control instance;

[0030] S313: If there is no idle stateless control function, the expansion process is triggered by the expansion and contraction engine, the scheduling function is called to expand the capacity to obtain a new control function, and the task is unloaded to the new control function;

[0031] S314: When the request delay is less than the preset delay threshold, the scaling engine is used to trigger the scaling process, and the top-level controller is used to select the stateless control function with the least tasks to perform the locking operation;

[0032] S315. When the current task is completed, the resource control authority is released, the current new control function is deleted, and the scaling down operation on the functionalized control plane is completed.

[0033] The present invention uses a scaling engine to monitor the load of each control function. When the load is high, the scaling process is triggered, and the scheduling function is called to select a deployment location for the control function instance. After the cold start is completed, a functionalized control plane is added to improve the throughput of the control plane.

[0034] In a more specific technical solution, S32 also includes:

[0035] S321. When the functionalized control plane is expanded, when the expansion process is triggered, the stateless control function is preferentially scheduled using the preset scheduling function instance in the control instance to obtain the scheduling result;

[0036] S322, causing the scheduling function instance to synchronize the scheduling result to the corresponding working node to perform a cold start operation;

[0037] S323, registering with the top-level controller using the new control function to obtain a resource view;

[0038] S324, when the top-level controller monitors the registration action of the new control instance, it calculates the resource view allocated to the new control instance, notifies the component holding the resource view, performs a view release operation, and transmits the resource view to the new control instance through a view reallocation operation;

[0039] S325. When the view reallocation operation is completed, the top-level controller is notified through the service discovery mechanism to monitor the new instance addresses of the remaining components that generate the new control instance changes.

[0040] In a more specific technical solution, S34 includes:

[0041] S341. During the dynamic partitioning and detection operation of the functionalized control plane, each control instance of the stateless control function preferentially selects applicable resources in the local resource view;

[0042] S342: If there are no applicable resources for the control instance in the current stateless control function, the stateless control function requests the other stateless control functions at the same level to select applicable resources.

[0043] In a more specific technical solution, during the detection operation of S341, a detection cache mechanism is used to record detection feedback and probes, and to mark resource restriction information of both parties in the detection operation to establish a control function resource cache table.

[0044] The present invention can establish a control function resource cache table in limited detection, skip instances that cannot meet the requirements in the cache, avoid redundant detection processes, and greatly improve the detection success rate.

[0045] In a more specific technical solution, the elastic control plane implementation system for server-aware computing scenarios includes:

[0046] The plane functionalization and instance scheduling module is used to functionalize the server imperceptible computing control plane, split the server imperceptible computing control plane into stateless control functions, so as to perform automatic scaling actions of the server imperceptible computing control plane. In the bottom functionalized control plane, the stateless control function is used to perform user interaction operations and execute request distribution operations of the server imperceptible computing control plane to perform instance scheduling. The server imperceptible computing control plane includes: a top-level control plane and a bottom-level functionalized control plane.

[0047] The functionalized plane design module is used to collaboratively design a functionalized control plane based on the stateless functions and dynamic partitioning strategies in the server-unaware computing control plane. The functionalized plane design module is connected with the plane functionalization and instance scheduling module.

[0048] The resource view partitioning module is used to design the control plane expansion and contraction strategy. Based on the converged hash, the control component resource view allocation strategy is designed to handle different demands when the load fluctuates, and to balance the decision speed and decision quality. The resource view is partitioned for the control instances corresponding to different stateless control functions. The resource view partitioning module is connected to the functional plane design module.

[0049] Compared with the prior art, the present invention has the following advantages:

[0050] The present invention proposes an elastic control plane technology for server-unaware computing scenarios, which supports automatic expansion and contraction of the control plane to meet the throughput needs of users. The present invention realizes dynamic expansion and contraction of the control plane by splitting the control plane into stateless functions, thereby improving the elasticity of the control plane, meeting the delay and resource requirements under different loads, and solving the problem of low scalability of the control plane in server-unaware computing scenarios.

[0051] The present invention utilizes the top-level controller to coordinate the calling relationship between control function instances and divides resource views for the bottom-level functionalized control plane instances to reduce the probability of decision conflicts.

[0052] The present invention realizes a functionalized control plane based on the collaborative design of stateless functions and dynamic partitioning strategies in server-unaware computing, and further designs a control plane expansion and contraction strategy and a control component resource view allocation strategy based on convergent hashing.

[0053] The present invention uses a scaling engine to monitor the load of each control function. When the load is high, the scaling process is triggered, and the scheduling function is called to select a deployment location for the control function instance. After the cold start is completed, a functionalized control plane is added to improve the throughput of the control plane.

[0054] The present invention can establish a control function resource cache table in limited detection, skip instances that cannot meet the requirements in the cache, avoid redundant detection processes, and greatly improve the detection success rate.

[0055] In order to solve the problem that the control plane architecture with fixed resource configuration is difficult to quickly adapt to the surge in the number of concurrent tasks due to insufficient consideration of expansion capabilities, this application proposes a new control plane architecture design, with special attention to scalability under fluctuating loads. By reconstructing the control plane architecture into a serverless computing function, the system can flexibly adjust resource configuration according to real-time load conditions to achieve automatic expansion and contraction. This architecture not only improves the response speed and stability of the system in high-concurrency scenarios, but also effectively avoids the degradation of service quality due to insufficient resources. Compared with the traditional control plane architecture, the design of this serverless computing function can better adapt to the variable load requirements in modern computing environments and provide higher elasticity and scalability. Whether facing sudden large-scale requests or daily stable loads, the new architecture can operate efficiently to ensure that the system continues to provide high-quality services. This innovation will significantly improve the performance of the scheduling system under various load conditions and provide a solid technical foundation for future applications.

[0056] The present invention solves the technical problems existing in the prior art that the control plane has poor elasticity and cannot meet the delay and resource requirements under different loads, resulting in low scalability of the control plane in the server-unaware computing scenario. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 This is a schematic diagram of the basic steps of a method for implementing an elastic control plane for server-aware computing scenarios according to Embodiment 1 of the present invention;

[0058] Figure 2 This is a schematic diagram of the system architecture for implementing an elastic control plane for server-aware computing scenarios according to Embodiment 1 of the present invention;

[0059] Figure 3 It is a schematic diagram of the principle of resource view allocation strategy of the control component of the converged hash in Example 1 of the present invention;

[0060] Figure 4 This is a schematic diagram of dynamic partitioning and detection of the control plane in Example 1 of the present invention;

[0061] Figure 5 This is a schematic diagram of data flow processing in the control plane expansion process of embodiment 1 of the present invention;

[0062] Figure 6 This is a schematic diagram of the control plane scheduling process of Embodiment 1 of the present invention;

[0063] Figure 7 This is a schematic diagram of throughput and delay comparison of Example 1 of the present invention;

[0064] Figure 8 This is a schematic diagram of the scalability of Embodiment 1 of the present invention;

[0065] Fig. 9 This is a schematic diagram of resource usage under fluctuating load in Example 1 of the present invention. DETAILED DESCRIPTION

[0066] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in combination with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0067] Example 1

[0068] like Figure 1 As shown, the method for implementing the elastic control plane for server-aware computing scenarios provided by the present invention includes the following basic steps:

[0069] S1. Functionalize the server-aware computing control plane and split it into stateless functions to achieve automatic scaling of the control plane.

[0070] In this embodiment, in order to make up for the shortcomings of the existing solutions and achieve the purpose of reducing the control plane overhead, an elastic control plane technology for server-unaware computing scenarios is proposed, and the existing server-unaware computing control plane is split into stateless functions to achieve dynamic scaling of the control plane. It can significantly reduce the overhead of communication between components and control plane decisions, and effectively reduce resource overhead in low-load scenarios.

[0071] like Figure 2 As shown, in this embodiment, the control plane includes: a top-level control plane and a bottom-level functionalized control plane. In this embodiment, the top-level control plane includes but is not limited to: a gateway, a top-level controller, and a global state storage.

[0072] In this embodiment, the gateway is used as the front-end API endpoint. The gateway accepts external function requests, determines the function to execute the request by calling the request distribution function, and forwards the request to the function. The top-level controller coordinates the calling relationship between the control function instances and divides the resource view for the underlying functional control plane instance to reduce the probability of decision conflicts.

[0073] S2. The present invention is based on the stateless function and dynamic partitioning strategy in the server-unaware computing, and collaboratively designs and implements a functionalized control plane;

[0074] In this embodiment, the underlying functionalized control plane includes: the stateless control function split from the original control plane, which is responsible for interacting with the user and realizing the request distribution, instance scheduling and other functions of the original control plane.

[0075] S3. Design control plane expansion and contraction strategies and control component resource view allocation strategies based on convergence hashing.

[0076] In this embodiment, the scaling engine monitors the load of each control function. When the load is high, it triggers the scaling process and calls the scheduling function to select a deployment location for the control function instance. After the cold start is completed, the functionalized control plane is added to improve the throughput of the control plane.

[0077] like Figure 3 As shown, in this embodiment, when the control plane is expanded, the system will splice the new instance (such as the scheduler instance) with the resources to be allocated to calculate the hash value and sort it with other instances. Resources will be allocated to the instance with the highest hash value. The converged hash can evenly distribute data, ensure load balancing of each instance of the control plane, and improve resource utilization. Its design only needs to redistribute a small amount of data when the number of nodes changes, significantly reducing the overhead of data migration and improving the stability and performance of the control plane when it is expanded or reduced.

[0078] In this embodiment, in response to decision conflicts and failures, resource views are dynamically divided into different control function instances to avoid decision conflicts among multiple control function instances.

[0079] like Figure 4 As shown in the figure, during the dynamic partitioning and detection operation of the control plane, each control function instance has a different resource view, and preferentially selects resources that meet the conditions in the local resource view; if no resources that meet the conditions can be found, the control function will ask the control function of the same level whether it can complete the task; when the load is low, a single control function has a larger resource view, high scheduling quality, and a higher detection success rate; when the load is high, there are more control function instances, and multiple control function instances make decisions together, which greatly improves the throughput speed of the control plane.

[0080] The aforementioned method meets different requirements under fluctuating loads and achieves a trade-off between decision speed and decision quality.

[0081] In this embodiment, in order to improve the detection speed of the control function, a detection cache mechanism is introduced. By recording detection feedback and probes, marking the resource limitations of both detection parties, a control function resource cache table can be established in limited detection. Instances that cannot meet the requirements in the cache are skipped to avoid redundant detection processes, greatly improving the detection success rate.

[0082] In the control plane expansion operation of this embodiment, request delay is used as the expansion and contraction indicator, which can effectively perceive the control plane load inclination and other situations; when the control function processing delay exceeds the threshold, the expansion and contraction engine will unload redundant tasks to idle control instances; if there is no idle control function, the expansion and contraction engine will trigger the expansion process, call the scheduling function to expand a new control function, and unload the task to the new control function; when the delay is less than the specified threshold, the expansion and contraction engine will trigger the contraction process, and the top-level controller will select the control function with the least tasks to lock it, wait for the current task to be completed, release the resource control right, delete the control function, and complete the contraction of the control plane.

[0083] like Figure 5 As shown, in the expansion process of the control plane in this embodiment, when the load of a control function instance is too high to trigger the expansion process, the scheduling of the control function is completed by the existing scheduling function instance with the highest priority, the same as the user function; after the scheduling is completed, the scheduling instance directly synchronizes the scheduling result to the corresponding working node for cold start; after the startup is completed, the control function will actively register with the top-level controller to obtain its resource view; after listening to the registration of the new control instance, the top-level controller will calculate the resource view allocated to the new instance, and notify the component holding the corresponding view to release the view and transmit it to the new control instance; after completing the view reallocation, the top-level controller will notify other components that listen to changes in this type of control instance of the address of the new instance through the service discovery mechanism.

[0084] Compared with existing work, the present invention can increase the maximum throughput of the control plane to 89,935 requests per second, and reduce resource overhead by more than 87%. The present invention can increase the throughput by more than 8.6 times under the same service level objective (SLO). The present invention proposes to split the control plane into stateless functions, and realize dynamic expansion and contraction of the control plane when facing fluctuating loads, which not only greatly improves the throughput of the control plane, but also greatly reduces the overhead of the control plane. Under the same load, the present invention can reduce the overhead by 87% compared with existing work.

[0085] For the control plane scheduling process, see Figure 6 . After receiving the request, the control plane queries for available function instances. If there is an available instance, the request is forwarded to the available function instance (hot call). If there is no available instance, the expansion process is triggered, and the scheduling request will be forwarded to the scheduler, which selects an available node to start the function instance. The cluster manager synchronizes the instance status and node status to the distributor and scheduler respectively. Finally, the distributor distributes the request to the new instance.

[0086] Experimental setup:

[0087] Since this application focuses on the scheduling overhead of the control plane, 210 containers are used to simulate nodes instead of real function instances. For details, see the comparison of throughput and latency. Figure 7 ; For extensibility, see Figure 8 ; For resource usage under fluctuating loads, see Fig. 9 .

[0088] index:

[0089] 1.99 percentile scheduling delay: The time from when the gateway receives the request to when the scheduling is completed is the scheduling delay.

[0090] 2. Throughput: the number of instances scheduled per second

[0091] Compare systems:

[0092] 1. OpenFaaS: Centralized Scheduling Architecture

[0093] 1.Ray: Bottom-up Scheduling Architecture

[0094] 2. Atoll: Two-tier scheduling architecture

[0095] 3.Sparrow: Fully distributed scheduling architecture

[0096] In summary, the present invention proposes an elastic control plane technology for server-unaware computing scenarios, which supports automatic expansion and contraction of the control plane to meet the throughput needs of users. The present invention realizes dynamic expansion and contraction of the control plane by splitting the control plane into stateless functions, thereby improving the elasticity of the control plane, meeting the delay and resource requirements under different loads, and solving the problem of low scalability of the control plane in server-unaware computing scenarios.

[0097] The present invention utilizes the top-level controller to coordinate the calling relationship between control function instances and divides resource views for the bottom-level functionalized control plane instances to reduce the probability of decision conflicts.

[0098] The present invention realizes a functionalized control plane based on the collaborative design of stateless functions and dynamic partitioning strategies in server-unaware computing, and further designs a control plane expansion and contraction strategy and a control component resource view allocation strategy based on convergent hashing.

[0099] The present invention uses a scaling engine to monitor the load of each control function. When the load is high, the scaling process is triggered, and the scheduling function is called to select a deployment location for the control function instance. After the cold start is completed, a functionalized control plane is added to improve the throughput of the control plane.

[0100] The present invention can establish a control function resource cache table in limited detection, skip instances that cannot meet the requirements in the cache, avoid redundant detection processes, and greatly improve the detection success rate.

[0101] The present invention solves the technical problems existing in the prior art that the control plane has poor elasticity and cannot meet the delay and resource requirements under different loads, resulting in low scalability of the control plane in the server-unaware computing scenario.

[0102] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for implementing an elastic control plane for server-aware computing scenarios, characterized in that: The method comprises: S1. Functionalize the server-unaware computing control plane, split the server-unaware computing control plane into stateless control functions to perform automatic scaling actions of the server-unaware computing control plane, and use the stateless control functions in the bottom functionalized control plane to perform user interaction operations and request distribution operations of the server-unaware computing control plane to perform instance scheduling, wherein the server-unaware computing control plane includes: a top-level control plane and a bottom-level functionalized control plane; S2. Based on the stateless function and dynamic partitioning strategy in the server-unaware computing control plane, collaboratively design a functionalized control plane; S3. Design a control plane expansion and contraction strategy. Based on the converged hash, design a control component resource view allocation strategy to handle differential demands when the load fluctuates, and balance decision speed and decision quality to divide the resource view for the control instances corresponding to different stateless control functions.

2. The method for implementing a flexible control plane for server-aware computing scenarios according to claim 1 is characterized in that: In S1, the top-level control plane includes: a gateway, a top-level controller, and a global state storage.

3. The method for implementing a flexible control plane for server-aware computing scenarios according to claim 2 is characterized in that: The gateway is used as the front-end API endpoint, and the gateway is used to accept external function requests. By calling the request distribution function, the execution function of the external function request is determined, and the external function request is forwarded to the execution function.

4. The method for implementing a flexible control plane for server-aware computing scenarios according to claim 2 is characterized in that: The top-level controller is used to coordinate the calling relationship between the control function instances, and the resource views are divided accordingly for the instances of the bottom-level functionalized control plane.

5. The method for implementing a flexible control plane for server-aware computing scenarios according to claim 1, characterized in that: The S3 includes: S31, using a preset scaling engine to monitor the load of each stateless control function, and when the load is greater than a preset load threshold, using the scaling engine to trigger a scaling process, and calling a preset scheduling function to select a deployment location for the control instance of the stateless control function; S32, performing a cold start operation; S33, performing an adding operation on the functionalized control plane to adjust a throughput parameter of the functionalized control plane; S34. When a decision conflict or failure occurs, the resource view is dynamically divided into different control instances to avoid instance decision conflicts of no less than two stateless control functions.

6. The method for implementing a flexible control plane for server-aware computing scenarios according to claim 5 is characterized in that: The S31 further includes: S311, when the functionalized control plane is expanded, obtaining and using request delay as an expansion and contraction indicator to perceive load tilt information of the functionalized control plane; S312: when the request delay of the stateless control function is greater than or equal to a preset delay threshold, using the scaling engine to offload redundant tasks to the idle control instance; S313: If there is no idle stateless control function, the expansion process is triggered by the expansion and contraction engine, the scheduling function is called to expand the capacity to obtain a new control function, and the task is unloaded to the new control function; S314: When the request delay is less than the preset delay threshold, the scaling engine is used to trigger the scaling process, and the top-level controller is used to select the stateless control function with the least tasks to perform a locking operation; S315. When the current task is completed, release the resource control authority, delete the current new control function, and complete the scaling down operation on the functionalized control plane.

7. The method for implementing a flexible control plane for server-aware computing scenarios according to claim 5, characterized in that: The S32 further includes: S321: When the functionalized control plane is expanded, when the expansion process is triggered, the stateless control function is preferentially scheduled using the preset scheduling function instance in the control instance to obtain a scheduling result; S322, enabling the scheduling function instance to synchronize the scheduling result to the corresponding working node to perform the cold start operation; S323, registering with the top-level controller using the new control function to obtain the resource view; S324, when the top-level controller monitors the registration action of the new control instance, calculates the resource view allocated to the new control instance, notifies the component holding the resource view, performs a view release operation, and transmits the resource view to the new control instance through a view reallocation operation; S325. When the view reallocation operation is completed, the top-level controller is notified through a service discovery mechanism to monitor new instance addresses of the remaining components that generate changes to the new control instance.

8. The method for implementing a flexible control plane for server-aware computing scenarios according to claim 5, characterized in that: The S34 includes: S341, during the dynamic partitioning and detection operation of the functionalized control plane, the control instance of each of the stateless control functions preferentially selects applicable resources in the local resource view; S342: If the control instance in the current stateless control function does not have the applicable resource, the stateless control function requests the other stateless control functions at the same level to select the applicable resource.

9. The method for implementing a flexible control plane for server-aware computing scenarios according to claim 8, characterized in that: During the detection operation of S341, a detection cache mechanism is used to record detection feedback and probes, and to mark resource restriction information of both parties in the detection operation, so as to establish a control function resource cache table.

10. The elastic control plane implementation system for server-unaware computing scenarios is characterized by: The system comprises: The plane functionalization and instance scheduling module is used to functionalize the server imperceptible computing control plane, split the server imperceptible computing control plane into stateless control functions, so as to perform automatic scaling actions of the server imperceptible computing control plane, and in the bottom functionalized control plane, use the stateless control function to perform user interaction operations and execute request distribution operations of the server imperceptible computing control plane to perform instance scheduling, wherein the server imperceptible computing control plane includes: a top-level control plane and a bottom-level functionalized control plane; A functionalized plane design module, used for collaboratively designing a functionalized control plane based on the stateless function and dynamic partitioning strategy in the server-unaware computing control plane, wherein the functionalized plane design module is connected to the plane functionalization and instance scheduling module; The resource view partitioning module is used to design the control plane expansion and contraction strategy. Based on the converged hash, the control component resource view allocation strategy is designed to handle different demands when the load fluctuates, and to balance the decision speed and decision quality. The resource view is partitioned for the control instances corresponding to different stateless control functions. The resource view partitioning module is connected to the functional plane design module.

Images