A server-unaware computing adaptive resource scheduling method, system and computer equipment

Abstract

The server-unaware computing adaptive resource scheduling method, system and computer equipment provided in the present application obtain workload log data of the server-unaware computing function, obtain characteristic indicator analysis of the workload log data, determine the threshold range θ of the function based on the characteristic indicator, determine the type of function, and select a corresponding scheduling strategy based on the threshold range θ of the function. The server-unaware computing adaptive resource scheduling method, system and computer equipment provided in the present application analyze the workload in the actual generation environment, and set thresholds through a multi-target threshold mechanism based on 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, providing a more optimized solution for server-unaware function resource scheduling, and thus improving the performance of server-unaware applications under complex function call modes.

Description

Technical Field

[0001] The present application relates to the field of cloud computing technology, and in particular to a server-unaware computing adaptive resource scheduling method, system, and computer equipment. Background Art

[0002] With the rapid development of information technology, humanity has entered an era of intelligent interconnection and interaction of all things. Cloud computing, as a new generation of information technology service model, is constantly evolving and playing an increasingly important role in society. Cloud native is a newly emerging technology that is essentially a collection of cloud computing technologies and management methods. Cloud native applications, as the name suggests, are applications suitable for running in cloud environments such as private, public, or hybrid clouds. Compared with traditional software applications, cloud native applications offer many advantages, such as low coupling, strong scalability, and high availability. Serverless computing is an advanced stage in the development of cloud native and a new trend in cloud computing. In recent years, the rapid development of serverless computing has brought significant advantages in reducing costs, reducing latency, improving scalability, and eliminating server-side management. Cloud service providers manage services in a transparent and auto-scaling manner, eliminating the need for developers to manage underlying service configurations and enabling rapid application deployment through functions. This has led to its widespread popularity and rapid growth.

[0003] As server-aware technology is widely adopted in fields such as the internet, transportation, and education, a wide variety of applications are being deployed on server-aware computing platforms. These applications can be broadly categorized as CPU-intensive, I / O-intensive, memory-intensive, and network-intensive. Different types of applications have varying resource and network requirements, posing significant challenges to server-aware computing performance. Optimizing the performance of server-aware application functions is key to improving server-aware computing utilization. Numerous factors influence the performance of server-aware application functions, including cold starts, performance isolation, scheduling policies, and cost.

[0004] In "Lambada: Interactive data analytics on cold data using serverless cloud infrastructure," Müller et al. designed a performance framework for serverless functions. Bayesian optimization was used to statistically learn the relationship between runtime cost and configuration of serverless functions and select the optimal configuration for them. In "Centralized core-granular scheduling for serverless functions," Kaffes et al. proposed a cluster-level, centralized, core-granular Serverless function scheduler. By maintaining a global view of cluster resources, it eliminates queue imbalances and reduces core-granular interference. In "SAND: Towards High-Performance Serverless Computing," Mohan et al. proposed that cold start startup time can be significantly reduced by pre-allocating network interfaces and then binding them to new function containers. SOCK optimizes the loading of Python functions in OpenLambda by intelligently caching library sets and using lightweight isolation mechanisms for functions.

[0005] The existing server-aware computing resource scheduling and management strategies are as follows:

[0006] (1) Prediction-based resource scheduling; (2) Optimizing memory and cost, and making a compromise between energy consumption and performance; (3) Reducing the impact of cold starts on the performance of server-unaware applications. The above strategies all consider different aspects, but a single scheduling strategy cannot properly solve the problem of running multiple types of applications in parallel under the complex dynamic function call model in the server-unaware computing platform. Summary of the Invention

[0007] In view of this, it is necessary to provide a server-unaware computing adaptive resource scheduling method, system and computer equipment that can improve the resource utilization of server-unaware computing functions and improve the performance of server-unaware applications under complex function call modes to address the defects in the existing technology.

[0008] To solve the above problems, this application adopts the following technical solutions:

[0009] One of the purposes of this application is to provide a server-unaware computing adaptive resource scheduling method, comprising the following steps:

[0010] Obtain workload log data of server-aware computing functions;

[0011] Obtain characteristic indicator analysis of the workload log data;

[0012] Determine the threshold range θ of the function according to the characteristic index and determine the type of the function;

[0013] A corresponding scheduling strategy is selected according to the threshold range θ of the function.

[0014] In some embodiments, the step of obtaining workload log data of server-unaware computing functions specifically includes: obtaining workload log data of server-unaware computing functions open sourced by the cloud service provider, wherein the workload log data includes request volume and function size.

[0015] In some embodiments, the step of obtaining characteristic indicator analysis of the workload log data specifically includes the following steps:

[0016] The Kmean clustering algorithm is used to perform cluster analysis on the workload based on the two characteristics of the request volume and the function size. Based on the clustering results, performance characterization modeling and portrait analysis are performed on the workload in the cloud provider's actual production environment to obtain corresponding characteristic indicators, and similarity analysis is performed on newly emerging function types. The characteristic indicators include CPU utilization, memory utilization and network bandwidth utilization.

[0017] In some embodiments, the step of determining the threshold range θ of the function according to the characteristic index and determining the type of the function specifically includes the following steps:

[0018] According to the characteristic index, the threshold range θ of the function is determined through the multi-objective threshold mechanism, the type of the function is determined, and the score of each node function is calculated through its algorithm. The formula is as follows:

[0019] θ(x)=(1-|C(x)-M(x)|)×10

[0020] Where M(x) is the memory utilization of node function x, C(x) is the CPU utilization of node function x, and θ(x) represents the score of node function x, which ranges from [0,10].

[0021] In some embodiments, the step of selecting a corresponding scheduling strategy according to the threshold range θ of the function specifically includes the following steps:

[0022] When θ(x) is less than or equal to 5, the Fixed CPU policy is selected, which allocates a CPU share proportional to the selected memory amount to each function instance. The formula is as follows:

[0023]

[0024] Where: c i The CPU share allocated to function application i, m i The memory size of function application i, cpu total is the sum of the CPU units of all machines in the server-unaware platform, mem total The total memory of all machines in the server-unaware platform;

[0025] The total CPU and memory allocated to the application function does not exceed the total available capacity, that is,

[0026]

[0027] ∑ i m i ≤mem total ;

[0028] When θ(x) is greater than 5, the Decoupled strategy is selected, which decouples the CPU and memory allocation of each function instance. The decoupled function will have a larger search space and adopt the time series prediction SVR algorithm to allocate resources to function instances based on the resources consumed by the previous function through time series prediction.

[0029] In some embodiments, the step of allocating resources to function instances by using a time series prediction SVR algorithm based on resources consumed by previous functions through time series prediction specifically includes the following steps:

[0030] According to the memory and network resources consumed by the function in the previous stage as input data, the input data is preprocessed, where the SVR algorithm can be expressed as:

[0031]

[0032] Where: k(x,x i ) is the kernel function, and the Lagrangian operator is introduced α i , select the polynomial kernel function as the kernel function of SVR, which is expressed as:

[0033] K(x i ,x j )=(x i q ·x j ) q

[0034] Where: q indicates that the kernel function is a q-order polynomial kernel function;

[0035] After determining the kernel function parameters, the PSO algorithm is used to optimize the kernel function to obtain the optimal parameters. The PSO algorithm formula is as follows:

[0036] v i =v i +c1×rand()×(pbest i -x i )+c2×rand()×(gbest i -x i )

[0037] x i =x i +v i

[0038] Where i = 1, 2, ..., N, N is the number of particles in the population, x i is the current position of the particle, v i is the velocity of the particle, c1 and c2 are learning factors;

[0039] The SVR model is used to predict the selected resource data, and the prediction results are used for resource scheduling in the next stage.

[0040] The second object of this application is to provide a server-unaware computing adaptive resource scheduling system, comprising:

[0041] Data collection unit: used to obtain workload log data of the server's non-aware computing function;

[0042] Analysis unit: used for obtaining characteristic index analysis of the workload log data;

[0043] A determination unit: configured to determine a threshold range θ of the function according to the characteristic index and determine the type of the function;

[0044] The selection unit is configured to select a corresponding scheduling strategy according to a threshold range θ of the function.

[0045] The third object of the present application is to provide a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, any of the methods described above is implemented.

[0046] This application adopts the above technical solution, and its beneficial effects are as follows:

[0047] The server-unaware computing adaptive resource scheduling method, system and computer equipment provided in the present application obtain workload log data of the server-unaware computing function, obtain characteristic indicator analysis of the workload log data, determine the threshold range θ of the function based on the characteristic indicator, determine the type of function, and select a corresponding scheduling strategy based on the threshold range θ of the function. The server-unaware computing adaptive resource scheduling method, system and computer equipment provided in the present application analyze the workload in the actual generation environment, and set thresholds through a multi-target threshold mechanism based on 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, providing a more optimized solution for server-unaware function resource scheduling, and thus improving the performance of server-unaware applications under complex function call modes. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments of the present application or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0049] Figure 1 This is a flowchart of the steps of the server-unaware computing adaptive resource scheduling method provided in Example 1 of the present application.

[0050] Figure 2 This is a flowchart of the steps for allocating resources to function instances through time series prediction using the time series prediction SVR algorithm provided in Example 1 of the present application based on the resources consumed by previous functions.

[0051] Figure 3 A structural diagram of the server-aware computing adaptive resource scheduling system provided in Example 2 of the present application.

[0052] Figure 4 A schematic diagram of the structure of the computer device provided in Example 3 of the present application. DETAILED DESCRIPTION

[0053] The following describes in detail embodiments of the present application, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0054] In the description of this application, it should be understood that the terms "upper", "lower", "horizontal", "inside", "outside", etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing this application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on this application.

[0055] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. Throughout the description of this application, "plurality" means two or more, unless otherwise specifically defined.

[0056] In order to make the purpose, technical solutions and advantages of this application more clear, this application is further described in detail below with reference to the accompanying drawings and embodiments.

[0057] Example 1

[0058] See also Figure 1 , which is a flowchart of a method for adaptive resource scheduling of server-unaware computing provided in Example 1, including the following steps:

[0059] Step S110: Obtain workload log data of the server's non-aware computing function.

[0060] In some embodiments, the step of obtaining workload log data of server-unaware computing functions specifically includes: obtaining workload log data of server-unaware computing functions open sourced by the cloud service provider, wherein the workload log data includes request volume and function size.

[0061] Step S120: Obtain characteristic indicator analysis of the workload log data.

[0062] In some embodiments, the step of obtaining characteristic indicator analysis of the workload log data specifically includes the following steps:

[0063] The Kmean clustering algorithm is used to perform cluster analysis on the workload based on the two characteristics of the request volume and the function size. Based on the clustering results, performance characterization modeling and portrait analysis are performed on the workload in the cloud provider's actual production environment to obtain corresponding characteristic indicators, and similarity analysis is performed on newly emerging function types. The characteristic indicators include CPU utilization, memory utilization and network bandwidth utilization.

[0064] Step S130: determining the threshold range θ of the function according to the characteristic index and determining the type of the function.

[0065] In some embodiments, the step of determining the threshold range θ of the function according to the characteristic index and determining the type of the function specifically includes the following steps:

[0066] According to the characteristic index, the threshold range θ of the function is determined through the multi-objective threshold mechanism, the type of the function is determined, and the score of each node function is calculated through its algorithm. The formula is as follows:

[0067] θ(x)=(1-|C(x)-M(x)|)×10

[0068] Where M(x) is the memory utilization of node function x, C(x) is the CPU utilization of node function x, and θ(x) represents the score of node function x, which ranges from [0,10].

[0069] Step S140: Select a corresponding scheduling strategy according to the threshold range θ of the function.

[0070] In some embodiments, the step of selecting a corresponding scheduling strategy according to the threshold range θ of the function specifically includes the following steps:

[0071] When θ(x) is less than or equal to 5, the Fixed CPU policy is selected, which allocates a CPU share proportional to the selected memory amount to each function instance. The formula is as follows:

[0072]

[0073] Where: c i The CPU share allocated to function application i, m i The memory size of function application i, cpu total is the sum of the CPU units of all machines in the server-unaware platform, mem total The total memory of all machines in the server-unaware platform;

[0074] The total CPU and memory allocated to the application function does not exceed the total available capacity, that is,

[0075]

[0076] ∑ i m i ≤mem total ;

[0077] When θ(x) is greater than 5, the Decoupled strategy is selected, which decouples the CPU and memory allocation of each function instance. The decoupled function will have a larger search space and adopt the time series prediction SVR algorithm to allocate resources to function instances based on the resources consumed by the previous function through time series prediction.

[0078] See also Figure 2 In the step of allocating resources to function instances by using the time series prediction SVR algorithm based on the resources consumed by the previous function, the following steps are specifically included:

[0079] Step S141: pre-process the input data based on the memory and network resources consumed by the function in the previous stage. The SVR algorithm can be expressed as:

[0080]

[0081] Where: k(x,x i ) is the kernel function, and the Lagrangian operator is introduced α i , select the polynomial kernel function as the kernel function of SVR, which is expressed as:

[0082]

[0083] Where: q indicates that the kernel function is a q-order polynomial kernel function;

[0084] Step S142: After the kernel function parameters are determined, the PSO algorithm is used to optimize the kernel function to obtain the optimal parameters. The PSO algorithm formula is as follows:

[0085] v i =v i +c1×rand()×(pbest i -x i )+c2×rand()×(gbest i -x i )

[0086] x i =x i +v i

[0087] Where i = 1, 2, ..., N, N is the number of particles in the population, x i is the current position of the particle, v i is the velocity of the particle, c1 and c2 are learning factors;

[0088] Step S143: Use the SVR model to predict the selected resource data, and the obtained prediction results are used for resource scheduling in the next stage.

[0089] The server-unaware computing adaptive resource scheduling method provided in the above-mentioned embodiment 1 of the present application analyzes the workload in the actual generation environment, and sets thresholds through a multi-target threshold mechanism based on 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, and providing a more optimized solution for server-unaware function resource scheduling, thereby improving the performance of server-unaware applications under complex function call modes.

[0090] Example 2

[0091] See also Figure 3 , which is a structural diagram of a server-unaware computing adaptive resource scheduling system provided in this embodiment 2, including a data collection unit 110: used to obtain workload log data of a server-unaware computing function; an analysis unit 120: used to obtain characteristic indicator analysis of the workload log data; a determination unit 130: used to determine the threshold range θ of the function according to the characteristic indicator and determine the type of the function; and a selection unit 140, used to select a corresponding scheduling strategy according to the threshold range θ of the function.

[0092] The server-unaware computing adaptive resource scheduling system provided in Example 2 of the present application has a detailed working method that can be referred to in Example 1 and will not be repeated here.

[0093] The server-unaware computing adaptive resource scheduling method provided in the above-mentioned embodiment 1 of the present application analyzes the workload in the actual generation environment, and sets thresholds through a multi-target threshold mechanism based on 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, and providing a more optimized solution for server-unaware function resource scheduling, thereby improving the performance of server-unaware applications under complex function call modes.

[0094] Example 3

[0095] See also Figure 4 , is a schematic diagram of the computer device structure of an embodiment of the present application. The computer device 50 includes a processor 51 and a memory 52 coupled to the processor 51.

[0096] The memory 52 stores program instructions for implementing a server-unaware computing adaptive resource scheduling method.

[0097] The processor 51 is used to execute program instructions stored in the memory 52 to implement a server-unaware computing adaptive resource scheduling method.

[0098] The processor 51 may also be referred to as a CPU (Central Processing Unit). The processor 51 may be an integrated circuit chip having signal processing capabilities. The processor 51 may also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component. The general-purpose processor may be a microprocessor or any conventional processor.

[0099] It can be understood that the various technical features of the above-described embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the various technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0100] The above are merely preferred embodiments of the present application and only specifically describe the technical principles of the present application. These descriptions are intended only to explain the principles of the present application and should not be construed in any way as limiting the scope of protection of the present application. Based on the explanations herein, any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present application, as well as other specific implementations of the present application that can be conceived by those skilled in the art without inventive effort, shall be included within the scope of protection of the present application.

Claims

1. A server-unaware computing adaptive resource scheduling method, characterized in that: The steps include: Obtain workload log data of server-aware computing functions; Obtain characteristic indicator analysis of the workload log data; Determining the threshold range θ of the function according to the characteristic index and determining the type of the function; specifically comprising the following steps: According to the characteristic index, the threshold range θ of the function is determined through the multi-objective threshold mechanism, the type of the function is determined, and the score of each node function is calculated through its algorithm. The formula is as follows: θ(x)=(1-|C(x)-M(x)|)×10 Where M(x) is the memory utilization of node function x, C(x) is the CPU utilization of node function x, and θ(x) represents the score of node function x, which ranges from [0, 10]. Selecting a corresponding scheduling strategy according to the threshold range θ of the function; specifically comprising the following steps: When θ(x) is less than or equal to 5, the Fixed CPU policy is selected, which allocates a CPU share proportional to the selected memory amount to each function instance. The formula is as follows: Where: c i The CPU share allocated to function application i, m i The memory size of function application i, cpu total is the sum of the CPU units of all machines in the server-unaware platform, mem total The total memory of all machines in the server-unaware platform; The total CPU and memory allocated to the application function does not exceed the total available capacity, that is, ∑ i m i ≤mem total ; When θ(x) is greater than 5, the Decoupled strategy is selected, which decouples the CPU and memory allocation of each function instance. The decoupled function will have a larger search space and adopt the time series prediction SVR algorithm to allocate resources to function instances based on the resources consumed by the previous function through time series prediction.

2. The server-unaware computing adaptive resource scheduling method according to claim 1, characterized in that: The step of obtaining workload log data of server-unaware computing functions specifically includes: obtaining workload log data of server-unaware computing functions open sourced by the cloud service provider, wherein the workload log data includes request volume and function size.

3. The server-unaware computing adaptive resource scheduling method according to claim 2, characterized in that: The step of obtaining characteristic indicator analysis of the workload log data specifically includes the following steps: The Kmean clustering algorithm is used to perform cluster analysis on the workload based on the two characteristics of the request volume and the function size. Based on the clustering results, performance characterization modeling and portrait analysis are performed on the workload in the cloud provider's actual production environment to obtain corresponding characteristic indicators, and similarity analysis is performed on newly emerging function types. The characteristic indicators include CPU utilization, memory utilization and network bandwidth utilization.

4. The server-unaware computing adaptive resource scheduling method according to claim 3, characterized in that: In the step of using the time series prediction SVR algorithm to allocate resources to function instances based on resources consumed by previous functions through time series prediction, the following steps are specifically included: The process of using the SVR model algorithm to predict and allocate resources is as follows: According to the memory and network resources consumed by the function in the previous stage as input data, the input data is preprocessed, where the SVR algorithm can be expressed as: Where: k(x,x i ) is the kernel function, and the Lagrangian operator is introduced α i , select the polynomial kernel function as the kernel function of SVR, which is expressed as: Where: q indicates that the kernel function is a q-order polynomial kernel function; After determining the kernel function parameters, the PSO algorithm is used to optimize the kernel function to obtain the optimal parameters. The PSO algorithm formula is as follows: v i =v i +c1×rand()×(pbest i -x i )+c2×rand()×(gbest i -x i ) x i =x i +v i Where i = 1, 2, ..., N, N is the total number of particles in the population, x i is the current position of the particle, v i is the velocity of the particle, c1 and c2 are learning factors; The SVR model is used to predict the selected resource data, and the prediction results are used for resource scheduling in the next stage.

5. A server-unaware computing adaptive resource scheduling system, characterized in that: include: Data collection unit: used to obtain workload log data of the server's non-aware computing function; Analysis unit: used for obtaining characteristic index analysis of the workload log data; A determination unit is configured to determine the threshold range θ of the function based on the characteristic index and determine the type of the function; specifically, the determination unit includes: According to the characteristic index, the threshold range θ of the function is determined through the multi-objective threshold mechanism, the type of the function is determined, and the score of each node function is calculated through its algorithm. The formula is as follows: θ(x)=(1-|C(x)-M(x)|)×10 Where M(x) is the memory utilization of node function x, C(x) is the CPU utilization of node function x, and θ(x) represents the score of node function x, which ranges from [0, 10]. The selection unit is configured to select a corresponding scheduling strategy according to the threshold range θ of the function; specifically comprising: When θ(x) is less than or equal to 5, the Fixed CPU policy is selected, which allocates a CPU share proportional to the selected memory amount to each function instance. The formula is as follows: Where: c i The CPU share allocated to function application i, m i The memory size of function application i, cpu total is the sum of the CPU units of all machines in the server-unaware platform, mem total The total memory of all machines in the server-unaware platform; The total CPU and memory allocated to the application function does not exceed the total available capacity, that is, ∑ i m i ≤mem total ; When θ(x) is greater than 5, the Decoupled strategy is selected, which decouples the CPU and memory allocation of each function instance. The decoupled function will have a larger search space and adopt the time series prediction SVR algorithm to allocate resources to function instances based on the resources consumed by the previous function through time series prediction.

6. A computer device, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method according to any one of claims 1 to 4 is implemented.

Images