An efficient Zygote container planning method and system for serverless startup in edge cloud

Abstract

This invention provides an efficient Zygote container planning method and system for serverless function startup in edge clouds, relating to the field of Zygote container planning. The method comprises the following steps: S1: Constructing an edge cloud system and setting constraints based on the edge cloud system; S2: Constructing a function startup time minimization expression using the edge cloud system and constraints; S3: Obtaining an approximate optimal feasible solution to the function startup time minimization expression using the Regression-Range (RR) algorithm; S4: Obtaining the Zygote container allocation strategy in the edge cloud system using the approximate optimal feasible solution. This invention solves the problem of low preheating efficiency of Zygote containers by using the RR algorithm to solve the function startup time minimization expression and finding an approximate optimal feasible solution that satisfies the constraints through continuous iteration. A Zygote container planning strategy is then designed based on this approximate optimal feasible solution, solving the previous problem of low Zygote container preheating efficiency. Ultimately, a more efficient Zygote container planning strategy is used in serverless edge cloud systems to minimize function startup time.

Description

Technical Field

[0001] This invention relates to the field of Zygote container planning, and in particular to an efficient Zygote container planning method and system for serviceless startup of edge clouds. Background Technology

[0002] In edge cloud environments, reducing function startup time becomes crucial due to the limited resources of edge servers. This involves two key concepts in serverless computing: warm-up and warm-up. Warm-up refers to keeping containers active by periodically triggering functions, thereby reducing cold start time. During warm-up, the function is not actually executed; it is only triggered to load the function's runtime environment and resources so that it can respond to requests immediately. Warm-up, on the other hand, refers to keeping the container active for a period of time after a warm start. It is usually shorter than warm-up and has the advantage of reducing the performance overhead of frequent container starts, eliminating the need for a cold start every time. Warm-up strategies offer better performance than warm-up but require more resources.

[0003] In their paper "{SOCK}: Rapid task provisioning with {Serverless-Optimized} containers," Oakes et al. mentioned that full container warm-up consumes a lot of memory, and thus proposed Zygote container warm-up. Zygote pre-creates and caches the runtime environment of functions by creating lightweight Zygote containers to achieve fast startup. Zygote containers pre-import commonly used packages. When a function is called, a new container is derived from the pre-warmed Zygote container, and packages not originally present in the Zygote container but required by the function are imported into it. Thus, starting a function only requires importing additional packages, avoiding time-consuming initialization operations. Furthermore, since different functions may have the same dependencies, the Zygote container can act as a 'parent container' for these functions, eliminating the need for frequent creation.

[0004] However, in practical applications, the preheating efficiency cannot be optimal due to two characteristics of the Zygote container: ① Exclusivity: This means that unnecessary packages are not allowed in the container during function execution, nor are additional packages allowed in the generated container. ② Additive-only: This means that packages can only be added to the container, not deleted. This means that packages imported into the Zygote container can only exist as a subset of function dependencies; otherwise, the preheated Zygote container cannot be used.

[0005] Therefore, the following problems in traditional Zygote container warm-up are not well resolved: 1) Which packages should be pre-imported in each Zygote, 2) On which edge server should Zygote be warmed up, 3) On which edge server should each function be called based on which Zygote, which is limited by memory and computing resources, and 4) On which registry should the image be downloaded during cold start.

[0006] Research shows that the Zygote container planning problem based on the above problem can be formalized into a quadratic integer programming problem. This problem is NP-hard, and it is difficult to find an efficient algorithm to solve it. Therefore, how to perform efficient Zygote container planning in the edge cloud environment to reduce function startup time is a technical problem that urgently needs to be solved. Summary of the Invention

[0007] In view of this, the purpose of this invention is to provide an efficient Zygote container planning method for edge cloud systems that can use a more efficient Zygote container planning strategy to minimize function startup time and enable serverless function startup in edge cloud environments.

[0008] This invention provides an efficient Zygote container planning method for serviceless startup in edge clouds, including:

[0009] S1: Construct an edge cloud system and set constraints based on the edge cloud system;

[0010] S2: Construct a function startup time expression by using edge cloud systems and constraint relationships;

[0011] S3: Obtain the approximate optimal feasible solution of the function startup time expression by calculating the RR algorithm;

[0012] S4: Obtain the allocation strategy of Zygote containers in the edge cloud system through the approximate optimal feasible solution.

[0013] Preferred:

[0014] The edge cloud system consists of: N servers and R registries; the servers are numbered n, n∈N; the registries are numbered r, r∈R;

[0015] The computing resource capacity of server n is C. n The network bandwidth between registry r and server n is B. nr The server n uses M memory to generate Zygote containers. n The imported package is p, the set of packages is P, and the import time of package p is t. p The size of package p is S pThe called function is f, which can be viewed as a raw container plus several dependent packages, where the size of the raw container is B; the set of called functions is F, and the request frequency of the called function f is λ. f The startup time of the called function f is t. fn The Zygote container generated on server n is called z, and the set of Zygote containers on server n is called Z.

[0016] Preferred:

[0017] The constraints include:

[0018] Define a binary variable s nz ∈{0,1} represents whether a Zygote container z is generated on server n, defined as a binary variable. Indicates whether the called function f uses the image stored in the registry r for cold boot;

[0019] Define binary variables Indicates whether package p is pre-imported into the Zygote container z on server n, setting the first constraint:

[0020]

[0021] Due to server memory limitations, a second constraint is set:

[0022]

[0023] Define a binary variable d fn ∈{0,1} indicates whether the called function f is invoked on server n, setting a third constraint:

[0024]

[0025] Define binary variables This indicates whether the called function f utilizes the Zygote container z on the same server n, setting a fourth constraint:

[0026]

[0027] And setting the fifth constraint:

[0028]

[0029] Define a binary variable β fp ∈{0,1} indicates whether the called function f needs to import package p. The sixth constraint is set as follows:

[0030]

[0031] The computational resources of all functions called on server n cannot exceed the computational resource capacity it possesses. A seventh constraint is set:

[0032]

[0033] Among them, C f The computational resources of the called function;

[0034] If the called function f is not invoked on server n, then the corresponding image does not need to be downloaded to server n. Set the eighth constraint:

[0035]

[0036] If calling the called function f on server n requires a cold start, this means there is no suitable Zygote container available for the called function f to utilize. Therefore, the ninth constraint is set:

[0037]

[0038] Define binary variables variable Set the tenth constraint:

[0039]

[0040]

[0041]

[0042] Define binary variables variable Set the eleventh constraint:

[0043]

[0044]

[0045]

[0046] Preferably, the expression for minimizing function startup time is as follows:

[0047]

[0048] Among them, t z (f)=t c (f)+t w (f);

[0049] Cold start time t c The expression for (f) is:

[0050]

[0051]

[0052] Hot start time t w The expression for (f) is:

[0053]

[0054] Preferably, step S3 specifically includes:

[0055] S31: Linearize the expression for minimizing the function startup time to obtain the solution expression;

[0056] S32: Set of binary variables The binary variables in the expression are relaxed to real numbers between [0,1]. The theoretical optimal solution is obtained by solving the real numbers in the expression using a linear programming solver.

[0057] S33: Set and initialize the set of variables Each variable in the variable set is rounded down to the nearest integer in the range {0, 1};

[0058] S34: For all z∈Z, n∈N, for Reassign value, The probability of taking the value 1 is For all f∈F, n∈N, Reassign value, The probability of taking the value 1 is

[0059] S35: For all p∈P, n∈N, z∈Z, if If it is 1, then it is a pair Reassign value, The probability of taking the value 1 is otherwise The value remains unchanged; if and If both are 1, then it is a pair Reassign value, The probability of taking the value 1 is otherwise The value remains unchanged;

[0060] S36: For all f∈F, n∈N, z∈Z, if and If both are 1, then it is a pair Reassign value, The probability of taking the value 1 is otherwise The value remains unchanged; for all f∈F, n∈N, z∈Z, p∈P, if and If both are 1, then it is a pair Reassign value, The probability of taking the value 1 is otherwise The value remains unchanged;

[0061] S37: For all f∈F, n∈N, r∈R, if If it is 1, then it is a pair Reassign value, The probability of taking the value 1 is otherwise The value remains unchanged;

[0062] S38: Repeat steps S34-S37 until the set of variables satisfies the constraints. The set of variables that satisfies the constraints is taken as the approximate optimal feasible solution.

[0063] Preferably, the solution expression is as follows:

[0064]

[0065] A storage medium storing instructions and data for implementing the efficient Zygote container planning method for serverless function startup in edge cloud environments.

[0066] A high-efficiency Zygote container planning system for edge cloud serverless startup includes: a processor and a storage medium; the processor loads and executes instructions and data in the storage medium to implement the high-efficiency Zygote container planning method for edge cloud serverless startup.

[0067] The present invention has the following beneficial effects:

[0068] An edge cloud system with N servers and R registries is constructed, and constraints are established for the system. Based on these constraints, a function startup time minimization expression is derived, enabling a formulaic representation of the startup time. This minimization expression is then linearized and solved using the Regression-Raphson (RR) algorithm. Through iterative processing, an approximately optimal feasible solution satisfying the constraints is found. Based on this approximately optimal feasible solution, a Zygote container planning strategy is designed, addressing the previous issue of low Zygote container warm-up efficiency. Finally, in the edge cloud system, a more efficient Zygote container planning strategy is used to minimize the function startup time. Attached Figure Description

[0069] Figure 1 This is a flowchart of a method according to an embodiment of the present invention;

[0070] Figure 2This is a first performance comparison chart of the RR algorithm with other algorithms;

[0071] Figure 3 This is a second performance comparison chart of the RR algorithm and other algorithms;

[0072] Figure 4 This is a third performance comparison chart of the RR algorithm and other algorithms;

[0073] Figure 5 A system architecture diagram for planning efficient Zygote containers for serviceless startup in edge cloud;

[0074] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0075] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0076] Reference Figure 1 This invention provides an efficient Zygote container planning method for service-free startup in edge clouds, comprising:

[0077] S1: Construct an edge cloud system and set constraints based on the edge cloud system;

[0078] S2: Construct a function startup time expression by using edge cloud systems and constraint relationships;

[0079] S3: Obtain the approximate optimal feasible solution of the function startup time expression by calculating the RR algorithm;

[0080] S4: Obtain the allocation strategy of Zygote containers in the edge cloud system through the approximate optimal feasible solution.

[0081] Furthermore, this invention establishes an edge cloud system consisting of a set of N heterogeneous edge servers with available resources and a set of R image registry entries, specifically defined as follows:

[0082] The edge cloud system consists of: N servers and R registries; the servers are numbered n, n∈N; the registries are numbered r, r∈R;

[0083] The computing resource capacity of server n is C. n The network bandwidth between registry r and server n is B. nr The server n uses M memory to generate Zygote containers. n The imported package is p, the set of packages is P, and the import time of package p is t. p The size of package p is S pThe called function is f, which can be viewed as a raw container plus several dependent packages, where the size of the raw container is B; the set of called functions is F, and the request frequency of the called function f is λ. f The startup time of the called function f is t. fn The Zygote container generated on server n is called z, and the set of Zygote containers on server n is called Z.

[0084] Specifically, for the registry r, it is assumed that it can provide various container images for different functions; memory M n This is used to limit the number of Zygote containers that can be pre-warmed on each server. Since Zygote may pre-import some packages, but these may not fully meet the requirements of the calling functions, some additional packages need to be imported before startup. Different packages of different sizes require different import times, therefore, the import time for package p needs to be defined as t. p .

[0085] Furthermore, based on the above definition of an edge cloud system, a series of constraints can be derived, which ultimately determine the Zygote container allocation strategy:

[0086] The constraints include:

[0087] Define a binary variable s nz ∈{0,1} represents whether a Zygote container z is generated on server n, defined as a binary variable. Indicates whether the called function f uses the image stored in the registry r for cold boot;

[0088] Define binary variables Indicates whether package p is pre-imported into the Zygote container z on server n, setting the first constraint:

[0089]

[0090] Due to server memory limitations, a second constraint is set:

[0091]

[0092] Define a binary variable d fn ∈{0,1} indicates whether the called function f is invoked on server n, setting a third constraint:

[0093]

[0094] Define binary variables This indicates whether the called function f utilizes the Zygote container z on the same server n, setting a fourth constraint:

[0095]

[0096] And setting the fifth constraint:

[0097]

[0098] Define a binary variable β fp ∈{0,1} indicates whether the called function f needs to import package p. The sixth constraint is set as follows:

[0099]

[0100] The computational resources of all functions called on server n cannot exceed the computational resource capacity it possesses. A seventh constraint is set:

[0101]

[0102] Among them, C f The computational resources of the called function;

[0103] If the called function f is not invoked on server n, then the corresponding image does not need to be downloaded to server n. Set the eighth constraint:

[0104]

[0105] If calling the called function f on server n requires a cold start, this means there is no suitable Zygote container available for the called function f to utilize. Therefore, the ninth constraint is set:

[0106]

[0107] Define binary variables variable Set the tenth constraint:

[0108]

[0109] Define binary variables variable Set the eleventh constraint:

[0110]

[0111] Furthermore, the expression for minimizing function startup time is as follows:

[0112]

[0113] Among them, t z (f)=t c (f)+t w (f);

[0114] Cold start time t c The expression for (f) is:

[0115]

[0116]

[0117] Specifically, we will only consider the import time of package p as the function's warm start time if both function f and container z are deployed on server n and container z has not imported the package p required by function f. Otherwise, the function's warm start will not take any time.

[0118] Hot start time t w The expression for (f) is:

[0119]

[0120] Furthermore, based on the complexity of the problem-solving process, this invention proposes the RR (Zygote Programming Based on Random Rounding) algorithm to obtain an approximate optimal feasible solution to the expression.

[0121] The RR algorithm strategy is as follows:

[0122] 1) First, relax all binary variables defined in the system to real numbers between [0,1], which makes them more suitable for solving linear programming problems.

[0123] 2) Use a linear programming solver to solve formula (16) and find the optimal solution of the slack variables in 1) as the best decision value.

[0124] 3) Use the real solution obtained in 2) as the probability value of the corresponding binary variable sign, and make the originally defined binary variable randomly rounded in [0,1] with this probability value. And check the constraint relationship of the final result each time.

[0125] 4) The above steps may yield a solution that satisfies the minimum time but not the constraints. Therefore, multiple iterations are required until a solution that minimizes the function startup time while satisfying the constraints is found, which is then considered an approximately optimal feasible solution.

[0126] Step S3 is as follows:

[0127] S31: Linearize the expression for minimizing the function startup time to obtain the solution expression;

[0128] S32: Set of binary variables The binary variables in the expression are relaxed to real numbers between [0,1]. The theoretical optimal solution is obtained by solving the real numbers in the expression using a linear programming solver.

[0129] S33: Set and initialize the set of variables Each variable in the variable set is rounded down to the nearest integer in the range {0, 1};

[0130] S34: For all z∈Z, n∈N, for Reassign value, The probability of taking the value 1 is For all f∈F, n∈N, Reassign value, The probability of taking the value 1 is

[0131] S35: For all p∈P, n∈N, z∈Z, if If it is 1, then it is a pair Reassign value, The probability of taking the value 1 is otherwise The value remains unchanged; if and If both are 1, then it is a pair Reassign value, The probability of taking the value 1 is otherwise The value remains unchanged;

[0132] S36: For all f∈F, n∈N, z∈Z, if and If both are 1, then it is a pair Reassign value, The probability of taking the value 1 is otherwise The value remains unchanged; for all f∈F, n∈N, z∈Z, p∈P, if and If both are 1, then it is a pair Reassign value, The probability of taking the value 1 is otherwise The value remains unchanged;

[0133] S37: For all f∈F, n∈N, r∈R, if If it is 1, then it is a pair Reassign value, The probability of taking the value 1 is otherwise The value remains unchanged;

[0134] S38: Repeat steps S34-S37 until the set of variables satisfies the constraints. The set of variables that satisfies the constraints is taken as the approximate optimal feasible solution.

[0135] Furthermore, the solution to the expression is as follows:

[0136]

[0137] Furthermore, step S4 obtains the allocation strategy of Zygote containers in the edge cloud system through an approximate optimal feasible solution, namely, which packages should be pre-imported in each Zygote, which edge server should be warmed up for Zygote, and which server each function should be called on based on which Zygote.

[0138] Results after implementation:

[0139] In this embodiment, an environment consisting of 10 edge servers and 5 image registries is considered. The computing resources and memory capacity of the edge servers are set to 2.0 GHz and 150 MB (average), respectively. The basic container used to generate Zygote is an Alpine Linux container with BusyBox pre-installed, requiring 25 MB of memory. Each edge server can obtain a container image from the image registry, the average network bandwidth between the edge server and the image registry is set to 500 Mbps, and 10 serverless functions are built.

[0140] Under the same conditions, it is compared with the OPT algorithm (the optimal solution obtained by solving mathematical programming formulas), the SOCK algorithm (greedy strategy scheduling and allocation of Zygote container resources), and the HHP algorithm (warming up the function with the highest request call rate).

[0141] according to Figure 2 , Figure 3 , Figure 4 It can be seen that in comparisons using different memory capacities, different computing resource capabilities, and different Zipf distributions as variables, the RR algorithm achieves faster function startup than the SOCK and HHP algorithms. Overall, compared to the SOCK and HHP algorithms, the RR algorithm can significantly reduce the overall startup time by an average of 21.2% and 30.1%, respectively.

[0142] The efficient Zygote container planning method proposed in this invention for serviceless function startup in edge cloud achieves a better reduction in function startup time compared to current planning algorithms.

[0143] A storage medium 403 stores instructions and data for implementing the efficient Zygote container planning method for serverless function startup in edge cloud environments.

[0144] A high-efficiency Zygote container planning system 401 for edge cloud serverless startup includes: a processor 402 and a storage medium 403; the processor 402 loads and executes instructions and data in the storage medium 403 to implement the high-efficiency Zygote container planning method for edge cloud serverless startup.

[0145] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0146] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments. In the unit claims listing several devices, several of these devices may be embodied by the same hardware item. The use of the terms first, second, and third, etc., does not indicate any order and can be interpreted as identifiers.

[0147] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. An efficient Zygote container planning method for serverless startup in edge cloud, characterized in that, include: S1: Construct an edge cloud system and set constraints based on the edge cloud system; S2: Construct a function startup time expression by using edge cloud systems and constraint relationships; S3: Obtain the approximate optimal feasible solution for minimizing the function startup time expression by calculating the RR algorithm; S4: Obtain the allocation strategy of Zygote containers in the edge cloud system through the approximate optimal feasible solution; The constraints include: Define a binary variable s nz ∈{0,1} represents whether a Zygote container z is generated on server n, defined as a binary variable. Indicates whether the called function f uses the image stored in the registry r for cold boot; Define binary variables Indicates whether package p is pre-imported into the Zygote container z on server n, setting the first constraint: Due to server memory limitations, a second constraint is set: Define a binary variable d fn ∈{0,1} indicates whether the called function f is invoked on server n, setting a third constraint: Define binary variables This indicates whether the called function f utilizes the Zygote container z on the same server n, setting a fourth constraint: And setting the fifth constraint: Define a binary variable β fp ∈{0,1} indicates whether the called function f needs to import package p. The sixth constraint is set as follows: The computational resources of all functions called on server n cannot exceed the computational resource capacity it possesses. A seventh constraint is set: Among them, C f The computational resources of the called function; If the called function f is not invoked on server n, then the corresponding image does not need to be downloaded to server n. Set the eighth constraint: If calling the called function f on server n requires a cold start, this means there is no suitable Zygote container available for the called function f to utilize. Therefore, the ninth constraint is set: Define binary variables variable Set the tenth constraint: Define binary variables variable Set the eleventh constraint:

2. The efficient Zygote container planning method for serviceless startup in edge cloud according to claim 1, characterized in that: The edge cloud system consists of: N servers and R registries; the servers are numbered n, n∈N; the registries are numbered r, r∈R; The computing resource capacity of server n is C. n The network bandwidth between registry r and server n is B. nr The server n uses M memory to generate Zygote containers. n The imported package is p, the set of packages is P, and the import time of package p is t. p The size of package p is S p The called function is f, which can be viewed as a raw container plus several dependent packages, where the size of the raw container is B; the set of called functions is F, and the request frequency of the called function f is λ. f The startup time of the called function f is t. fn The Zygote container generated on server n is called z, and the set of Zygote containers on server n is called Z.

3. The efficient Zygote container planning method for serviceless startup in edge cloud according to claim 1, characterized in that, The expression for minimizing function startup time is as follows: Among them, t z (f)=t c (f)+t w (f); Cold start time t c The expression for (f) is: Hot start time t w The expression for (f) is:

4. The efficient Zygote container planning method for serviceless startup in edge cloud according to claim 1, characterized in that, Step S3 is as follows: S31: Linearize the expression for minimizing the function startup time to obtain the solution expression; S32: Set of binary variables The binary variables in the expression are relaxed to real numbers between [0,1]. The theoretical optimal solution is obtained by solving the real numbers in the expression using a linear programming solver. S33: Set and initialize the set of variables Each variable in the variable set is rounded down to the nearest integer in the range {0, 1}; S34: For all z∈Z, n∈N, for Reassign value, The probability of taking the value 1 is For all f∈F, n∈N, Reassign value, The probability of taking the value 1 is S35: For all p∈P, n∈N, z∈Z, if If it is 1, then it is a pair Reassign value, The probability of taking the value 1 is otherwise The value remains unchanged; if and If both are 1, then it is a pair Reassign value, The probability of taking the value 1 is otherwise The value remains unchanged; S36: For all f∈F, n∈N, z∈Z, if and If both are 1, then it is a pair Reassign value, The probability of taking the value 1 is otherwise The value remains unchanged; for all f∈F, n∈N, z∈Z, p∈P, if and If both are 1, then it is a pair Reassign value, The probability of taking the value 1 is otherwise The value remains unchanged; S37: For all f∈F, n∈N, r∈R, if If it is 1, then it is a pair Reassign value, The probability of taking the value 1 is otherwise The value remains unchanged; S38: Repeat steps S34-S37 until the set of variables satisfies the constraints. The set of variables that satisfies the constraints is taken as the approximate optimal feasible solution.

5. The efficient Zygote container planning method for serviceless startup in edge cloud according to claim 4, characterized in that, The specific solution to the expression is as follows:

6. A storage medium, characterized in that: The storage medium stores instructions and data to implement the efficient Zygote container planning method for serverless startup of edge cloud as described in any one of claims 1 to 5.

7. A high-efficiency Zygote container planning system for service-free startup in edge clouds, characterized in that: include: A processor and a storage medium; the processor loads and executes instructions and data in the storage medium to implement the efficient Zygote container planning method for serverless startup of edge cloud as described in any one of claims 1 to 5.

Images