A method and system for evaluating reliability of a computing power network considering functional self-healing

Abstract

This invention discloses a method and system for reliability assessment of computing power networks considering functional self-healing, belonging to the field of computing power network reliability assessment technology. The invention includes: acquiring operational data of a computing power network, wherein the computing power network includes a main component and an auxiliary subsystem composed of multiple auxiliary components; constructing a reliability model based on the operational data, wherein the reliability model describes the closed-loop coupling relationship between the main component and the auxiliary subsystem, wherein the failure rate of the auxiliary components increases with the increase of the number of failed auxiliary components, and the self-healing rate of the main component decreases with the increase of the number of failed auxiliary components; solving the time-domain reliability function based on the reliability model, and determining the reliability index of the computing power network. This invention, by establishing a closed-loop reliability model of the main-auxiliary coupling structure, characterizes the resource competition relationship between load-sharing failure and state-dependent self-healing, achieving accurate assessment of computing power network reliability and providing a quantitative basis for network operation and maintenance decisions.

Description

Technical Field

[0001] This invention relates to the field of computing network reliability assessment technology, and specifically to a computing network reliability assessment method and system that considers functional self-healing. Background Technology

[0002] Computing networks are critical infrastructure for integrating distributed computing resources and providing on-demand computing services; their reliability directly affects service quality and user experience. During the operation of computing networks, problems such as computing node failures, network outages, and resource contention frequently occur, creating an urgent need for network reliability assessment.

[0003] Functional self-healing is an important technical means to improve the reliability of computing networks. Functional self-healing refers to the ability of a network system to automatically complete processes such as fault detection, root cause localization, fault handling, and service restoration without human intervention when a fault or performance degradation is detected. Common self-healing mechanisms include fault isolation and degradation, dynamic resource scheduling, dynamic path reconstruction, and elastic capacity compensation.

[0004] However, traditional reliability assessment methods fail to effectively utilize the aforementioned self-healing mechanism information. They typically assume that system failure is an irreversible process, meaning that once a component fails, it permanently loses its function, or they only consider the restorative effect of external repairs, ignoring the system's inherent self-healing capabilities. This neglect leads to conservative assessment results that fail to accurately reflect the actual operating state of the computing network.

[0005] In recent years, some research has begun to focus on the self-healing characteristics of systems, proposing self-healing reliability assessment methods for single-component systems. These methods treat self-healing capability as an inherent property of the component, independent of the system state. Their self-healing mechanism is time-threshold type; that is, if the interval between two impacts exceeds a set threshold, the damage caused by the previous impact is fully recovered; otherwise, it is not. However, computing networks have a master-slave coupled structure, and their self-healing capability has a complex dependency on the system state. Furthermore, the self-healing effect is usually continuously changing, rather than a simple all-or-nothing situation. If existing assessment methods are directly applied to computing networks with a master-slave coupled structure, the coupling relationship between the master and slave components will be ignored, thus losing crucial system state information and leading to significant deviations between the assessment results and the actual situation.

[0006] Therefore, how to establish a reliability assessment model that can characterize the coupling relationship between the self-healing mechanism and the system state for the master-slave coupled structure of computing power networks is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a method and system for evaluating the reliability of computing networks that consider functional self-healing. By constructing a main-auxiliary coupled structure model consisting of a main component and k / n auxiliary components, a bidirectional coupling relationship is established between the load-sharing failure of auxiliary components and the state-dependent self-healing of the main component. Based on this coupling relationship, an analytical solution method is used to achieve quantitative evaluation of the reliability of computing networks that consider functional self-healing, thereby improving the evaluation accuracy of computing networks.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] This invention provides a reliability assessment method for computing power networks that considers functional self-healing, comprising:

[0010] S1. Obtain the operation data of the computing power network, wherein the computing power network includes a main component and an auxiliary subsystem, and the auxiliary subsystem is composed of multiple auxiliary components;

[0011] S2. Construct a reliability model based on the operational data. The reliability model describes the closed-loop coupling relationship between the main component and the auxiliary subsystem. The closed-loop coupling relationship includes:

[0012] Among the auxiliary components, the failure rate of the remaining auxiliary components increases with the increase of the number of failed auxiliary components;

[0013] The self-healing rate of the main component decreases as the number of failed auxiliary components increases;

[0014] S3. Based on the reliability model, solve the time-domain reliability function of the computing power network, and determine the reliability index of the computing power network according to the time-domain reliability function. The reliability index includes at least one of reliability, failure probability or mean time to first failure.

[0015] Furthermore, the auxiliary subsystem consists of n auxiliary components. When the number of failed auxiliary components reaches k, the entire auxiliary subsystem fails, where n≥1 and 1≤k≤n.

[0016] Furthermore, the failure rate of the auxiliary component is expressed as:

[0017]

[0018] in, As the baseline failure rate, This represents the number of failed auxiliary components. The number of failed auxiliary parts is Failure rate of remaining auxiliary components. The total workload undertaken by the auxiliary subsystem. The number of auxiliary components in the auxiliary subsystem. For about It is a monotonically increasing function.

[0019] Furthermore, the self-healing rate of the main component depends on the current operating environment of the computing network, which is described by a continuous-time Markov process, having an environment state space. and state transition rate matrix , For the number of environmental states, For from the first The environmental state transitions to the first... The transition rate of each environmental state.

[0020] Furthermore, the self-healing rate of the main component is expressed as:

[0021]

[0022] in, The self-healing rate of the main component is expressed as the rate at which the component heals in the first... Under various environmental conditions, when the number of failed auxiliary components is u, what is the repair ratio of the main component to the impact damage? ; The number of auxiliary components in the auxiliary subsystem. In the first Under what environmental conditions, regarding A monotonically increasing function. .

[0023] Furthermore, the main component is also subject to external random impacts during operation, and these random impacts employ a homogeneous Poisson process. Describe the arrival rate. ,in For time variables, Indicates the first Under each environmental state, within the time interval (0, ... The number of impacts occurring within; In the The average number of impacts per unit time under each environmental condition; the damage caused to the main component by each impact is an independent and identically distributed non-negative random variable.

[0024] Furthermore, in S3, solving the reliability function of the computing network includes:

[0025] An integral equation for the conditional reliability function is established based on the reliability model. The integral equation is then transformed into an algebraic equation by applying a double Laplace transform. After solving the algebraic equation, the time-domain reliability function is obtained through numerical inversion using the inverse Laplace transform.

[0026] Furthermore, based on the time-domain reliability function, the mean time to first failure (MTBF) of the computing network is calculated. The formula is:

[0027]

[0028] in This is the time-domain reliability function.

[0029] This invention also proposes a reliability assessment system for computing power networks that considers functional self-healing, for implementing the above method, comprising:

[0030] The data processing module is used to acquire the operating data of the computing power network, which includes a main component and an auxiliary subsystem, and the auxiliary subsystem is composed of multiple auxiliary components.

[0031] The model building module is used to build a reliability model based on the operational data. The reliability model is used to describe the closed-loop coupling relationship between the main component and the auxiliary subsystem. The closed-loop coupling relationship includes: the failure rate of the auxiliary component increases with the increase of the number of failed auxiliary components, and the self-healing rate of the main component decreases with the increase of the number of failed auxiliary components.

[0032] The solution calculation module is used to solve the reliability index of the computing power network based on the reliability model.

[0033] The output module is used to output the evaluation results.

[0034] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0035] (1) This invention establishes a closed-loop reliability model that includes both forward and reverse coupling for the main and auxiliary components of a computing power network. The number of failed auxiliary components acts in two directions simultaneously: an increase in the number of failed auxiliary components leads to an increase in the failure rate of the remaining auxiliary components, accelerating the failure of more auxiliary components; at the same time, an increase in the number of failed auxiliary components leads to a decrease in the self-healing rate of the main components, accelerating the failure of the main components. The two effects are coupled to each other through the same variable, the number of failed auxiliary components, forming a positive feedback loop, which completely preserves the key state information of the main and auxiliary coupling structure of the computing power network and overcomes the defects of traditional methods in evaluating distortion.

[0036] (2) Existing reliability assessment methods that consider self-healing mechanisms treat self-healing capability as an inherent attribute of components, independent of system state, and its self-healing trigger is discrete. This invention characterizes the resource-dependent nature of the self-healing capability of computing networks, that is, the self-healing capability depends on the abundance of available auxiliary component resources in redundant auxiliary subsystems, and the self-healing rate decreases continuously and monotonically with the increase of the number of failed auxiliary components, accurately quantifying the resource competition relationship in which auxiliary components both bear the computing load and provide self-healing support.

[0037] (3) This invention establishes a load-sharing failure model for auxiliary components and couples it with a state-dependent self-healing model through the same variable, forming a complete resource competition closed loop: auxiliary component failure—increased remaining load—increased failure rate—more auxiliary component failures—reduced self-healing resources—decreased self-healing rate—accelerated failure of main components. This closed-loop coupling relationship characterizes the inherent contradiction between load sharing and self-healing capability supply in computing power networks, providing a theoretical basis for accurately assessing the reliability of computing power networks.

[0038] (4) This invention addresses the system state determined by both the operating environment state and the number of failed auxiliary components. It employs a dual Laplace transform and numerical inversion to solve the problem. Based on the time-domain reliability function, it can output various reliability indicators: reliability, failure probability, and mean time to first failure. Compared with existing technologies, this invention differs fundamentally in five dimensions: system structure, failure logic, self-healing mechanism, state space, and evaluation output. The evaluation results can directly guide the optimization of redundant configuration, design of self-healing strategies, and operation and maintenance decisions of computing networks. Attached Figure Description

[0039] Figure 1 A flowchart illustrating the reliability assessment method for computing power networks considering self-healing functionality, as described in this invention.

[0040] Figure 2 This is a schematic diagram of the sample path for network performance degradation. Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] Example

[0043] refer to Figure 1 This embodiment provides a reliability assessment method for computing power networks that considers functional self-healing, including the following steps:

[0044] S1. Obtain the operating data of the computing power network, which includes main components and auxiliary subsystems.

[0045] The main component is the core business processing node in the computing power network. It is responsible for receiving computing tasks, allocating computing resources, performing basic calculations, and aggregating calculation results. When the main component fails, the business of the computing power network cannot be completed, and the network fails.

[0046] The auxiliary subsystem consists of multiple auxiliary components, including computing acceleration nodes such as GPU servers and edge computing nodes, which provide additional computing capabilities under the scheduling of the main component. Failure of an auxiliary component will degrade the performance of the computing network, but will not directly cause it to fail. Multiple auxiliary components are connected via a k / n redundancy structure, where n is the total number of auxiliary components and k is the failure threshold. When the number of failed auxiliary components reaches k, the entire auxiliary subsystem fails, where n ≥ 1 and 1 ≤ k ≤ n.

[0047] The operational data of the computing network includes: failure data of each component, failure threshold, topological relationship of the main-auxiliary coupling structure, historical fault records, load data, impact event data, self-healing records of the main component, status data of the auxiliary component, etc.

[0048] In this embodiment, the computing network is an edge AI inference system. The main component is an edge AI inference gateway, which receives inference requests from terminal devices, performs basic AI inference, schedules and allocates resources, and aggregates inference results. The auxiliary components are multiple GPU inference nodes, which provide high-precision AI inference acceleration under the scheduling of the edge AI inference gateway. For example, n=8 and k=5.

[0049] S2. Based on the operational data, construct a reliability model to describe the closed-loop coupling relationship between the main component and the auxiliary subsystem. This specifically includes the following sub-steps:

[0050] S201. Based on the historical fault records and load data obtained in S1, the operating environment of the computing network is modeled. The operating environment of the computing network (such as the service load level) changes dynamically over time and affects the self-healing rate of the main components. This embodiment uses a continuous-time Markov process. Describe the dynamic changes in the operating environment. Let be the time variable, where the environment state space of the continuous-time Markov process is denoted as . and state transition rate matrix This data was obtained from historical statistics.

[0051] , In this embodiment, the number of environmental states is [number]. =3, corresponding to low load, medium load, and high load operating environments, respectively. The initial environment probability distribution is denoted as... , For the first The initial probabilities of each environment state, and the state transitions between environments are determined by the transition rate matrix. Regulation, For from the first The environmental state transitions to the first... The transition rate of each environmental state. In this embodiment, , for:

[0052]

[0053] S202. Based on the impact event data in the fault records obtained in S1, model the external random impacts experienced by the main component under different operating environments. In this embodiment, a homogeneous Poisson process is used. Describe the arrival pattern of the impact. Indicates the first Under each environmental state, within the time interval (0, ... The number of impacts occurring within a given timeframe, and the arrival rate of the homogeneous Poisson process. , Indicates the first The average number of shocks occurring per unit time under a given environmental condition. For example, =0.8 / day, meaning that on the first day... An impact occurs on average every 1.25 days under various environmental conditions. The impact arrival rate and damage distribution are obtained by fitting historical data. Specifically, the average number of impacts per unit time in historical records is used as an estimate of the impact arrival rate. Statistical analysis is performed on the original damage recorded in historical impact events to fit the probability distribution parameters of the damage. In this embodiment, an exponential distribution is used for fitting, and its rate parameter is determined by statistical analysis of historical data. For example, the mean of historical impact damage can be calculated, and its reciprocal is the estimate of the rate parameter, or conventional statistical methods such as maximum likelihood estimation can be used to determine it.

[0054] No. In the first environment The original damage to the main component caused by the secondary impact. These are independent and identically distributed non-negative random variables. In this embodiment, an exponential distribution is used to describe them. The probability distribution, whose probability density function is... for:

[0055]

[0056] in, For the first The rate parameter of the exponential distribution under various environmental conditions. for One of the possible values, In this embodiment, , , .

[0057] S203. Based on the auxiliary component failure data and load data obtained in S1, construct a forward-coupled sub-model to describe the load-sharing failure process of the auxiliary components. The lifetime of each auxiliary component follows an exponential distribution, and the rate parameter of this distribution is the failure rate of the auxiliary component. All auxiliary components share the total working load. When some auxiliary components fail, the load borne by the remaining auxiliary components increases, leading to an increase in the failure rate. The failure rate of the remaining auxiliary components is expressed as:

[0058]

[0059] in, As the baseline failure rate, This represents the number of failed auxiliary components. The number of failed auxiliary parts is The failure rate of the remaining auxiliary components is given by the cumulative distribution function of the remaining auxiliary components at this time. , This represents a given lifespan threshold. The specific value can be determined according to actual needs, such as 24 days, 50 days, etc. This indicates that the remaining auxiliary components have a lifespan of less than or equal to The probability of; The total workload undertaken by the auxiliary subsystem. For about It is a monotonically increasing function. In this embodiment, we take... , =20, The specific expression was determined by fitting auxiliary component failure records and load data.

[0060] S204. Based on the self-healing records of the main component and the state data of the auxiliary component obtained in S1, construct a reverse-coupled sub-model. The self-healing rate of the main component... Represented as:

[0061]

[0062] in, The self-healing rate of the main component is expressed as the rate at which the component heals in the first... Under various environmental conditions, when the number of failed auxiliary components is u, what is the repair ratio of the main component to the impact damage? ; In the first Under what environmental conditions, regarding The monotonically increasing function, the form of which is determined by fitting the self-healing records of the main component and the status data of the auxiliary component (such as health status, failure time, load records, etc.), .

[0063] For the The first environmental state Secondary impact, actual damage to main components for:

[0064]

[0065] in, This is the current self-healing rate of the main component. From this formula, we can see that the self-healing rate... The higher the value, the greater the actual damage. The smaller the value, the higher the self-healing rate. The lower the value, the greater the actual damage. The larger.

[0066] No. Under certain environmental conditions, when the main component is Actual damage accumulated over time Exceeding the failure threshold When the main component fails:

[0067]

[0068] In this step, the number of failed auxiliary components It acts simultaneously on both the forward-coupled sub-model and the reverse-coupled sub-model. In the forward coupling... The increase leads to a higher load on the remaining auxiliary components, a higher failure rate, and accelerates the failure of more auxiliary components; in reverse coupling, Increased activity leads to a decrease in the self-healing rate of the main components, making them more prone to failure.

[0069] Auxiliary component failure rate The increase and rise are related to the self-healing rate of the main components. The effects of increasing and decreasing are both achieved through the same variable. The components are mutually coupled, forming a positive feedback loop: the self-healing ability of auxiliary components decreases upon failure, leading to accelerated failure of main components. This invention achieves a precise characterization of the main-auxiliary coupled structure of a computing network through this closed-loop coupling relationship. (Reference) Figure 2 , Figure 2 The diagram illustrates a sample path of performance degradation in computing networks, vividly demonstrating the change in the amount of cumulative damage to the main component under random impacts and self-healing mechanisms, as well as the impact of auxiliary component failures on self-healing capabilities.

[0070] S3. Based on the reliability model established in S2, solve for the time-domain reliability function. The specific process is as follows:

[0071] S301. Constructing the reliability function based on the reliability model established in S2. , Indicates the failure threshold is Under these conditions, the computing power network in the time interval The probability that no main component failure will occur. Based on the law of total probability, the reliability function... This can be expressed as the sum of the joint probabilities of the operating environment state and the number of failed auxiliary components:

[0072]

[0073] in, The number of failed auxiliary components at the initial time is The probability; in this embodiment, it is assumed that the computing network is in a completely new state at the initial moment, that is, all auxiliary components are working normally, i.e. , In practical applications, It can also be obtained from historical operating data statistics, or set according to the status after system maintenance.

[0074] Let be the conditional reliability function, representing the condition when the initial environmental state is the th . The environmental conditions and the number of initially failed auxiliary components are as follows: Under the condition that the system has a failure threshold of The time is Reliability at that time.

[0075] S302, Regarding the reliability function Applying the double Laplace transform, i.e., applying the time variable separately... and failure threshold variable Performing a Laplace transform converts the partial differential-integral equation into an algebraic equation, yielding the Laplace domain reliability function. Reliability function about and The double Laplace transform is:

[0076]

[0077] in, For time variables The corresponding Laplace variable, Failure threshold The corresponding Laplace variable.

[0078] S303. Solve for the Laplace domain reliability function in the Laplace domain to obtain the Laplace domain analytic function. :

[0079]

[0080] in, It is the joint probability distribution vector of the operating environment and the number of failed auxiliary components at the initial moment, which is... , It is the identity matrix; It is a coefficient matrix, whose elements are determined by the Laplace transform of model parameters (failure rate of auxiliary components, self-healing rate of main components, environmental transfer rate, impact arrival rate, etc.). The function matrix is ​​known and determined by the initial and boundary conditions; This is a column vector of all 1s. This expression is a formal solution to the Laplace field analytic function. and The specific form is derived from the parameters in the reliability model using the standard Laplace transform method. In this embodiment, The dimension is , and All dimensions are , The dimension is .

[0081] S304. Perform numerical inversion of the solution Laplace domain reliability function using inverse Laplace transform to obtain the time domain reliability function. This embodiment employs the Gaver-Stehfest algorithm for numerical inversion, which approximates the time-domain function by representing the Laplace domain function as a weighted sum of a series of exponential functions.

[0082] S4. Based on the time-domain reliability function obtained from the solution. Determine various reliability indicators for computing power networks:

[0083] Reliability The probability that the system will function normally at time t, i.e., through It can be calculated directly;

[0084] Failure probability The probability that the system has failed at time t is calculated by the following formula:

[0085]

[0086] Mean Time to First Failure The expected time before the system's first failure is calculated using the following formula:

[0087]

[0088] Users can select appropriate metrics based on different evaluation needs: when it is necessary to evaluate the operational reliability of the system at a specific point in time, output reliability. When risk analysis is required, output the failure probability. When it is necessary to evaluate the overall system reliability or compare different solutions, the mean time to first failure (MTBF) is output. .

[0089] The specific embodiments of the present invention are provided to enable those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention.

[0090] It should be understood that the present invention is not limited to the content already described above, and various modifications and changes can be made without departing from its scope. The scope of the present invention is limited only by the appended claims.

Claims

1. A reliability assessment method for computing power networks considering functional self-healing, characterized in that, include: S1. Obtain the operation data of the computing power network, wherein the computing power network includes a main component and an auxiliary subsystem, and the auxiliary subsystem is composed of multiple auxiliary components; S2. Construct a reliability model based on the operational data. The reliability model is used to describe the closed-loop coupling relationship between the main component and the auxiliary subsystem. The closed-loop coupling relationships include: Among the auxiliary components, the failure rate of the remaining auxiliary components increases with the increase of the number of failed auxiliary components; The self-healing rate of the main component decreases as the number of failed auxiliary components increases; S3. Based on the reliability model, solve the time-domain reliability function of the computing power network, and determine the reliability index of the computing power network according to the time-domain reliability function. The reliability index includes at least one of reliability, failure probability or mean time to first failure.

2. The reliability assessment method for computing power networks considering functional self-healing as described in claim 1, characterized in that, The auxiliary subsystem consists of n auxiliary components. When the number of failed auxiliary components reaches k, the entire auxiliary subsystem fails, where n≥1 and 1≤k≤n.

3. The reliability assessment method for computing power networks considering functional self-healing as described in claim 1, characterized in that, The failure rate of the auxiliary component is expressed as follows: in, As the baseline failure rate, This represents the number of failed auxiliary components. The number of failed auxiliary parts is Failure rate of remaining auxiliary components. The total workload undertaken by the auxiliary subsystem. The number of auxiliary components in the auxiliary subsystem. For about It is a monotonically increasing function.

4. The reliability assessment method for computing power networks considering functional self-healing as described in claim 1, characterized in that, The self-healing rate of the main component depends on the current operating environment of the computing network, which is described by a continuous-time Markov process, having an environment state space. and state transition rate matrix , For the number of environmental states, For from the first The environmental state transitions to the first... The transition rate of each environmental state.

5. The reliability assessment method for computing power networks considering functional self-healing according to claim 4, characterized in that, The self-healing rate of the main component is expressed as: in, The self-healing rate of the main component is expressed as the rate at which the component heals in the first... Under various environmental conditions, when the number of failed auxiliary components is u, what is the repair ratio of the main component to the impact damage? ; The number of auxiliary components in the auxiliary subsystem. In the first Under what environmental conditions, regarding A monotonically increasing function. .

6. The reliability assessment method for computing power networks considering functional self-healing as described in claim 5, characterized in that, The main component is also subject to external random impacts during operation, and these random impacts are homogeneous Poisson processes. Describe the arrival rate. ,in For time variables, Indicates the first Under each environmental state, within the time interval (0, ... The number of impacts occurring within; In the The average number of impacts per unit time under each environmental condition; the damage caused to the main component by each impact is an independent and identically distributed non-negative random variable.

7. The reliability assessment method for computing power networks considering functional self-healing as described in claim 1, characterized in that, In S3, solving the reliability function of the computing network includes: An integral equation for the conditional reliability function is established based on the reliability model. The integral equation is then transformed into an algebraic equation by applying a double Laplace transform. After solving the algebraic equation, the time-domain reliability function is obtained through numerical inversion using the inverse Laplace transform.

8. The reliability assessment method for computing power networks considering functional self-healing according to claim 1, characterized in that, Calculate the mean time to first failure (MTBF) of the computing network based on the time-domain reliability function. The formula is: in This is the time-domain reliability function.

9. A reliability assessment system for computing power networks considering functional self-healing, used to implement the method described in any one of claims 1 to 8, characterized in that, include: The data processing module is used to acquire the operating data of the computing power network, which includes a main component and an auxiliary subsystem, and the auxiliary subsystem is composed of multiple auxiliary components. The model building module is used to build a reliability model based on the operational data. The reliability model is used to describe the closed-loop coupling relationship between the main component and the auxiliary subsystem. The closed-loop coupling relationship includes: the failure rate of the auxiliary component increases with the increase of the number of failed auxiliary components, and the self-healing rate of the main component decreases with the increase of the number of failed auxiliary components. The solution calculation module is used to solve the reliability index of the computing power network based on the reliability model. The output module is used to output the evaluation results.

Images