Server cluster security control method combined with multi-level encryption strategy and joint scheduling

Abstract

The present invention relates to the field of computer technology and discloses a server cluster security control method combined with multi-level encryption strategy and joint scheduling, comprising the following steps: S1, performing sensitivity assessment on computing tasks and data within the cluster, and classifying the data into low-sensitivity data, medium-sensitivity data, and high-sensitivity data according to predetermined rules; S2, encrypting low-sensitivity data using a first encryption algorithm, encrypting medium-sensitivity data using a second encryption algorithm, and encrypting high-sensitivity data using a homomorphic encryption algorithm; S3, a joint scheduling mechanism based on reinforcement learning, obtaining cluster load, computing task requirements, and data sensitivity information in real time. Through multi-level encryption strategies, dynamic resource scheduling based on deep Q learning, and a security protection mechanism combining a zero-trust architecture with deep learning, the problems of low cluster resource utilization efficiency, insufficient data security, and inability to respond to security threats in real time are resolved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a server cluster security control method combined with multi-level encryption strategy joint scheduling. Background Art

[0002] With the widespread adoption of server clusters in fields like big data and cloud computing, efficient cluster resource management and data security have become critical issues. Traditional cluster management solutions often rely on static configurations and rules, failing to dynamically adjust to changes in cluster load, task requirements, and data sensitivity. This static resource scheduling approach not only fails to effectively utilize cluster resources but also easily leads to compute node overload or idle resources, impacting overall cluster performance.

[0003] In terms of data security, existing encryption technologies typically use a unified encryption strategy, lacking flexible encryption schemes for data of varying sensitivity, resulting in significant performance overhead for computing tasks. Furthermore, traditional encryption methods often experience performance bottlenecks in large clusters, especially when processing large amounts of data, where performance losses are more pronounced.

[0004] Traditional security measures for clusters often rely on trust-based models, neglecting fine-grained access control and dynamic permission management, and failing to effectively prevent unauthorized access and potential insider threats. Furthermore, the diverse security threats and complex attack patterns present in cluster environments pose significant challenges to traditional security measures, necessitating the need for more intelligent and efficient security control strategies. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides a server cluster security control method combined with multi-level encryption strategy joint scheduling, which solves the problems of low resource scheduling efficiency, insufficient data security protection and insufficient ability to deal with dynamic security threats in server clusters.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a server cluster security control method combined with multi-level encryption strategy joint scheduling, comprising the following steps:

[0007] S1. Perform sensitivity assessment on computing tasks and data within the cluster, and classify data into low-sensitivity data, medium-sensitivity data, and high-sensitivity data according to pre-defined rules.

[0008] S2. Encrypt low-sensitivity data using the first encryption algorithm, encrypt medium-sensitivity data using the second encryption algorithm, and encrypt high-sensitivity data using the homomorphic encryption algorithm;

[0009] S3, a joint scheduling mechanism based on reinforcement learning, dynamically adjusts the allocation of computing resources and task scheduling by obtaining real-time information about cluster load, computing task requirements, and data sensitivity;

[0010] S4. Deploy a zero-trust architecture in the cluster, requiring each request to undergo authentication and authorization, and implement fine-grained control over task permissions through role-based access control policies.

[0011] S5: Monitor computing tasks and data transmission in the cluster in real time, use deep learning algorithms to detect anomalies in cluster behavior, and identify potential security threats;

[0012] S6. When a potential threat is detected, the security response mechanism is automatically executed to provide protection by isolating abnormal nodes and restricting resource access of malicious tasks.

[0013] Preferably, the sensitivity assessment of the data in the cluster in step S1 specifically includes the following steps:

[0014] S1.1. Collect data on computing tasks within the cluster and analyze the data type, content, and access frequency.

[0015] S1.2. Conduct a sensitivity assessment of the data using a trained machine learning model or specified statistical methods, and classify the data into low-sensitivity data, medium-sensitivity data, and high-sensitivity data based on the assessment results;

[0016] S1.3. Classify the data based on the assessment results and assign corresponding encryption algorithms and security policies according to the categories;

[0017] S1.4. Update the sensitivity level of data regularly or in real time based on changes in cluster resources, and dynamically adjust encryption policies.

[0018] Preferably, the step S2 specifically includes the following steps:

[0019] S2.1. Encrypt low-sensitivity data using the AES-128 encryption algorithm to generate the corresponding encrypted ciphertext;

[0020] S2.2. Encrypt the sensitive data using the AES-256 encryption algorithm to generate the corresponding encrypted ciphertext;

[0021] S2.3. Encrypt the highly sensitive data using a homomorphic encryption algorithm, wherein the homomorphic encryption algorithm is the Paillier encryption algorithm;

[0022] S2.4. All encryption processes generate and manage keys through a centralized key management system, including key lifecycle management, key storage and distribution.

[0023] Preferably, the joint scheduling mechanism based on reinforcement learning in step S3 includes:

[0024] S3.1. Collect the current cluster resource usage, including the load, memory and CPU usage of the computing nodes, as well as the computing requirements and priorities of the tasks;

[0025] S3.2. Based on deep Q-learning, dynamically adjust task scheduling and resource allocation strategies while satisfying resource allocation constraints.

[0026] S3.3. Based on the encryption level, data sensitivity, and computing resource requirements of the task, select a computing node with idle resources that can meet the encryption computing requirements;

[0027] S3.4. When scheduling computing tasks, consider resource utilization, load balancing, and security.

[0028] Preferably, the implementation of the zero trust architecture in step S4 specifically includes:

[0029] S4.1. Implement OAuth2.0 or OpenIDConnect-based authentication for every request within the cluster.

[0030] S4.2. Implement attribute-based access control to control access rights based on the sensitivity of the task, the identity and role information of the requester;

[0031] S4.3. Each access request is subject to permission verification;

[0032] S4.4. Dynamically adjust permissions so that permissions are updated in real time based on task requirements and access conditions during the task life cycle.

[0033] Preferably, the step S5 specifically includes:

[0034] S5.1. Use the Prometheus monitoring system to monitor the cluster's compute node resource usage in real time and visualize it using Grafana.

[0035] S5.2. Use deep learning algorithms to model cluster behavior and identify abnormal patterns that deviate from normal behavior;

[0036] S5.3. Use model-based anomaly detection methods to automatically trigger alarms and notify administrators when cluster resources or task behaviors are abnormal.

[0037] S5.4. Combine the monitoring system with anomaly detection results to generate real-time security reports to help administrators take timely response measures.

[0038] Preferably, the security response mechanism in step S6 includes:

[0039] S6.1. When a security threat is detected, the malicious node is automatically isolated and the computing tasks on the node are stopped to prevent the security threat from spreading.

[0040] S6.2. Limit resource access rights of affected nodes or tasks to reduce interference with normal computing tasks.

[0041] S6.3. Automatically adjust cluster access permissions based on the severity of abnormal behavior to prohibit unauthorized users or nodes from accessing sensitive data.

[0042] S6.4. Keep detailed records of all security incidents, generate audit logs, and provide them to administrators for subsequent analysis and investigation.

[0043] Preferably, the joint scheduling mechanism of reinforcement learning in step S3 optimizes the scheduling strategy through the following mathematical model: ; Among them, R(t) represents the reward value of the scheduling strategy, r i represents the available capacity of resource i, t j represents the demand of task j, x ij is a Boolean variable indicating whether task j is assigned to resource i, α and β are regularization coefficients, n is the number of resource nodes, m is the number of tasks, and d j is the resource requirement of task j, p j is the priority of task j.

[0044] The present invention provides a server cluster security control method that combines multi-level encryption strategy with joint scheduling. It has the following beneficial effects:

[0045] 1. The present invention implements encryption protection for data of different sensitivity levels by performing sensitivity assessments on computing tasks and data within the cluster and selecting appropriate encryption algorithms based on the assessment results. Compared with traditional unified encryption schemes, the multi-level encryption strategy of the present invention can perform customized encryption based on the specific sensitivity of the data, greatly improving the flexibility and security of data encryption. Through this flexible encryption management, the encryption burden of low-sensitivity data can be effectively reduced, ensuring stronger protection for highly sensitive data, thereby improving resource utilization efficiency while ensuring the overall security of the cluster.

[0046] 2. By introducing reinforcement learning algorithms, specifically a joint scheduling mechanism based on deep Q-learning, this invention can acquire cluster load, task requirements, and data sensitivity information in real time, and dynamically adjust the allocation of computing resources and task scheduling. Compared with traditional static resource allocation or simple scheduling algorithms, the dynamic scheduling mechanism based on reinforcement learning can optimize according to the real-time changes in workload and task requirements, automatically selecting the most appropriate computing resources for task scheduling, thereby improving cluster resource utilization efficiency and task processing capabilities.

[0047] 3. The present invention combines zero-trust architecture with deep learning algorithms to achieve strict access control and real-time abnormal behavior detection within the cluster. All requests must undergo identity authentication and permission authorization. At the same time, fine-grained access control is performed based on the sensitivity of the task and the user's role information. Using deep learning algorithms, the system can monitor cluster behavior in real time and identify abnormal patterns. Once a potential security threat is discovered, it automatically triggers a security response mechanism, such as isolating abnormal nodes and restricting malicious task resource access, thereby effectively preventing the spread of security threats. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 Flowchart of the present invention. DETAILED DESCRIPTION

[0049] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the present specification. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0050] Please see the attached Figure 1 The embodiment of the present invention provides a server cluster security control method combined with multi-level encryption strategy joint scheduling, including the following steps:

[0051] S1. Perform sensitivity assessment on computing tasks and data within the cluster, and classify data into low-sensitivity data, medium-sensitivity data, and high-sensitivity data according to predetermined rules.

[0052] The sensitivity assessment of the data in the cluster in step S1 specifically includes the following steps:

[0053] S1.1. Collect data on computing tasks within the cluster and analyze the data type, content, and access frequency.

[0054] S1.2. Conduct a sensitivity assessment of the data using a trained machine learning model or specified statistical methods, and classify the data into low-sensitivity data, medium-sensitivity data, and high-sensitivity data based on the assessment results;

[0055] S1.3. Classify the data based on the assessment results and assign corresponding encryption algorithms and security policies according to the categories;

[0056] S1.4. Update the sensitivity level of data regularly or in real time based on changes in cluster resources, and dynamically adjust encryption policies.

[0057] Specifically, in step S1.1, the system first collects all relevant computing task data from each compute node in the cluster. Data collection not only considers the type and content of the data, but also considers the access frequency, access patterns, and historical usage. For example, some data may be frequently accessed by different compute nodes, while other data may only be used in specific tasks. Therefore, access frequency and patterns are one of the important criteria for sensitivity assessment. This information can be obtained in real time through the cluster management system or a dedicated log analysis system to ensure the comprehensiveness and accuracy of data collection.

[0058] In step S1.2, based on the collected task data, the system uses trained machine learning (such as support vector machines, decision trees, random forests, etc.) or statistical methods to assess the sensitivity of the data. Common sensitivity assessment models can comprehensively determine the sensitivity level of each data item based on characteristics such as data type (such as financial data, user information), content (such as sensitive fields, private data), and access frequency. Models used in the data assessment process can be trained using historical data and continuously adjusted and optimized to improve assessment accuracy.

[0059] The output of the sensitivity assessment divides the data into three categories:

[0060] Low-sensitivity data: This data may be public data or data with low security requirements, and can be protected by conventional encryption or simple access control measures.

[0061] Medium sensitive data: This data requires moderate protection measures and is processed using strong encryption algorithms and appropriate access controls.

[0062] Highly sensitive data: These data are extremely sensitive core data and require the highest level of protection. They are usually encrypted using advanced encryption technologies such as homomorphic encryption, and their access rights are strictly controlled. Formula: Let the dataset be ; where d i For the i-th data, the data sensitivity assessment is performed using the following model: ; Among them, S (d i ) represents data d i The sensitivity evaluation result of , f is the trained sensitivity evaluation model, and the output is the sensitivity level of the data.

[0063] Based on the evaluation results, the system classifies the data according to its sensitivity level. Specifically, for different categories of data, the system specifies different encryption algorithms and access control policies:

[0064] For low-sensitivity data, simpler encryption algorithms such as AES-128 are used, and basic authentication and access control are used to protect it.

[0065] For medium-sensitive data, AES-256 encryption algorithm is used for protection, and stricter access rights are configured for the data;

[0066] For highly sensitive data, homomorphic encryption algorithms (such as the Paillier encryption algorithm) are used for encryption to ensure that the data can still be processed and calculated even in the encrypted state.

[0067] During this step, the system maps encryption algorithms to data sensitivity levels, ensuring appropriate protection measures for each data category. Furthermore, the system allocates appropriate computing resources and scheduling strategies to different data levels based on task requirements, ensuring that encrypted computing does not significantly impact cluster performance.

[0068] As computing tasks within the cluster change and data access patterns adjust, the sensitivity of the data may change. Therefore, in this step, the system needs to update the data sensitivity level periodically or based on real-time conditions. This process is achieved through continuous monitoring and data assessment to ensure that the data maintains the appropriate security protection level throughout its lifecycle.

[0069] During the update process, the system dynamically adjusts data encryption policies based on new sensitivity assessment results. For example, if access frequency of certain moderately sensitive data increases significantly, the system may upgrade it to highly sensitive data and switch to a stronger encryption algorithm. Furthermore, the system optimizes encryption computing resource allocation based on task priorities and changes in computing resources to ensure a balance between data protection and cluster resources.

[0070] This embodiment combines machine learning methods with statistical analysis to assess data sensitivity and implements a multi-level encryption strategy based on the assessment results, ensuring that data within the cluster is appropriately protected based on diverse security requirements. During the sensitivity assessment process, the system incorporates multiple factors, such as task type and data access frequency, using an intelligent assessment model to ensure the accuracy of sensitivity classification. This method offers high flexibility and scalability, effectively addressing various security threats in a cluster environment.

[0071] S2. Encrypt low-sensitivity data using the first encryption algorithm, encrypt medium-sensitivity data using the second encryption algorithm, and encrypt high-sensitivity data using the homomorphic encryption algorithm;

[0072] Step S2 specifically includes the following steps:

[0073] S2.1. Encrypt low-sensitivity data using the AES-128 encryption algorithm to generate the corresponding encrypted ciphertext;

[0074] S2.2. Encrypt the sensitive data using the AES-256 encryption algorithm to generate the corresponding encrypted ciphertext;

[0075] S2.3. Use the homomorphic encryption algorithm to encrypt highly sensitive data. The homomorphic encryption algorithm is the Paillier encryption algorithm.

[0076] S2.4. All encryption processes generate and manage keys through a centralized key management system, including key lifecycle management, key storage and distribution.

[0077] Specifically, in this embodiment, step S2 encrypts the data, which specifically includes the following:

[0078] For less sensitive data, the AES-128 (Advanced Encryption Standard, 128-bit key length) encryption algorithm is used. AES-128 is a symmetric encryption algorithm with a relatively simple encryption process and fast computation speed. It is suitable for scenarios with large data volumes and relatively low security requirements. While data encrypted with the AES-128 algorithm provides some protection, due to its low encryption strength, it is suitable for protecting non-sensitive data or data that is less susceptible to abuse, such as public system logs or other data with lower security requirements.

[0079] In this step, the system encrypts the low-sensitivity data using a 128-bit key, generating the corresponding encrypted ciphertext. The encrypted data is then stored or transmitted to the cluster's compute nodes, where it is protected from unauthorized access during transmission.

[0080] Sensitive data often includes critical business information (such as user profiles and certain commercial data), and therefore requires stronger encryption. Compared to AES-128, AES-256 uses a 256-bit key for encryption, providing stronger security and effectively preventing brute force attacks.

[0081] In this step, the system generates a 256-bit key for the moderately sensitive data and encrypts it using the AES-256 algorithm. AES-256 encryption provides greater security and withstands stronger attacks, making it suitable for protecting data that requires a certain degree of confidentiality but is not extremely sensitive.

[0082] Highly sensitive data, such as financial data, personal privacy data, and core business data, is the most confidential and in need of protection in a cluster. To ensure that this data can still be calculated and analyzed after encryption, this embodiment uses a homomorphic encryption algorithm, specifically the Paillier encryption algorithm.

[0083] Homomorphic encryption is a specialized encryption method that allows certain computations to be performed directly on encrypted data without decryption, thus avoiding the risk of exposing the data during processing. The Paillier encryption algorithm is an additive homomorphic encryption algorithm particularly suitable for scenarios requiring addition operations on encrypted data, such as summing encrypted financial data without decryption. The Paillier algorithm performs encryption and decryption by generating public and private keys: the public key is used to encrypt data, and the private key is used to decrypt data.

[0084] Through this encryption algorithm, highly sensitive data can be calculated in an encrypted state, preventing data leakage while ensuring the security and accuracy of the calculation results.

[0085] Key management is a core component of encryption systems. All encryption processes rely on a centralized key management system to generate, distribute, and store keys. This system not only generates the keys required by encryption algorithms (such as AES-128, AES-256, and Paillier keys) but also manages the key lifecycle, including key creation, renewal, revocation, storage, and distribution.

[0086] Specifically, a key management system must ensure secure key storage, preventing unauthorized access or disclosure. To this end, keys are typically stored in hardware security modules (HSMs) or key management hardware, providing a high level of protection. Key management systems also need to ensure that the key distribution process is secure, ensuring that only authenticated and authorized nodes can access the keys.

[0087] In addition, the key management system regularly updates keys to reduce the risk of key cracking. For keys that have expired or are no longer needed, the system will revoke them to ensure the validity and security of the keys.

[0088] This embodiment uses a multi-level encryption strategy to apply AES-128, AES-256, and Paillier homomorphic encryption algorithms to low-sensitivity data, medium-sensitivity data, and high-sensitivity data within the cluster, respectively. Different encryption algorithms ensure the security of data of different sensitivity levels while minimizing the consumption of computing resources. All encryption processes use a centralized key management system for key generation, storage, distribution, and lifecycle management to ensure the security and effectiveness of the keys. This solution provides a flexible and efficient solution for data security in a cluster environment, helping to optimize cluster resource utilization while ensuring security.

[0089] S3, a joint scheduling mechanism based on reinforcement learning, dynamically adjusts the allocation of computing resources and task scheduling by obtaining real-time information about cluster load, computing task requirements, and data sensitivity;

[0090] The joint scheduling mechanism based on reinforcement learning in step S3 includes:

[0091] S3.1. Collect the current cluster resource usage, including the load, memory and CPU usage of the computing nodes, as well as the computing requirements and priorities of the tasks;

[0092] S3.2. Based on deep Q-learning, dynamically adjust task scheduling and resource allocation strategies while satisfying resource allocation constraints.

[0093] S3.3. Based on the encryption level, data sensitivity, and computing resource requirements of the task, select a computing node with idle resources that can meet the encryption computing requirements;

[0094] S3.4. When scheduling computing tasks, consider resource utilization, load balancing, and security.

[0095] The joint scheduling mechanism of reinforcement learning in step S3 optimizes the scheduling strategy through the following mathematical model: ; Among them, R(t) represents the reward value of the scheduling strategy, r i represents the available capacity of resource i, t j represents the requirements of task j, x ij is a Boolean variable indicating whether task j is assigned to resource i, α and β are regularization coefficients, n is the number of resource nodes, m is the number of tasks, and d j is the resource requirement of task j, p j is the priority of task j.

[0096] Specifically, in this embodiment, the resource usage of the cluster includes the load, memory and CPU usage of each computing node, as well as the computing requirements and priority information of the current computing task. Specifically:

[0097] Computing node load collection: The cluster management system monitors the resource usage of each node in real time. The load status of computing node i is determined by cpu i and mem i Two indicators are used to measure the load function of the node. ; Where f is the load function, which means that the load of the node is comprehensively judged by calculating the CPU and memory resource consumption of the node.

[0098] Task computing requirements and priorities: Each computing task j has its computing resource requirements ;in: and are the CPU and memory resource requirements of task j, p j The priority of the task, the task with higher priority will be scheduled first.

[0099] In this way, the system can accurately collect the resource usage and task demand status of all nodes in the cluster, providing data support for subsequent scheduling decisions.

[0100] After collecting cluster resource and task information, the system then uses the deep Q learning method in reinforcement learning to implement dynamic task scheduling and resource allocation. The specific steps are as follows:

[0101] Definition of state space:

[0102] The state space of the system contains the resource status of all computing nodes and the demand information of the tasks to be scheduled. t It can be represented as the resource status of all nodes in the cluster and the set of tasks currently to be scheduled: ;in, ; Indicates the resource status of the i-th node, ; represents the resource requirements and priority of the jth task.

[0103] Definition of action space: action space a t represents the choice of the current scheduling decision. For each task j, the system can choose whether to assign it to node i action x ij is a Boolean variable indicating whether task j is assigned to resource node i: Reward function design: The goal of the reward function is to maximize the efficiency of cluster resource utilization, optimize load balancing, and consider task priorities. Based on the task allocation results, the reward function R(t) can be defined by the following formula: ; Among them, R(t) represents the reward value of the scheduling strategy, r i represents the available capacity of resource i, t j represents the demand of task j, x ijis a Boolean variable indicating whether task j is assigned to resource i, α and β are regularization coefficients, n is the number of resource nodes, m is the number of tasks, and d j is the resource requirement of task j, p j is the priority of task j.

[0104] This reward function aims to optimize resource scheduling by considering resource utilization and task priority, so that tasks can be effectively scheduled to appropriate computing nodes.

[0105] Q-value update rule: In reinforcement learning, Q-value represents the expected reward of taking a certain action in a certain state. After each scheduling decision, the system updates the Q-value based on the reward obtained. The Q-value update follows the classic Q-learning formula: ; Among them: α is the learning rate, which controls the speed of learning, γ is the discount factor, which determines the importance of future rewards, r t is the reward obtained at the current time step. ; is the next state s t+1 The maximum Q value of all possible actions.

[0106] In this embodiment, task scheduling not only considers its computing resource requirements and priority, but also the encryption level and data sensitivity of the task. Based on the different encryption requirements of the task, the system will schedule the task to the appropriate computing node to meet security and computing requirements:

[0107] Encryption level: Low-sensitivity tasks use standard encryption (such as AES-128), medium-sensitivity tasks use stronger encryption (such as AES-256), and high-sensitivity tasks use homomorphic encryption algorithms (such as Paillier encryption).

[0108] Encryption computing resource requirements: Since encryption algorithms consume different computing resources, the system dynamically calculates its resource requirements based on the encryption level of the task. It can be calculated by the following model: ; Among them, f enc is the encryption requirement function, which represents the resource consumption of tasks under different encryption levels.

[0109] Resource allocation: Based on the encryption requirements of the task, the system selects computing nodes that can provide sufficient computing power to ensure that the encryption task can be completed smoothly.

[0110] During the scheduling process, the system takes the following factors into consideration:

[0111] Resource utilization and load balancing: To avoid overloading certain nodes in the cluster, the system will make a trade-off between load balancing and resource utilization. For example, if the load on a node exceeds 80%, the system will avoid assigning new tasks to it.

[0112] Security: For highly sensitive tasks, the system will give priority to scheduling nodes with reinforced security protection to ensure that the transmission and calculation process of task data meets security requirements.

[0113] S4. Deploy a zero-trust architecture in the cluster, requiring each request to undergo authentication and authorization, and implement fine-grained control over task permissions through role-based access control policies.

[0114] The implementation of the zero trust architecture in step S4 specifically includes:

[0115] S4.1. Implement OAuth2.0 or OpenIDConnect-based authentication for every request within the cluster.

[0116] S4.2. Implement attribute-based access control to control access rights based on the sensitivity of the task, the identity and role information of the requester;

[0117] S4.3. Each access request is subject to permission verification;

[0118] S4.4. Dynamically adjust permissions so that permissions are updated in real time based on task requirements and access conditions during the task life cycle.

[0119] Specifically, in this embodiment, to ensure that each request undergoes strict identity verification, the cluster uses authentication protocols based on OAuth2.0 or OpenIDConnect for identity verification. These protocols can effectively confirm the user's identity and ensure the legitimacy of the requester's identity.

[0120] OAuth 2.0 authentication: The OAuth 2.0 protocol allows client applications to access resource servers on behalf of users. The client must obtain an authorization token from the authorization server as a credential for the request. Each time a client requests a cluster resource, it must carry the OAuth 2.0 access token, which the server uses to verify its identity with the authorization server.

[0121] OpenID Connect authentication: OpenID Connect is an extension of the OAuth 2.0 protocol that provides additional authentication capabilities. Through OpenID Connect, the cluster can verify the requester's identity and obtain basic user information (such as name, email address, and role). Each request is authenticated by the identity provider and an ID token is generated for subsequent authorization verification.

[0122] Authentication process: The requester first requests authentication from the authentication server through the OAuth2.0 or OpenIDConnect protocol and obtains an identity token.

[0123] The token is sent along with the request to the cluster server, and the server confirms the identity of the requester by verifying the validity of the token.

[0124] If the identity authentication is successful, the subsequent permission authorization stage will be entered; if the verification fails, the request will be rejected and the corresponding error message will be returned.

[0125] Attribute-based access control is a more flexible and dynamic permission control mechanism that determines access rights by evaluating multiple attributes of the requester, such as identity, role, and task sensitivity. Compared to traditional role-based access control, it provides more refined permission management.

[0126] Access control policy: In attribute-based access control, the access control policy relies on a set of defined attributes, which can be divided into the following categories:

[0127] Requester attributes: including user identity information, role, group, etc.

[0128] Task attributes: including the sensitivity level of the task, the life cycle status of the task, etc.

[0129] Environmental attributes: including request time, geographic location, request source, etc.

[0130] Policy example:

[0131] For example, a highly sensitive task (task sensitivity is high) can only be accessed by a specific user group (such as the system administrator group) and can only be accessed during working hours (such as 9:00-18:00). This policy can be defined as: Through such policy definitions, attribute-based access control can ensure dynamic and flexible permission determination based on the specific circumstances of tasks and requesters in the cluster.

[0132] Implementation method: The access control system in the cluster will match the request with the relevant attributes and determine whether to allow access based on the preset access control policy.

[0133] If the requester meets all policy conditions, the system will allow the request to be executed, otherwise it will return a response of denying access.

[0134] In a zero-trust architecture, every request must undergo permission verification, regardless of whether the request originates from a legitimate source or a known user. All requested access permissions are temporary and dynamic, so each access triggers permission verification.

[0135] Permission verification process: Whenever the cluster receives a request, the system performs permission verification based on the context of the current request (including the requester's identity, request content, task sensitivity, etc.).

[0136] Permission verification usually includes the following aspects:

[0137] Authentication: Checks whether the requester has successfully authenticated via the OAuth 2.0 or OpenID Connect protocols.

[0138] Role and attribute matching: Checks whether the requester has sufficient permissions to access the target resource through attribute-based access control or role-based access control models.

[0139] Task sensitivity and access control: Based on the sensitivity of the task and the attributes of the requester, further determine whether the access conditions are met.

[0140] Verification failure: If the permission verification fails, the system will immediately terminate the execution of the request and return the corresponding error code (such as "403Forbidden") to the requester.

[0141] To better adapt to changes in the task lifecycle and different access modes, permissions need to be adjusted in real time based on task requirements and access conditions. The system can dynamically update permissions to ensure the security of tasks at different stages.

[0142] Permission adjustment triggers: When the sensitivity of a task changes, the system automatically adjusts access permissions. For example, when a task reaches a high-sensitivity stage, the system increases the strictness of permissions, restricting access to users with specific roles.

[0143] When the access frequency of a task increases, the system may temporarily adjust resources and permissions based on the load to prevent security risks caused by over-authorization or access to too many resources.

[0144] Implementation: Based on the task lifecycle and access logs, the cluster management system dynamically updates access permissions according to pre-set rules. For example, if a task enters a sensitive phase, the permission check level is automatically increased.

[0145] In addition, the cluster will log every permission change for audit and tracking purposes. Every permission adjustment operation will generate an audit log and synchronize it to the management platform in real time for monitoring and management.

[0146] S5: Monitor computing tasks and data transmission in the cluster in real time, use deep learning algorithms to detect anomalies in cluster behavior, and identify potential security threats;

[0147] Step S5 specifically includes:

[0148] S5.1. Use the Prometheus monitoring system to monitor the cluster's compute node resource usage in real time and visualize it using Grafana.

[0149] S5.2. Use deep learning algorithms to model cluster behavior and identify abnormal patterns that deviate from normal behavior;

[0150] S5.3. Use model-based anomaly detection methods to automatically trigger alarms and notify administrators when cluster resources or task behaviors are abnormal.

[0151] S5.4. Combine the monitoring system with anomaly detection results to generate real-time security reports to help administrators take timely response measures.

[0152] Specifically, to enable real-time monitoring of cluster computing tasks and resources, this example uses Prometheus as the cluster resource monitoring system. Prometheus regularly pulls resource usage metrics from each cluster node and saves them as time series data. Grafana is used to visualize Prometheus monitoring data, allowing administrators to intuitively understand the health of the cluster.

[0153] Prometheus monitoring system: Prometheus is an open source monitoring system that can regularly collect resource usage information for each node and supports efficient storage, query, and alarm. It mainly monitors the following aspects:

[0154] Compute node resource usage: such as CPU usage, memory usage, disk IO, network bandwidth, etc.

[0155] Task execution status: such as task runtime, status, resource requirements, etc.

[0156] Prometheus collects and saves resource information of each computing node through "exporters" to generate time series data.

[0157] Grafana visualization: Grafana, a visualization tool for Prometheus, presents real-time resource usage data to administrators in the form of charts and dashboards. Through Grafana, administrators can view the resource usage of each node in real time and identify potential performance bottlenecks or anomalies.

[0158] Implementation process:

[0159] Deploy Prometheus. Install a monitoring agent or exporter on each computing node and configure Prometheus to pull data regularly.

[0160] Configure Grafana to connect with Prometheus, and use Grafana to configure a visual dashboard to display the system's operating status, resource usage, task execution status, etc.

[0161] Generate monitoring reports regularly to facilitate administrators to do performance analysis.

[0162] In order to more efficiently identify abnormal behavior in the cluster, the system uses deep learning algorithms to model cluster behavior, identify patterns that deviate from normal behavior, and then detect potential security threats.

[0163] Deep learning model construction: Using resource monitoring data, task execution data, network traffic, and other data collected in the cluster as input, deep learning algorithms are used to train models. These models can learn normal operating and behavioral patterns.

[0164] Autoencoder: Through the autoencoder model, the normal pattern of cluster behavior is measured by the reconstruction loss of the input and output. When abnormal behavior occurs, the reconstruction error will increase significantly, so it can be used to detect behavioral anomalies.

[0165] Training process: First, collect a large amount of normal operating cluster data, including computing resource usage, task status, data transmission status, etc.

[0166] This data is trained using a deep learning algorithm (such as an LSTM model) to obtain a "normal behavior model" of the cluster.

[0167] In actual operation, real-time monitoring data is input into the deep learning model, and the loss value is calculated to determine whether there is anomaly.

[0168] Abnormal pattern recognition: When the behavior of real-time monitoring data deviates from the normal behavior learned by the model, the model will output a large anomaly score, indicating that the current cluster is abnormal.

[0169] These anomalies include but are not limited to: abnormal computing node load, sudden increase in network traffic, long task execution time, abnormal data transmission speed, etc.

[0170] Model-based anomaly detection compares cluster behavior with a trained model of normal behavior to determine if an anomaly has occurred. When an anomaly is detected, the system automatically triggers an alarm, notifying administrators to take timely action.

[0171] Anomaly Detection: The system compares real-time monitoring data with the normal patterns in the deep learning model to determine if anomalies exist. If there is a significant discrepancy between the model's predicted output and the actual monitoring data, the cluster's behavior deviates from normal patterns and is therefore identified as an anomaly.

[0172] Threshold-based alarm: When the anomaly score of a resource exceeds the set threshold, an alarm is triggered.

[0173] Automatic alarm mechanism: The system can automatically trigger alarms based on pre-defined policies and thresholds. These alarms include:

[0174] System performance alarms: such as high CPU usage, abnormal memory usage, etc.

[0175] Task execution exception alarm: such as task execution timeout, excessive resource request, etc.

[0176] Network anomaly alarm: such as sudden increase in network traffic, abnormal data transmission behavior, etc.

[0177] Notification mechanism: Alarm information will be pushed to the administrator in real time, and the notification content includes the abnormality type, impact range, specific indicators and possible solutions.

[0178] Administrators can be notified via various channels, such as email, SMS, instant messaging, etc.

[0179] The generation of real-time security reports is the final output of cluster monitoring and anomaly detection, helping administrators understand the current health status of the cluster and quickly respond to possible security incidents.

[0180] Report generation: During cluster operation, the monitoring system collects real-time data on resource usage of each node and generates a detailed security report based on deep learning anomaly detection results. The report typically includes:

[0181] Current resource usage: CPU, memory, and network bandwidth usage of each node.

[0182] Anomaly detection results: whether there is abnormal behavior, the type and scope of the anomaly.

[0183] Alarm records: Detailed records of abnormal events that occur within the system.

[0184] Report format: Real-time security reports can be displayed in various forms such as charts, logs, and warnings. After the report is generated, it will be sent to the administrator or security operations team immediately.

[0185] The report includes analysis of abnormal behavior, potential risk assessment, and recommended countermeasures, such as restricting access to abnormal nodes and optimizing resource allocation.

[0186] Countermeasures: Reports help administrators quickly understand the health status of the cluster and promptly identify and respond to potential security threats.

[0187] For example, if abnormal network traffic causes security risks, administrators can immediately isolate related nodes or restrict access to certain services.

[0188] S6. When a potential threat is detected, the security response mechanism is automatically executed to provide protection by isolating abnormal nodes and restricting resource access of malicious tasks.

[0189] The S6 step security response mechanism includes:

[0190] S6.1. When a security threat is detected, the malicious node is automatically isolated and the computing tasks on the node are stopped to prevent the security threat from spreading.

[0191] S6.2. Limit resource access rights of affected nodes or tasks to reduce interference with normal computing tasks.

[0192] S6.3. Automatically adjust cluster access permissions based on the severity of abnormal behavior to prohibit unauthorized users or nodes from accessing sensitive data.

[0193] S6.4. Keep detailed records of all security incidents, generate audit logs, and provide them to administrators for subsequent analysis and investigation.

[0194] Specifically, when the system detects a malicious node or potential security threat, isolation measures must be taken quickly to prevent the threat from spreading to other computing nodes or affecting other parts of the system.

[0195] Malicious node detection: The system uses deep learning models, behavioral analysis algorithms, and real-time monitoring data to determine whether a node has abnormal behavior, such as task timeouts, abnormal resource usage, and abnormal network traffic.

[0196] If the node behavior is detected to be significantly different from the normal pattern and meets the characteristics of malicious activities (such as malicious scripts, etc.), the system will determine that the node is a "malicious node".

[0197] Automatically isolate malicious nodes: Once a node is identified as malicious, the system will immediately isolate it. Isolation measures include:

[0198] Network isolation: Disconnect malicious nodes from the cluster's internal network to prevent them from communicating with other normal nodes.

[0199] Task stop: Forcefully stop all computing tasks on the node to prevent malicious tasks from being further executed.

[0200] Resource recycling: Immediately reclaim the computing resources occupied by the node and release them to other nodes for normal task processing.

[0201] Preventing the spread of threats: By quickly isolating malicious nodes, malicious behavior can be prevented from spreading to other nodes in the cluster or affecting critical data processing tasks, ensuring the overall security of the cluster.

[0202] After a threat is detected, in addition to isolating the malicious nodes, it is also necessary to restrict the resource access rights of the affected nodes or tasks to prevent them from interfering with normal computing tasks.

[0203] Resource access restrictions: For affected nodes or tasks, the system will take measures to restrict their resource access rights, including:

[0204] CPU / Memory Limit: Set quotas on the CPU and memory resources of the affected tasks to prevent them from consuming excessive resources.

[0205] Network access restriction: Limit the network bandwidth or network access capabilities of the affected nodes to reduce their interference with other nodes.

[0206] Storage access restriction: Limit the access of affected nodes to the storage system to prevent malicious data tampering or leakage.

[0207] Reduce interference: By precisely controlling the resource usage of affected nodes or tasks, we can ensure that the impact on other normally running computing tasks is minimized, while also effectively isolating potential security risks.

[0208] Based on the severity of the detected abnormal behavior, the system needs to dynamically adjust the access permissions in the cluster to ensure that sensitive data in the cluster is not accessed by unauthorized nodes or users.

[0209] Severity Assessment of Abnormal Behavior:

[0210] The system assesses the severity of detected abnormal behavior based on its nature and impact. For example, if a node contains malicious code or a network attack, the system will assess the potential threat to data.

[0211] Severity assessment criteria: The severity of abnormal behavior can be assessed based on factors such as the duration of the abnormal behavior, the scope of impact, and the sensitivity of the task.

[0212] Access rights adjustment:

[0213] If abnormal behavior is assessed to be severe, the system will automatically adjust access permissions based on the cluster's security policy:

[0214] Prohibit unauthorized users from accessing sensitive data: If a node is identified as a malicious node, all requests from that node to access sensitive data will be prohibited.

[0215] Strengthened authentication: Strengthen authentication for all users and nodes to ensure that only authorized nodes can access critical resources and sensitive data.

[0216] Restrict data transmission permissions: Restrict data transmission between tasks to prevent sensitive data from being illegally accessed or leaked.

[0217] Dynamic adjustment: Access rights can be adjusted in real time, and normal access rights policies can be restored after abnormal behavior is corrected or the threat is eliminated.

[0218] During the execution of the security response mechanism, the system will record all security incidents in detail and generate audit logs for administrators to conduct subsequent analysis and investigation, ensuring the transparency and traceability of the incidents.

[0219] Security event logging: All operations related to security threat detection, isolation, and permission adjustments will be recorded in the log. The log content includes:

[0220] Event time: The specific time when the security incident occurred.

[0221] Event type: For example, "Malicious node isolation", "Task stop", "Permission adjustment", etc.

[0222] Impact scope: affected nodes, tasks, or data.

[0223] Response measures: Security response measures taken, such as network isolation and resource restrictions.

[0224] Audit log generation: Audit logs will be stored in a secure log management system to ensure that the logs are not tampered with or deleted.

[0225] These log records can help administrators track and trace security incidents and further analyze the source, process, and impact of malicious attacks.

[0226] Subsequent analysis and investigation: Audit logs are not only an important reference for security response, but also provide data support for post-event analysis. Administrators can use the log content to:

[0227] Conduct in-depth analysis of malicious behavior and identify attack patterns.

[0228] Evaluate the effectiveness of cluster security policies and adjust security policies based on discovered vulnerabilities.

[0229] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A server cluster security control method combining multi-level encryption strategy and joint scheduling, characterized in that: The following steps are involved: S1. Perform sensitivity assessment on computing tasks and data within the cluster, and classify data into low-sensitivity data, medium-sensitivity data, and high-sensitivity data according to pre-defined rules. S2. Encrypt low-sensitivity data using the first encryption algorithm, encrypt medium-sensitivity data using the second encryption algorithm, and encrypt high-sensitivity data using the homomorphic encryption algorithm; S2.

1. Encrypt low-sensitivity data using the AES-128 encryption algorithm to generate the corresponding encrypted ciphertext; S2.

2. Encrypt the sensitive data using the AES-256 encryption algorithm to generate the corresponding encrypted ciphertext; S2.

3. Encrypt the highly sensitive data using a homomorphic encryption algorithm, wherein the homomorphic encryption algorithm is the Paillier encryption algorithm; S2.

4. All encryption processes generate and manage keys through a centralized key management system, including key lifecycle management, key storage, and distribution; S3, a joint scheduling mechanism based on reinforcement learning, dynamically adjusts the allocation of computing resources and task scheduling by obtaining real-time information about cluster load, computing task requirements, and data sensitivity; S3.

1. Collect the current cluster resource usage, including the load, memory and CPU usage of the computing nodes, as well as the computing requirements and priorities of the tasks; S3.

2. Based on deep Q-learning, dynamically adjust task scheduling and resource allocation strategies while satisfying resource allocation constraints. S3.

3. Based on the encryption level, data sensitivity, and computing resource requirements of the task, select a computing node with idle resources that can meet the encryption computing requirements; S3.

4. When scheduling computing tasks, consider resource utilization, load balancing, and security; S4. Deploy a zero-trust architecture in the cluster, requiring each request to undergo authentication and authorization, and implement fine-grained control over task permissions through role-based access control policies. S4.

1. Implement OAuth2.0 or OpenIDConnect-based authentication for every request within the cluster. S4.

2. Implement attribute-based access control to control access rights based on the sensitivity of the task, the identity and role information of the requester; S4.

3. Each access request is subject to permission verification; S4.

4. Dynamically adjust permissions so that permissions are updated in real time based on task requirements and access status during the task lifecycle; S5: Monitor computing tasks and data transmission in the cluster in real time, use deep learning algorithms to detect anomalies in cluster behavior, and identify potential security threats; S5.

1. Use the Prometheus monitoring system to monitor the cluster's compute node resource usage in real time and visualize it using Grafana. S5.

2. Use deep learning algorithms to model cluster behavior and identify abnormal patterns that deviate from normal behavior; S5.

3. Use model-based anomaly detection methods to automatically trigger alarms and notify administrators when cluster resources or task behaviors are abnormal. S5.

4. Generate real-time security reports based on the monitoring system and anomaly detection results to help administrators take timely countermeasures. S6. When a potential threat is detected, the security response mechanism is automatically executed to provide protection by isolating abnormal nodes and restricting resource access of malicious tasks.

2. The server cluster security control method combined with multi-level encryption strategy joint scheduling according to claim 1 is characterized in that: The sensitivity assessment of the data in the cluster in step S1 specifically includes the following steps: S1.

1. Collect data on computing tasks within the cluster and analyze the data type, content, and access frequency. S1.

2. Conduct a sensitivity assessment of the data using a trained machine learning model or specified statistical methods, and classify the data into low-sensitivity data, medium-sensitivity data, and high-sensitivity data based on the assessment results; S1.

3. Classify the data based on the assessment results and assign corresponding encryption algorithms and security policies according to the categories; S1.

4. Update the sensitivity level of data regularly or in real time based on changes in cluster resources, and dynamically adjust encryption policies.

3. The server cluster security control method combined with multi-level encryption strategy joint scheduling according to claim 1 is characterized in that: The S6 step security response mechanism includes: S6.

1. When a security threat is detected, the malicious node is automatically isolated and the computing tasks on the node are stopped to prevent the security threat from spreading. S6.

2. Limit resource access rights of affected nodes or tasks to reduce interference with normal computing tasks. S6.

3. Automatically adjust cluster access permissions based on the severity of abnormal behavior to prohibit unauthorized users or nodes from accessing sensitive data. S6.

4. Keep detailed records of all security incidents, generate audit logs, and provide them to administrators for subsequent analysis and investigation.

4. The server cluster security control method combined with multi-level encryption strategy joint scheduling according to claim 1 is characterized in that: The joint scheduling mechanism of reinforcement learning in step S3 optimizes the scheduling strategy through the following mathematical model: Among them, R(t) represents the reward value of the scheduling strategy, r i represents the available capacity of resource i, t j represents the demand of task j, x ij is a Boolean variable indicating whether task j is assigned to resource i, α and β are regularization coefficients, n is the number of resource nodes, m is the number of tasks, and d j is the resource requirement of task j, p j is the priority of task j.

Images