Intelligent security monitoring method and system in virtualization environment

By constructing a polymorphic behavior baseline and a three-layer mapping relationship, the cross-layer attack path in the virtualization environment is traced in reverse, which solves the problem of false alarms and false negatives in the monitoring strategy after virtual machine migration and realizes intelligent security monitoring of the entire life cycle of virtual machines.

CN123067992APending Publication Date: 2026-11-10BEIJING RUIYUAN XINDA INFORMATION TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611240458.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-17
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

In existing virtualization environments, monitoring strategies after virtual machine migration cannot effectively trace attack paths, leading to false alarms or missed alarms. Furthermore, there is a lack of overall tracing capabilities for cross-layer resource call chains, making it impossible to identify cross-layer attacks.

Method used

Construct a set of polymorphic behavior baselines, analyze the resource call dependencies between the virtual machine instance layer, the virtualization management layer, and the physical resource layer, trace back security events, mark cross-layer attack paths, and update monitoring policies based on feedback data.

Benefits of technology

It enables intelligent security monitoring of the entire lifecycle of virtual machines, reduces false positive rates, accurately identifies cross-layer attacks, dynamically adapts to environmental changes, and improves detection efficiency and response speed.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123067992A_ABST
    Figure CN123067992A_ABST
Patent Text Reader

Abstract

This invention relates to the field of security monitoring technology for virtualized environments, and in particular to an intelligent security monitoring method and system for virtualized environments. It constructs a set of polymorphic behavior baselines by extracting spatiotemporal trajectory features of virtual machine migration and establishes a three-layer mapping relationship between virtual machine instances, virtualization management, and physical resources. It reverse-tracks the resource call chain triggered by security events, marks cross-layer attack paths, locates threat virtual machine instances and attack stages, generates control commands, and updates the baselines and mapping relationship weights based on feedback to generate a full lifecycle monitoring strategy, thereby achieving accurate detection of cross-layer attacks and adaptive security monitoring.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of security monitoring technology in virtualized environments, and in particular to intelligent security monitoring methods and systems in virtualized environments. Background Technology

[0002] In cloud computing virtualization environments, security monitoring methods typically rely on intrusion detection systems deployed at the virtualization layer or host-based security agents to monitor virtual machine behavior. These systems establish behavioral baselines based on statistical distributions or machine learning models by collecting system call sequences, network traffic characteristics, or process activity data within the virtual machine, and use this to detect abnormal events that deviate from normal patterns. Some solutions add access control policies to the virtualization management layer to verify resource scheduling requests, but the monitoring scope is often limited to either the virtual machine instance layer or the virtualization management layer, lacking the ability to trace the entire cross-layer resource call chain. For virtual machine migration scenarios, existing monitoring strategies usually treat virtual machines as independent entities, re-initializing the baseline or using the static configuration before migration after migration, without fully considering the impact of changes in physical location and resource dependencies caused by migration on the behavioral model.

[0003] Conventional practices, with monitoring modules distributed across different layers of the virtualization environment, make it difficult to effectively correlate security events between layers. For example, the relationship between abnormal behavior at the virtual machine layer and unauthorized scheduling requests at the virtualization management layer, or cross-tenant access at the physical resource layer, is difficult to identify as a coherent attack process. This makes security mechanisms based on single-layer anomaly detection prone to missing cross-layer attacks exploiting inter-layer dependencies and unable to accurately trace attack paths. Existing behavioral baselines lack dynamic adaptability to migration contexts. After a virtual machine migrates, its operating environment, physical resource topology, and network paths all change, causing the original baseline model built based on static features or historical data to quickly become inaccurate, resulting in a large number of false positives or false negatives. Furthermore, existing systems typically only trigger resampling or retraining processes when responding to migration events, without integrating threat handling feedback with baseline updates in a closed loop. This prevents monitoring strategies from adaptively optimizing throughout the virtual machine's lifecycle, reducing the continuous effectiveness of security monitoring. Summary of the Invention

[0004] The present invention provides an intelligent security monitoring method and system in a virtualized environment, which can solve the problems in the prior art.

[0005] A first aspect of this invention provides an intelligent security monitoring method in a virtualized environment, comprising:

[0006] Extract physical location switching features and resource access mode change features from the spatiotemporal migration trajectories of virtual machines in cloud computing data centers, and construct a set of polymorphic behavior baselines that integrate migration contexts;

[0007] By analyzing the resource call dependencies between the virtual machine instance layer, the virtualization management layer, and the physical resource layer, a three-layer mapping relationship of the virtualization environment is constructed.

[0008] In the three-layer mapping relationship, the resource call chain triggered by the security event is traced in reverse. When the behavior characteristics of the virtual machine instance layer are detected to violate the polymorphic behavior baseline set, and the violation generates an unauthorized resource scheduling request in the virtualization management layer and is mapped as a cross-tenant shared resource access operation in the physical resource layer, the complete resource call chain from the virtual machine instance layer to the physical resource layer is marked as a cross-layer attack path.

[0009] Based on the polymorphic behavior baseline set and the cross-layer attack path, locate the set of virtual machine instances of security threats and their corresponding attack stage identifiers, generate control instructions, and collect feedback data after executing the control instructions;

[0010] Based on the migration events and threat handling results of virtual machine instances in the feedback data, the migration context and behavior boundaries of the virtual machine instances corresponding to the polymorphic behavior baseline set are updated, and the dependency weights of the three-layer mapping relationship are adjusted to generate a monitoring strategy covering the entire lifecycle of the virtual machine instance.

[0011] Extracting physical location switching features and resource access pattern change features from the spatiotemporal migration trajectories of virtual machines in cloud computing data centers, a set of polymorphic behavioral baselines integrating migration contexts is constructed, including:

[0012] Identify physical host switching events of virtual machine instances from the resource scheduling event sequence of the virtualization management layer;

[0013] Extract the physical host machine identifier and the timestamp of the switching time before and after the physical host machine switching event to form the physical location switching feature;

[0014] Extract the resource access operation types and access frequencies of the virtual machine instance within a preset time window before and after the switch, calculate the differences in the distribution of resource access operation types and the rate of change in access frequency before and after the physical host switch event, and constitute the resource access mode change characteristics.

[0015] The physical location switching feature, the resource access mode change feature, and the virtual machine instance identifier are associated and bound to form a feature tuple carrying the migration context.

[0016] For each virtual machine instance, based on the resource access mode change characteristics corresponding to different physical host identifiers in the feature tuple, multiple physical location-related behavioral baseline rules are established, and the multiple behavioral baseline rules together constitute the polymorphic behavioral baseline of a single virtual machine instance.

[0017] The polymorphic behavior baselines of all virtual machine instances within the cloud computing data center are aggregated to form the polymorphic behavior baseline set.

[0018] By resolving the resource call dependencies between the virtual machine instance layer, the virtualization management layer, and the physical resource layer, the three-layer mapping relationship of the virtualization environment is constructed, including:

[0019] Record the identifier of the resource request initiated by the virtual machine instance;

[0020] Extract virtual machine instance layer resource request records and corresponding target resource types, virtualization management layer resource scheduling records and corresponding physical resource location information, and physical resource layer access response records and corresponding physical resource identifiers;

[0021] Based on the time-series constraints of timestamps, records whose request initiation timestamps, scheduling completion timestamps, and response completion timestamps satisfy a sequentially increasing relationship are initially matched.

[0022] In the initially matched records, by mapping the requested target resource type to the physical resource location information, and mapping the physical resource location information to the physical resource identifier, a resource semantic mapping path from the virtual machine instance layer to the physical resource layer is established, and the record combination with the complete resource semantic mapping path is marked as a valid association;

[0023] Using the resource request identifier in the valid association as the cross-layer association key, establish an association index between the virtual machine instance layer resource request record, the virtualization management layer resource scheduling record, and the physical resource layer access response record;

[0024] Based on the associated index, a resource call dependency chain is constructed from the virtual machine instance layer resource request to the physical resource layer access response, and the resource call dependency chain forms the three-layer mapping relationship.

[0025] In the three-layer mapping relationship, the resource call chain triggered by the security event is traced backwards. When the behavior characteristics of the virtual machine instance layer violate the polymorphic behavior baseline set, and the violation generates an unauthorized resource scheduling request at the virtualization management layer and maps it to a cross-tenant shared resource access operation at the physical resource layer, the complete resource call chain from the virtual machine instance layer to the physical resource layer is marked as a cross-layer attack path, including:

[0026] The resource access operations monitored by the virtual machine instance layer are matched with the behavior baseline rules of the corresponding virtual machine instance in the polymorphic behavior baseline set. When the resource access operation exceeds the boundary range defined by the behavior baseline rule, it is marked as a violation and the resource request identifier is recorded.

[0027] By using the associated index in the three-layer mapping relationship, the resource scheduling record of the virtualization management layer corresponding to the resource request identifier is searched in reverse;

[0028] When the scheduling instruction permission identifier in the virtualization management layer resource scheduling record does not have a corresponding authorization record in the authorization configuration associated with the virtual machine instance identifier, the virtualization management layer resource scheduling record is marked as an unauthorized resource scheduling request.

[0029] Based on the associated index, reverse search is performed on the physical resource access response record corresponding to the unauthorized resource scheduling request at the physical resource layer, and the physical resource identifier and the tenant identifier that initiated the access operation are extracted from the physical resource access response record.

[0030] When the physical resource corresponding to the physical resource identifier is a shared resource shared by multiple tenants, and the tenant identifier is inconsistent with the tenant identifier to which the virtual machine instance identifier belongs, it is marked as a cross-tenant shared resource access operation.

[0031] The complete resource call chain from the violation at the virtual machine instance layer, the unauthorized resource scheduling request at the virtualization management layer, to the cross-tenant shared resource access operation at the physical resource layer is marked as the cross-layer attack path.

[0032] Based on the polymorphic behavior baseline set and the cross-layer attack path, the set of virtual machine instances of the security threat and the corresponding attack stage identifier are located, and control instructions are generated, including:

[0033] Extract the virtual machine instance identifier corresponding to the violation behavior of the virtual machine instance layer from the cross-layer attack path, and use it as the initial threat virtual machine instance;

[0034] Based on the associated index in the three-layer mapping relationship, find all resource request identifiers initiated by the initial threat virtual machine instance within a preset time window, and trace the associated physical resource layer access response records through the resource request identifiers;

[0035] Find all physical resource access response records in the physical resource layer access response records that were accessed within the preset time window for the accessed physical resource identifier. From all physical resource access response records, trace back the virtual machine instance identifier that initiated the access and mark the virtual machine instance identifier that initiated the access as an associated threat virtual machine instance.

[0036] The initial threat virtual machine instance and the associated threat virtual machine instance are merged to form the virtual machine instance set;

[0037] For each virtual machine instance identifier in the virtual machine instance set, the corresponding attack stage identifier is determined and bound based on the violation behavior, unauthorized resource scheduling request, and cross-tenant shared resource access operation in the cross-layer attack path;

[0038] When the attack phase is identified as the reconnaissance phase or the penetration phase, a resource access restriction instruction is generated; when the attack phase is identified as the privilege escalation phase or the lateral movement phase, a virtual machine isolation instruction is generated; when the attack phase is identified as the cross-tenant penetration phase or the persistence phase, a virtual machine termination instruction is generated.

[0039] For each virtual machine instance identifier in the virtual machine instance set, the corresponding attack stage identifier is determined and bound based on the violation behavior, unauthorized resource scheduling request, and cross-tenant shared resource access operation in the cross-layer attack path, including:

[0040] For each virtual machine instance identifier in the virtual machine instance set, extract the corresponding violation behavior, unauthorized resource scheduling request, and cross-tenant shared resource access operation from the cross-layer attack path;

[0041] Analyze the resource access operation types that exceed the polymorphic behavior baseline set in the violation behavior. When the resource access operation type is a network scanning operation, mark the attack stage as the reconnaissance stage. When the resource access operation type is a vulnerability exploitation operation, mark the attack stage as the penetration stage.

[0042] Parse the scheduling instruction permission identifier in the unauthorized resource scheduling request. When the scheduling instruction permission identifier involves virtual machine configuration modification permission, mark the attack phase identifier as the permission escalation phase. When the scheduling instruction permission identifier involves cross-virtual machine communication permission, mark the attack phase identifier as the lateral movement phase.

[0043] Parse the physical resource identifier in the cross-tenant shared resource access operation. When the physical resource type corresponding to the physical resource identifier is storage resource, mark the attack stage as cross-tenant penetration stage. When the physical resource type corresponding to the physical resource identifier is computing resource, mark the attack stage as persistence stage.

[0044] The attack phase identifier is bound to the corresponding virtual machine instance identifier.

[0045] Based on the migration events and threat handling results of virtual machine instances in the feedback data, the migration context and behavior boundaries of the virtual machine instances corresponding to the polymorphic behavior baseline set are updated, and the dependency weights of the three-layer mapping relationship are adjusted to generate a monitoring strategy covering the entire lifecycle of the virtual machine instance, including:

[0046] Parse the source physical node identifier and target physical node identifier in the migration event record, and query the corresponding physical resource layer configuration differences in the three-layer mapping relationship as migration context change features;

[0047] The matching degree between the successfully intercepted attack paths in the threat handling results and the cross-layer attack paths is statistically analyzed. When the matching degree exceeds a preset matching threshold, the resource access operation types involved in the attack paths are extracted as high-risk operation types.

[0048] For the behavior baseline rules of the corresponding virtual machine instance in the polymorphic behavior baseline set, the migration context change feature is added to the migration context field, and the high-risk operation type is added to the prohibited operation list to narrow the behavior boundary range.

[0049] The anomaly detection contribution rate of the virtual machine instance layer, virtualization management layer and physical resource layer in the threat handling result is calculated in the three-layer mapping relationship. The anomaly detection contribution rate is used as the dependency weight of the corresponding layer. When the dependency weight of one layer is lower than the preset weight threshold, the priority of the association index in the three-layer mapping relationship is reduced.

[0050] Based on the updated polymorphic behavior baseline set and the adjusted three-layer mapping relationship, corresponding monitoring policies are generated for the creation, operation, migration and destruction phases of virtual machine instances. The monitoring policies include behavior baseline rules, associated index priorities and anomaly detection trigger conditions for each phase.

[0051] A second aspect of this invention provides an intelligent security monitoring system in a virtualized environment, comprising:

[0052] The polymorphic baseline unit is used to extract physical location switching features and resource access mode change features in the spatiotemporal migration trajectory of virtual machines in cloud computing data centers, and to construct a set of polymorphic behavioral baselines that integrate migration context.

[0053] The three-layer mapping unit is used to construct the three-layer mapping relationship of the virtualization environment by resolving the resource call dependencies between the virtual machine instance layer, the virtualization management layer, and the physical resource layer.

[0054] The attack path marking unit is used to trace the resource call chain triggered by the security event in reverse in the three-layer mapping relationship. When the behavior characteristics of the virtual machine instance layer are detected to violate the polymorphic behavior baseline set, and the violation generates an unauthorized resource scheduling request in the virtualization management layer and is mapped as a cross-tenant shared resource access operation in the physical resource layer, the complete resource call chain from the virtual machine instance layer to the physical resource layer is marked as a cross-layer attack path.

[0055] The threat location and control unit is used to locate the set of virtual machine instances of security threats and their corresponding attack stage identifiers based on the polymorphic behavior baseline set and the cross-layer attack path, generate control instructions, and collect feedback data after executing the control instructions.

[0056] The baseline mapping adjustment unit is used to update the migration context and behavior boundary of the virtual machine instance corresponding to the polymorphic behavior baseline set based on the migration events and threat handling results of the virtual machine instance in the feedback data, and adjust the dependency weight of the three-layer mapping relationship to generate a monitoring strategy covering the entire life cycle of the virtual machine instance.

[0057] A third aspect of the present invention provides an electronic device, comprising:

[0058] processor;

[0059] Memory used to store processor-executable instructions;

[0060] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0061] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0062] The beneficial effects of this application are as follows:

[0063] The polymorphic behavior baseline set integrates physical location switching and resource access pattern changes in the virtual machine migration trajectory, adaptively characterizing the normal behavior profile in the virtualization environment and significantly reducing the false positive rate for legitimate migrations and resource accesses. The construction of a three-layer mapping relationship enables transparent resolution of resource call dependencies between the virtual machine instance layer, the virtualization management layer, and the physical resource layer. This allows security monitoring to penetrate the virtualization abstraction layer and reach the underlying physical resource interactions, avoiding the blind spots of traditional single-layer monitoring. The reverse tracing mechanism links violations of the polymorphic baseline with unauthorized scheduling requests from the virtualization management layer and cross-tenant shared access operations from the physical resource layer, completely capturing the cross-layer attack path from the virtual machine to the physical resource, effectively blocking covert penetration attacks targeting the virtualization environment.

[0064] Automatic labeling and threat localization of cross-layer attack paths can accurately identify infected sets of virtual machine instances and their attack stages, guiding the generation of targeted control commands. Migration events and threat handling results of virtual machine instances in the feedback data are used to dynamically update the migration context and behavioral boundaries of the polymorphic behavior baseline, enabling the baseline model to continuously adapt to environmental changes and preventing attackers from exploiting behavioral drift to bypass detection. The dependency weights of the three-layer mapping relationship are adjusted in real time based on feedback data, strengthening the monitoring of frequently attacked or high-risk links and improving the resource allocation efficiency of the detection model.

[0065] The monitoring strategy covers the entire lifecycle of virtual machine instances, maintaining consistent security protection capabilities from initial creation to migration or destruction. This approach reduces the need for manual intervention in security incident response, achieving self-healing and evolution of security policies through an automated feedback loop. The overall architecture balances detection accuracy and system performance, maintaining low-latency threat response even in scenarios with large-scale dynamic changes in virtualized environments, providing a hierarchical and evolvable intelligent security monitoring solution for cloud computing data centers. Attached Figure Description

[0066] Figure 1 This is a flowchart illustrating an intelligent security monitoring method in a virtualized environment.

[0067] Figure 2 A flowchart illustrating the monitoring strategy that covers the entire lifecycle of a virtual machine instance. Detailed Implementation

[0068] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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.

[0069] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0070] Figure 1 This is a flowchart illustrating the intelligent security monitoring method in a virtualized environment according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0071] Extract physical location switching features and resource access mode change features from the spatiotemporal migration trajectories of virtual machines in cloud computing data centers, and construct a set of polymorphic behavior baselines that integrate migration contexts;

[0072] By resolving the resource call dependencies between the virtual machine instance layer, the virtualization management layer, and the physical resource layer, a three-layer mapping relationship for the virtualization environment is constructed.

[0073] In the three-layer mapping relationship, the resource call chain triggered by the security event is traced in reverse. When the behavior characteristics of the virtual machine instance layer are detected to violate the polymorphic behavior baseline set, and the violation generates an unauthorized resource scheduling request in the virtualization management layer and is mapped as a cross-tenant shared resource access operation in the physical resource layer, the complete resource call chain from the virtual machine instance layer to the physical resource layer is marked as a cross-layer attack path.

[0074] Based on the polymorphic behavior baseline set and the cross-layer attack path, locate the set of virtual machine instances of security threats and their corresponding attack stage identifiers, generate control instructions, and collect feedback data after executing the control instructions;

[0075] Based on the migration events and threat handling results of virtual machine instances in the feedback data, the migration context and behavior boundaries of the virtual machine instances corresponding to the polymorphic behavior baseline set are updated, and the dependency weights of the three-layer mapping relationship are adjusted to generate a monitoring strategy covering the entire lifecycle of the virtual machine instance.

[0076] In one optional implementation, physical location switching features and resource access mode change features are extracted from the spatiotemporal trajectories of virtual machine migration in cloud computing data centers, and a set of polymorphic behavioral baselines that integrate migration contexts is constructed, including:

[0077] Identify physical host switching events of virtual machine instances from the resource scheduling event sequence of the virtualization management layer;

[0078] Extract the physical host machine identifier and the timestamp of the switching time before and after the physical host machine switching event to form the physical location switching feature;

[0079] Extract the resource access operation types and access frequencies of the virtual machine instance within a preset time window before and after the switch, calculate the differences in the distribution of resource access operation types and the rate of change in access frequency before and after the physical host switch event, and constitute the resource access mode change characteristics.

[0080] The physical location switching feature, the resource access mode change feature, and the virtual machine instance identifier are associated and bound to form a feature tuple carrying the migration context.

[0081] For each virtual machine instance, based on the resource access mode change characteristics corresponding to different physical host identifiers in the feature tuple, multiple physical location-related behavioral baseline rules are established, and the multiple behavioral baseline rules together constitute the polymorphic behavioral baseline of a single virtual machine instance.

[0082] The polymorphic behavior baselines of all virtual machine instances within the cloud computing data center are aggregated to form the polymorphic behavior baseline set.

[0083] The virtualization management layer continuously collects resource scheduling event sequences, filters and parses scheduling records involving virtual machine instance host changes, and identifies physical host switch events. When performing virtual machine migration operations, the virtualization management layer generates scheduling logs containing source host identifiers, target host identifiers, migration trigger reasons, and completion status. Real-time streaming parsing of these logs allows for accurate location of the occurrence time and involved host node information for each physical host switch event. During identification, it is necessary to distinguish between planned migrations (such as hot migrations triggered by load balancing) and unplanned migrations (such as disaster recovery migrations triggered by host failures). These two types of migrations are assigned different context labels when establishing subsequent behavior baselines to avoid misjudging normal disaster recovery migration behavior as abnormal.

[0084] After extracting the physical host switch event, record the physical host identifier before the switch. The physical host identifier after switching and the timestamp of the switch completion time. These three elements together constitute the physical location switching characteristic. The physical host identifier uses a node ID with unified coding within the data center, which can uniquely map to a specific physical server, thus supporting precise tracing from the virtual machine instance layer to the physical resource layer in subsequent cross-layer analysis. Timestamp The accuracy is maintained at the millisecond level to ensure time alignment with resource access operation records and to ensure that resource access sampling within the preceding and following windows is not affected by clock skew.

[0085] While acquiring physical location switching features, Based on this, preset time windows are captured forward and backward respectively. Records of resource access operations within the system. Preset time window. The value needs to be configured according to the business type of the virtual machine instance. For high-frequency transaction instances, a shorter window (e.g., 30 seconds) can be set, while for batch processing instances, a longer window (e.g., 5 minutes) can be set to ensure that a sufficient number of operation samples are collected for statistical analysis. Within the window, resource access operations are categorized and statistically analyzed by type, including but not limited to CPU calls, memory read / write, disk I / O, network transmission / reception, and cross-virtual machine communication. The frequency distribution of each type of operation is calculated by counting the number of times it appears within the window.

[0086] For the resource access operation type distribution of the window before and after the switch, calculate the difference metric between the two distributions. Let the operation types in the window before the switch be... The frequency is Switch the operation type in the window The frequency is The differences in the distribution of operation types can be quantified by the mean of the absolute values ​​of the frequency differences of each type, i.e. ,in Total number of operation types. Rate of change in access frequency. Defined as the ratio of the total frequency of window operations after switching to the total frequency of window operations before switching, minus 1. ,in and These represent the total operation frequency within the window before and after the switch. and Together, they constitute the characteristics of resource access pattern changes, which can characterize the degree of change in virtual machine behavior patterns before and after migration from two dimensions: type structure and total quantity.

[0087] Physical location switching features ( , , Characteristics of changes in resource access patterns , and all types , The unique identifier of the corresponding virtual machine instance. The migration context is then linked and bound to form feature tuples. These feature tuples are stored as structured records, with fields including virtual machine instance identifier, migration type label, source host identifier, target host identifier, migration timestamp, before and after window frequencies for each resource type, and calculated difference metrics. The migration type label in the feature tuple distinguishes between planned and unplanned migrations, enabling the establishment of separate behavioral rules for different migration scenarios during subsequent baseline building, thus avoiding interference between behavioral patterns triggered by different reasons.

[0088] For each virtual machine instance, aggregate all its historical feature tuples, categorized by target host identifier. Group the feature tuples, each different This corresponds to a set of behavioral observation samples after migration to this physical host. Within each set of samples, the frequency of each resource access operation type is... Perform statistical modeling and calculate the mean. and standard deviation The normal frequency range of this operation type on the physical host is determined by adding or subtracting a certain number of standard deviations from the mean, forming a behavioral baseline rule. This process is repeated for all operation types to obtain a set of behavioral baseline rules covering all resource access types. This set is then correlated with the physical host identifier. Binding establishes a location-dependent behavioral baseline for the virtual machine instance at its corresponding physical location. When a virtual machine instance has migrated to multiple different physical hosts, a separate location-dependent behavioral baseline is established for each historical target host. These multiple location-dependent behavioral baselines together constitute the polymorphic behavioral baseline of the virtual machine instance. The "polymorphic" characteristic of the polymorphic behavioral baseline is reflected in the different normal behavior patterns exhibited by the same virtual machine instance in different physical locations. For example, after migrating to a high-bandwidth node, the normal upper limit of network access frequency will increase accordingly, and after migrating to a storage-intensive node, the normal range of disk I / O will also be adjusted accordingly. A single static baseline cannot accurately describe this location dependency, while the polymorphic behavioral baseline achieves a precise characterization of location dependency through binding with the physical host identifier.

[0089] For newly deployed virtual machine instances or those with limited migration history, when the number of historical samples is insufficient to support statistical modeling, behavioral data from similar virtual machine instances on the same physical host can be used for cold start initialization. Subsequently, as the instance's own migration events accumulate, these data can be gradually replaced with instance-specific baseline parameters to ensure that the baseline rules remain effective throughout the entire lifecycle.

[0090] The polymorphic behavior baselines of all virtual machine instances within the cloud computing data center are aggregated to form a polymorphic behavior baseline set. This set is identified by virtual machine instances. The primary key is used for indexing, and each record contains all location-related behavioral baselines for that instance and its associated physical host identifier. In actual security monitoring, when a virtual machine instance is detected to be located on a host machine... At that time, the instance is searched from the polymorphic behavior baseline set. The corresponding behavioral baseline rules serve as a reference for real-time behavioral comparison, thereby achieving accurate baseline verification that matches the current physical location of the virtual machine. The polymorphic behavioral baseline set supports dynamic incremental updates. Whenever a new migration event occurs and the feature tuples are collected and calculated, the polymorphic behavioral baseline of the corresponding virtual machine instance is triggered for a local update, maintaining the overall timeliness and accuracy of the set.

[0091] In one optional implementation, the three-layer mapping relationship of the virtualization environment is constructed by resolving the resource call dependencies between the virtual machine instance layer, the virtualization management layer, and the physical resource layer, including:

[0092] Record the identifier of the resource request initiated by the virtual machine instance;

[0093] Extract virtual machine instance layer resource request records and corresponding target resource types, virtualization management layer resource scheduling records and corresponding physical resource location information, and physical resource layer access response records and corresponding physical resource identifiers;

[0094] Based on the time-series constraints of timestamps, records whose request initiation timestamps, scheduling completion timestamps, and response completion timestamps satisfy a sequentially increasing relationship are initially matched.

[0095] In the initially matched records, by mapping the requested target resource type to the physical resource location information, and mapping the physical resource location information to the physical resource identifier, a resource semantic mapping path from the virtual machine instance layer to the physical resource layer is established, and the record combination with the complete resource semantic mapping path is marked as a valid association;

[0096] Using the resource request identifier in the valid association as the cross-layer association key, establish an association index between the virtual machine instance layer resource request record, the virtualization management layer resource scheduling record, and the physical resource layer access response record;

[0097] Based on the associated index, a resource call dependency chain is constructed from the virtual machine instance layer resource request to the physical resource layer access response, and the resource call dependency chain forms the three-layer mapping relationship.

[0098] In a virtualization environment, the resource access relationships between the virtual machine instance layer, the virtualization management layer, and the physical resource layer are complex. The dispersed storage and heterogeneous formats of cross-layer data make direct tracing of dependencies difficult. Therefore, the first step in constructing this three-layer mapping is to assign a unique resource request identifier to each resource request initiated by a virtual machine instance. This identifier is carried throughout the request's lifecycle, spanning the entire chain from the virtual machine instance layer issuing the request, the virtualization management layer scheduling it, to the physical resource layer completing the response. The resource request identifier can be generated using a globally unique identifier (UUID) mechanism, combined with the unique identifier of the virtual machine instance. The system uses millisecond-level timestamps to construct the request, ensuring that request identifiers from different virtual machine instances do not collide in high-concurrency scenarios.

[0099] After completing the identification recording, raw log records are collected from three levels. At the virtual machine instance level, the target resource type of each resource request record is extracted, such as compute resources (vCPU), memory resources (vMEM), storage resources (vDisk), or network resources (vNIC), and this type is bound and stored with the corresponding resource request identifier. At the virtualization management layer, physical resource location information is extracted from resource scheduling records. This information describes the scheduling decisions of the hypervisor in mapping virtual resource requests to specific physical nodes, including the target physical host number, physical resource pool identifier, and scheduling policy type. At the physical resource layer, the physical resource identifier is extracted from access response records. This identifier precisely points to the actual accessed physical device or resource unit, such as the physical CPU core number, physical memory address range, disk partition identifier, or network interface card port identifier. Each of the three types of records must retain its own timestamp field during the collection phase, recorded as the request initiation timestamp. Scheduling completion timestamp and response completion timestamp These three timestamps form the core basis for subsequent temporal constraint matching.

[0100] When performing preliminary matching of three-level records based on timestamp-based temporal constraints, the following conditions must be met: The time-series increasing relationship. Specifically, for a resource request record at the virtual machine instance layer, a search is conducted in the scheduling record of the virtualization management layer to find a record that satisfies the resource request. Furthermore, the difference between the scheduling completion timestamp and the request initiation timestamp does not exceed the preset scheduling timeout threshold. The record, that is Further search the response records at the physical resource layer to find those that meet the requirements. and The records, among which The preset response timeout threshold is used. Time-series constraint filtering can effectively eliminate erroneous association candidates caused by log out-of-order, clock drift, or system latency, significantly reducing the number of noisy records in the initial matching candidate set, laying the foundation for subsequent semantic verification.

[0101] After initial matching, resource semantic mapping path verification is performed on candidate record combinations. The requested target resource type in the virtual machine instance layer record is mapped to the physical resource location information in the virtualization management layer record. This verifies whether the physical resource type pointed to by the scheduling decision is semantically consistent with the requested resource type. For example, a virtual memory request should correspond to the location information of a physical memory resource, not the location information of a storage device. Furthermore, the physical resource location information is mapped to the physical resource identifier in the physical resource layer record, verifying whether the actually accessed physical resource is indeed within the range of the physical node pointed to by the scheduling decision. Only when both levels of semantic mapping pass verification—that is, when the requested target resource type from the virtual machine instance layer can be completely and unambiguously mapped to the physical resource identifier from the physical resource layer via the physical resource location information of the virtualization management layer—is the record combination considered to have a complete resource semantic mapping path and marked as a valid association. For record combinations with broken semantic mapping (such as the node pointed to by the physical resource location information being inconsistent with the node where the physical resource identifier is located) or semantic type mismatches, they are included in the invalid candidate set and do not participate in subsequent index construction, but the original records are retained for anomaly analysis.

[0102] After valid association is confirmed, a relational index structure is established using the resource request identifier in the record combination as the cross-layer association key. The relational index adopts a storage method combining hash tables and inverted indexes: with the resource request identifier as the primary key, the storage pointers or record numbers of the corresponding virtual machine instance layer resource request records, virtualization management layer resource scheduling records, and physical resource layer access response records are uniformly mounted under the same index entry. This design allows for quick retrieval of the related records of the other two layers given a record at any layer, supporting bidirectional and even multidirectional cross-layer queries. Simultaneously, the relational index also includes a stored resource type label, physical host number, and a confidence score for valid association. The confidence score comprehensively considers the temporal constraint margin and the matching accuracy of the semantic mapping, providing a reference for subsequent weight assignment of dependency relationship links.

[0103] Based on the constructed association index, the three-layer records corresponding to each valid association are linked together into a resource call dependency chain. This chain starts with the resource request record at the virtual machine instance layer, passes through the resource scheduling record at the virtualization management layer as an intermediate node, and ends with the access response record at the physical resource layer, forming a directed chain structure. Multiple resource call dependency chains are organized according to the virtual machine instance and time dimensions, together constituting the three-layer mapping relationship of the virtualization environment. The three-layer mapping relationship is stored in the form of a directed acyclic graph (DAG), where nodes represent resource record entities at each layer, directed edges represent cross-layer dependencies, and the edge weights are initialized with the confidence scores of valid associations. This graph structure provides a complete data foundation for subsequent reverse tracing of resource call chains triggered by security events in the three-layer mapping relationship. This enables the rapid location of abnormal scheduling records at the virtualization management layer and actual access operations at the physical resource layer along the directed edges when abnormal behavior at the virtual machine instance layer is detected, supporting accurate marking of cross-layer attack paths and full-chain source tracing analysis of security threats.

[0104] In one optional implementation, the resource call chain triggered by the security event is traced back in the three-layer mapping relationship. When the behavior characteristics of the virtual machine instance layer are detected to violate the polymorphic behavior baseline set, and the violation generates an unauthorized resource scheduling request at the virtualization management layer and is mapped as a cross-tenant shared resource access operation at the physical resource layer, the complete resource call chain from the virtual machine instance layer to the physical resource layer is marked as a cross-layer attack path, including:

[0105] The resource access operations monitored by the virtual machine instance layer are matched with the behavior baseline rules of the corresponding virtual machine instance in the polymorphic behavior baseline set. When the resource access operation exceeds the boundary range defined by the behavior baseline rule, it is marked as a violation and the resource request identifier is recorded.

[0106] By using the associated index in the three-layer mapping relationship, the resource scheduling record of the virtualization management layer corresponding to the resource request identifier is searched in reverse;

[0107] When the scheduling instruction permission identifier in the virtualization management layer resource scheduling record does not have a corresponding authorization record in the authorization configuration associated with the virtual machine instance identifier, the virtualization management layer resource scheduling record is marked as an unauthorized resource scheduling request.

[0108] Based on the associated index, reverse search is performed on the physical resource access response record corresponding to the unauthorized resource scheduling request at the physical resource layer, and the physical resource identifier and the tenant identifier that initiated the access operation are extracted from the physical resource access response record.

[0109] When the physical resource corresponding to the physical resource identifier is a shared resource shared by multiple tenants, and the tenant identifier is inconsistent with the tenant identifier to which the virtual machine instance identifier belongs, it is marked as a cross-tenant shared resource access operation.

[0110] The complete resource call chain from the violation at the virtual machine instance layer, the unauthorized resource scheduling request at the virtualization management layer, to the cross-tenant shared resource access operation at the physical resource layer is marked as the cross-layer attack path.

[0111] In the three-layer mapping relationship, the resource call chain triggered by the security event is traced backward. Relying on the pre-built association index structure between the virtual machine instance layer, the virtualization management layer, and the physical resource layer, the resource access operations are compared and marked layer by layer. The entire reverse tracing process starts with the behavior detection of the virtual machine instance layer, extends down the call chain to the physical resource layer, and finally forms a complete cross-layer attack path marking result.

[0112] At the virtual machine instance layer, resource access operations collected during runtime are matched in real time. A set of behavioral baseline rules is maintained for each virtual machine instance within the polymorphic behavior baseline set. These rules include boundary constraints across multiple dimensions, such as the allowed set of resource access types, upper and lower bounds on access frequency, access target range, and operation permission level. When the attribute value of a resource access operation exceeds the aforementioned boundary range, the operation is determined to be a violation. Upon a violation, the corresponding resource request identifier is immediately extracted. This identifier is unique throughout the entire call chain and is used to link the processing records of each layer during subsequent reverse lookup. The resource request identifier is typically generated based on a combined hash of the virtual machine instance identifier, request initiation timestamp, and operation sequence number to ensure no conflicts occur in high-concurrency scenarios.

[0113] After obtaining the resource request identifier, the resource scheduling record corresponding to that resource request identifier in the virtualization management layer is searched backwards using the pre-established association index in the three-layer mapping relationship. The association index, using the resource request identifier as the key, records information such as the sequence of scheduling instructions triggered by the request in the virtualization management layer, the permission identifier of the scheduling instructions, and the scheduling execution time. After obtaining the resource scheduling record in the virtualization management layer, the scheduling instruction permission identifier is extracted and compared with the authorization configuration associated with the virtual machine instance identifier. The authorization configuration is stored in a separate permission management repository, recording the set of legal scheduling operation permissions granted to each virtual machine instance. If the scheduling instruction permission identifier does not have a corresponding authorization record in the authorization configuration, the resource scheduling record is marked as an unauthorized resource scheduling request, and the corresponding scheduling instruction identifier is recorded for subsequent traceability.

[0114] After an unauthorized resource scheduling request is confirmed, the system continues to reverse-engineer the associated index to find the physical resource access response record generated at the physical resource layer for that scheduling request. The physical resource access response record contains fields such as the identifier of the accessed physical resource, the type of access operation, the tenant identifier at the time the operation was initiated, and the access result status. The physical resource identifier is used to uniquely locate a specific physical resource entity in the data center, such as a physical storage volume, physical network port, or physical compute node. The tenant identifier indicates the actual tenant affiliation of the entity that initiated the access operation; this identifier originates from the context information carried by the virtualization management layer when issuing the scheduling command.

[0115] After extracting the physical resource identifier and tenant identifier, two parallel decision-making logics are executed. First, the physical resource sharing attribute table is queried to confirm whether the physical resource corresponding to the physical resource identifier is shared by multiple tenants. The sharing attribute table records the tenant affiliation list for each physical resource. If the affiliation list contains two or more different tenant identifiers, the physical resource is identified as a multi-tenant shared resource. Second, the tenant identifier in the physical resource access response record is compared with the tenant identifier to which the virtual machine instance identifier that initiated the violation belongs. The affiliation relationship between the virtual machine instance identifier and the tenant identifier is stored in the tenant asset management database; the corresponding tenant affiliation can be queried through the instance identifier. If the two tenant identifiers are inconsistent, it indicates that the access operation crossed the tenant boundary and belongs to a different tenant than the virtual machine instance that initiated the request.

[0116] When both of the above conditions are met simultaneously—that is, the physical resource is a multi-tenant shared resource and the tenant identifier of the access operation is inconsistent with the tenant identifier to which the virtual machine instance belongs—the physical resource access response record is marked as a cross-tenant shared resource access operation. At this point, a complete three-layer resource call chain is formed, from the violation record at the virtual machine instance layer, the unauthorized resource scheduling request record at the virtualization management layer, to the cross-tenant shared resource access operation record at the physical resource layer. All node information, call sequence relationships, and marking statuses at each layer in this chain are integrated into a cross-layer attack path record and written to the attack path database.

[0117] Cross-layer attack path records contain several key fields, specifically including a unique identifier for the attack path, identifiers of the virtual machine instances involved, a description of the type of violation, identifiers of unauthorized scheduling instructions, identifiers of the accessed physical resources, tenant identifier pairs crossed, and timestamp sequences of events at each layer. These fields provide complete contextual information for subsequent threat localization, attack phase identification, and response control instruction generation. In scenarios where multiple cross-layer attack paths coexist, it is also necessary to perform correlation analysis on the paths to determine whether there are combined attack behaviors initiated by multiple virtual machine instances in concert. By comparing the overlap of physical resource identifiers and the degree of overlap of time windows in each path, potential coordinated attack sets can be identified.

[0118] To improve the efficiency of reverse tracing, the associated index is organized using a hierarchical inverted index structure. At the virtual machine instance level, a forward index is established using the resource request identifier as the primary key; at the virtualization management layer, the scheduling instruction identifier is used as the primary key, and a reverse mapping from resource request identifiers to scheduling instruction identifiers is established; at the physical resource layer, the physical resource access response record identifier is used as the primary key, and a reverse mapping from scheduling instruction identifiers to response record identifiers is established. Through this hierarchical inverted index structure, only three index lookups are needed to complete the entire link from the virtual machine instance layer to the physical resource layer during reverse tracing, meeting the performance requirements for real-time detection in large-scale virtualization environments.

[0119] In actual deployment, scheduling records at the virtualization management layer and access response records at the physical resource layer are continuously written in a streaming manner. Updates to the associated indexes are synchronized with record writing to ensure the latest link status can be obtained during reverse tracing. For record loss due to network latency or out-of-order logs, a timeout compensation mechanism is set up. If the corresponding layer's record is not received within a preset waiting time, the link node is marked as pending completion, and the link is automatically pieced together upon the arrival of subsequent records, avoiding missed detections of cross-layer attack paths due to temporary data loss.

[0120] In one optional implementation, based on the polymorphic behavior baseline set and the cross-layer attack path, locating the set of virtual machine instances of the security threat and their corresponding attack stage identifiers, and generating control instructions, includes:

[0121] Extract the virtual machine instance identifier corresponding to the violation behavior of the virtual machine instance layer from the cross-layer attack path, and use it as the initial threat virtual machine instance;

[0122] Based on the associated index in the three-layer mapping relationship, find all resource request identifiers initiated by the initial threat virtual machine instance within a preset time window, and trace the associated physical resource layer access response records through the resource request identifiers;

[0123] Find all physical resource access response records in the physical resource layer access response records that were accessed within the preset time window for the accessed physical resource identifier. From all physical resource access response records, trace back the virtual machine instance identifier that initiated the access and mark the virtual machine instance identifier that initiated the access as an associated threat virtual machine instance.

[0124] The initial threat virtual machine instance and the associated threat virtual machine instance are merged to form the virtual machine instance set;

[0125] For each virtual machine instance identifier in the virtual machine instance set, the corresponding attack stage identifier is determined and bound based on the violation behavior, unauthorized resource scheduling request, and cross-tenant shared resource access operation in the cross-layer attack path;

[0126] When the attack phase is identified as the reconnaissance phase or the penetration phase, a resource access restriction instruction is generated; when the attack phase is identified as the privilege escalation phase or the lateral movement phase, a virtual machine isolation instruction is generated; when the attack phase is identified as the cross-tenant penetration phase or the persistence phase, a virtual machine termination instruction is generated.

[0127] The virtual machine instance (VM) identifiers corresponding to violations are extracted from cross-layer attack paths, using the violation records at the VM instance layer within the cross-layer attack path as the entry point. Each cross-layer attack path carries a VM instance identifier field that triggered the violation when marked. The VM instance identifiers in this field are extracted to form an initial set of threat VM instances. It is important to note that the same cross-layer attack path involves coordinated violations by multiple VM instances at different time points. Therefore, it is necessary to traverse all violation records in the path, deduplicate and merge the extracted VM instance identifiers to ensure the integrity of the initial set of threat VM instances. The initial set of threat VM instances is the starting point for subsequent correlation and diffusion analysis, and its accuracy directly affects the coverage of the final set of VM instances.

[0128] Based on the association index in the three-layer mapping relationship, all resource request identifiers initiated by the initial threat virtual machine instance within a preset time window are retrieved. During construction, the three-layer mapping relationship maintains an association index structure for each virtual machine instance in the virtual machine instance layer, with the virtual machine instance identifier as the key and a list of resource request identifiers as values. This index structure stores the correspondence between the initiation time of resource requests and the request identifiers in chronological order. Using the preset time window as a filter, all resource request identifiers falling within that time window are retrieved from the association index. Subsequently, the resource request identifiers are matched against the access response record table in the physical resource layer to obtain the physical resource layer access response record corresponding to each resource request. The record content includes the accessed physical resource identifier, the virtual machine instance identifier of the initiator, the access timestamp, and the access result status. This step extends the behavioral trajectory of the initial threat virtual machine instance from the virtual machine instance layer down to the physical resource layer, forming a set of physical resource access response records that can be used for subsequent correlation analysis.

[0129] After obtaining the physical resource layer access response record set, a list of accessed physical resource identifiers is extracted from it. For each physical resource identifier, all access response records that have accessed that physical resource are retrieved within a preset time window, forming an expanded physical resource access response record set. This expanded set covers the behavior records of other virtual machine instances that accessed the same physical resource as the initial threat virtual machine instance within the same time window. The virtual machine instance identifiers that initiated the access are traced back from the expanded set. After excluding the initial threat virtual machine instance itself, the remaining virtual machine instance identifiers are marked as associated threat virtual machine instances. The identification logic of associated threat virtual machine instances is based on the attack propagation medium of shared physical resources. When conducting cross-tenant penetration, attackers often use shared storage, shared network interfaces, or shared compute nodes as springboards for lateral movement. Therefore, virtual machine instances that share the same physical resource as the initial threat virtual machine instance and have access behavior within a similar time window have a high probability of being associated threats.

[0130] The initial set of threat virtual machine instances and the set of associated threat virtual machine instances were merged, and a deduplication operation was performed to obtain the final set of virtual machine instances. This set comprehensively covers virtual machine instances that directly participated in the violations in this security incident, as well as potential threat virtual machine instances that were associated through physical resource sharing paths, providing a complete scope of analysis objects for determining the identification of subsequent attack phases.

[0131] For each virtual machine instance identifier in the virtual machine instance set, the attack stage identifier corresponding to the virtual machine instance is determined by combining the violation behavior type, unauthorized resource scheduling request content, and cross-tenant shared resource access operation characteristics recorded in the cross-layer attack path. The attack phase is determined according to the following logic: If the violation behavior of the virtual machine instance in the cross-layer attack path is a slight deviation from the resource access pattern in the polymorphic behavior baseline set, and no unauthorized resource scheduling requests are detected, then the attack phase is identified as the reconnaissance phase; if the violation behavior is a probing access to other virtual machine instances or physical resources, and low-privilege unauthorized scheduling requests appear in the virtualization management layer, then it is identified as the penetration phase; if the violation behavior is accompanied by unauthorized scheduling requests for high-privilege resources in the virtualization management layer, and the request attempts to obtain resource control capabilities beyond the original privileges of the virtual machine instance, then it is identified as the privilege escalation phase; if the violation behavior has been mapped to cross-tenant shared resource access operations at the physical resource layer, and there are records of accessing other tenant virtual machine instances on the same physical host, then it is identified as the lateral movement phase; if the cross-tenant shared resource access operation has been successfully completed and the target is the core resources of other tenants, then it is identified as the cross-tenant penetration phase; if the violation behavior leaves persistent access records at the physical resource layer, and the behavioral boundaries of the corresponding virtual machine instance in the polymorphic behavior baseline set have been continuously breached, then it is identified as the persistence phase. Each virtual machine instance identifier is bound to its corresponding attack stage identifier, forming a mapping table between virtual machine instance identifiers and attack stage identifiers, which is used by the control command generation stage.

[0132] Control commands are generated based on the severity of the attack phase, implementing differentiated response strategies. When the attack phase is identified as reconnaissance or penetration, the corresponding virtual machine instance is still in the early stages of the attack, and actual damage has not yet spread. Resource access restriction commands are generated in this case. These commands include: rate limiting outbound network access requests to the virtual machine instance, downgrading its access permissions to specific types of physical resources to read-only mode, and adding additional authentication verification tags to the virtual machine instance's resource requests in the Layer 3 mapping relationship to ensure that subsequent resource requests undergo enhanced auditing.

[0133] When the attack phase is identified as either the privilege escalation phase or the lateral movement phase, the attack has already had a substantial impact on the virtualization management layer and has begun to spread laterally, generating a virtual machine isolation command. Once triggered, the virtual machine isolation command switches the target virtual machine instance's network interface to an isolated network area at the virtualization management layer, blocking its communication links with other virtual machine instances and external networks. Simultaneously, it freezes all resource scheduling permissions at the virtualization management layer, preventing it from initiating new resource requests, but retains the virtual machine instance's running state for subsequent forensic analysis.

[0134] When the attack phase is identified as the cross-tenant penetration phase or the persistence phase, the attack has already caused cross-tenant data access or established a persistent access channel at the physical resource layer, posing the highest level of harm, and a virtual machine termination command is generated. When the virtual machine termination command is executed, the following operations are performed sequentially: at the physical resource layer, the virtual machine instance's access authorization to all shared physical resources is revoked; at the virtualization management layer, all scheduling credentials of the virtual machine instance are cancelled; finally, at the virtual machine instance layer, the running process of the virtual machine instance is forcibly terminated, and its runtime memory image and disk state snapshot are saved to an isolated storage area to support subsequent attack tracing and forensics. All three types of control commands, after generation, carry the corresponding virtual machine instance identifier and attack phase identifier as metadata to ensure that the command execution results accurately correspond to subsequent feedback data collection stages.

[0135] In one optional implementation, for each virtual machine instance identifier in the virtual machine instance set, determining and binding the corresponding attack stage identifier based on the violation behavior, unauthorized resource scheduling request, and cross-tenant shared resource access operation in the cross-layer attack path includes:

[0136] For each virtual machine instance identifier in the virtual machine instance set, extract the corresponding violation behavior, unauthorized resource scheduling request, and cross-tenant shared resource access operation from the cross-layer attack path;

[0137] Analyze the resource access operation types that exceed the polymorphic behavior baseline set in the violation behavior. When the resource access operation type is a network scanning operation, mark the attack stage as the reconnaissance stage. When the resource access operation type is a vulnerability exploitation operation, mark the attack stage as the penetration stage.

[0138] Parse the scheduling instruction permission identifier in the unauthorized resource scheduling request. When the scheduling instruction permission identifier involves virtual machine configuration modification permission, mark the attack phase identifier as the permission escalation phase. When the scheduling instruction permission identifier involves cross-virtual machine communication permission, mark the attack phase identifier as the lateral movement phase.

[0139] Parse the physical resource identifier in the cross-tenant shared resource access operation. When the physical resource type corresponding to the physical resource identifier is storage resource, mark the attack stage as cross-tenant penetration stage. When the physical resource type corresponding to the physical resource identifier is computing resource, mark the attack stage as persistence stage.

[0140] The attack phase identifier is bound to the corresponding virtual machine instance identifier.

[0141] For each virtual machine instance identifier in the virtual machine instance set, three types of attack elements associated with that instance are precisely extracted from the marked cross-layer attack paths: violation behavior, unauthorized resource scheduling requests, and cross-tenant shared resource access operations. The extraction process uses the virtual machine instance identifier as the index key, performing precise matching within the structured records of the cross-layer attack paths to ensure that the attack elements corresponding to each virtual machine instance are not confused with data from other instances. After extraction, the three types of attack elements are stored in the temporary attack context buffer of the corresponding instance for use in subsequent parsing stages. The core purpose of this step is to decompose the global attack information in the cross-layer attack paths at the instance level, providing accurate data input for subsequent layer-by-layer attack stage determination.

[0142] When analyzing violations, the focus is on resource access operation types that exceed the boundaries defined by the polymorphic behavior baseline set. The polymorphic behavior baseline set maintains a whitelist of legitimate operation types associated with each virtual machine instance's current migration context, along with corresponding frequency threshold ranges. When the actual behavioral characteristics of a certain type of resource access operation exceed the whitelist range or surpass the frequency threshold, the operation is identified as a violation and recorded in the cross-layer attack path. During the attack phase determination stage, if the resource access operation type analyzed from the violation is a network scanning operation—that is, a large-scale probe of the external or internal network address space, including port scanning, protocol probing, and service discovery—then the attack phase of that virtual machine instance is marked as the reconnaissance phase. The reconnaissance phase means that the attacker is in the target environment information gathering stage, has not yet carried out active intrusion, but has already posed a potential threat to the network topology of the cloud computing data center. If the analyzed operation type is a vulnerability exploitation operation—that is, constructing malicious input against known or unknown software vulnerabilities and attempting to gain unauthorized code execution capabilities—then the attack phase is marked as the penetration phase. The penetration phase indicates that the attacker has shifted from external probing to proactive attacks, and the threat level is significantly higher than that of the reconnaissance phase, requiring a higher priority response strategy.

[0143] When parsing unauthorized resource scheduling requests, the focus is on analyzing the permission identifier field carried in the scheduling command. An unauthorized resource scheduling request refers to a scheduling command initiated by a virtual machine instance within the virtualization management layer that is unauthorized or exceeds the preset permission scope. The scheduling command permission identifier is an encoded description of the operation permission declared or attempted by the command, typically existing in the metadata of the scheduling request in the form of a permission bitmask or permission enumeration value. When the scheduling command permission identifier involves virtual machine configuration modification permissions—such as attempting to modify core configuration items like CPU quotas, memory limits, network interface bindings, or security group rules—the attack phase is marked as a privilege escalation phase. A privilege escalation phase indicates that the attacker has attempted to expand their control at the virtualization management layer, manipulating virtual machine configurations to circumvent security restrictions or create conditions for subsequent attacks. When the scheduling command permission identifier involves cross-virtual machine communication permissions—that is, attempting to establish unauthorized direct communication channels between different virtual machine instances to bypass tenant isolation mechanisms for data transfer or command issuance—the attack phase is marked as a lateral movement phase. The lateral movement phase indicates that the attacker has breached the boundaries of a single virtual machine instance and is spreading to other tenants or instances within the cloud computing data center. It is a critical node in the cross-tenant attack chain and requires immediate isolation response.

[0144] When analyzing cross-tenant shared resource access operations, the physical resource identifier is the core analytical dimension. The physical resource identifier is a unique code for each type of shared physical resource in the physical resource layer, precisely pointing to a specific physical storage device, physical compute node, or physical network device. By querying the physical resource type registry using the physical resource identifier, it can be determined whether the accessed shared physical resource belongs to storage or compute resources. When the physical resource type corresponding to the physical resource identifier is a storage resource—such as distributed block storage, shared file system, or object storage backend—the attack phase is marked as the cross-tenant penetration phase. The cross-tenant penetration phase means that the attacker has successfully accessed storage space originally belonging to other tenants, posing a substantial risk of data theft, data tampering, or data injection, and is one of the most dangerous attack scenarios in cloud computing security. When the physical resource type corresponding to the physical resource identifier is a compute resource—such as a physical CPU core, GPU compute unit, or FPGA acceleration resource—the attack phase is marked as the persistence phase. The persistence phase indicates that the attacker is using cross-tenant physical compute resources to maintain their attack capabilities or implant persistent malicious loads. Even if the original compromised virtual machine instance is destroyed, the attack capabilities can still be continued through residual code or configuration at the physical compute resource level.

[0145] After determining the attack phase across the three dimensions mentioned above, all identified attack phase identifiers are bound to their corresponding virtual machine instance identifiers, forming an instance-level attack phase mapping record. The binding operation uses structured key-value pairs, with the virtual machine instance identifier as the key and the list of attack phase identifiers as the value. This supports scenarios where a single virtual machine instance carries multiple attack phase identifiers within the same attack event. For example, a virtual machine instance simultaneously marked as being in the penetration and privilege escalation phases indicates that it attempted to expand privileges at the virtualization management layer while simultaneously exploiting the vulnerability. This ability to concurrently mark multiple phases more accurately reflects the overlapping characteristics of attack behaviors in complex attack chains, avoiding the loss of attack phase information due to single-phase determination logic. The bound instance-level attack phase mapping record will serve as direct input for subsequent control command generation, ensuring that response actions for different attack phases accurately correspond to specific virtual machine instances, achieving refined security responses based on attack phase awareness.

[0146] In one optional implementation, based on the migration events and threat handling results of virtual machine instances in the feedback data, the migration context and behavior boundaries of the virtual machine instances corresponding to the polymorphic behavior baseline set are updated, and the dependency weights of the three-layer mapping relationship are adjusted to generate a monitoring strategy covering the entire lifecycle of the virtual machine instance, including:

[0147] Parse the source physical node identifier and target physical node identifier in the migration event record, and query the corresponding physical resource layer configuration differences in the three-layer mapping relationship as migration context change features;

[0148] The matching degree between the successfully intercepted attack paths in the threat handling results and the cross-layer attack paths is statistically analyzed. When the matching degree exceeds a preset matching threshold, the resource access operation types involved in the attack paths are extracted as high-risk operation types.

[0149] For the behavior baseline rules of the corresponding virtual machine instance in the polymorphic behavior baseline set, the migration context change feature is added to the migration context field, and the high-risk operation type is added to the prohibited operation list to narrow the behavior boundary range.

[0150] The anomaly detection contribution rate of the virtual machine instance layer, virtualization management layer and physical resource layer in the threat handling result is calculated in the three-layer mapping relationship. The anomaly detection contribution rate is used as the dependency weight of the corresponding layer. When the dependency weight of one layer is lower than the preset weight threshold, the priority of the association index in the three-layer mapping relationship is reduced.

[0151] Based on the updated polymorphic behavior baseline set and the adjusted three-layer mapping relationship, corresponding monitoring policies are generated for the creation, operation, migration and destruction phases of virtual machine instances. The monitoring policies include behavior baseline rules, associated index priorities and anomaly detection trigger conditions for each phase.

[0152] like Figure 2 As shown, the method includes:

[0153] The analysis of feedback data begins with migration event records. Each migration event record contains source and target physical node identifiers. These identifiers are used to locate the corresponding physical resource layer configuration items in the three-layer mapping relationship. Differences between the two nodes in dimensions such as CPU architecture type, memory page size policy, storage I / O scheduler type, network bandwidth quota, and NUMA topology are extracted. This set of differences is used as the migration context change feature for this migration. The granularity of the migration context change feature directly affects the update accuracy of the behavioral baseline: if the storage I / O scheduler types of the source and target nodes are different, the disk access latency distribution generated by the virtual machine instance after migration will undergo a systematic shift. If this difference is not included in the migration context field, the baseline rules will continue to trigger false alarms. Therefore, the completeness of configuration differences is the core requirement for extracting migration context change features.

[0154] The statistics on threat handling results revolve around attack path matching. Successfully intercepted attack paths are compared segment by segment with cross-layer attack paths obtained by tracing back the three-layer mapping relationship. The overlap ratio of the two in the resource call node sequence is calculated to obtain a matching metric. Let the matching metric be... The preset matching threshold is ,when If the interception result is deemed to have high confidence in confirming a cross-layer attack path, then the resource access operation types involved in that attack path are extracted and marked as high-risk operation types. The extraction of high-risk operation types is refined to specific operation categories, such as cross-tenant shared memory mapping operations, unauthorized vCPU affinity modification operations, and DMA operations that bypass the hypervisor to directly access physical devices. These operation types will be added to a prohibited operation list in subsequent steps as a basis for narrowing behavior boundaries.

[0155] The update of behavior baseline rules is performed on the corresponding virtual machine instance entries in the polymorphic behavior baseline set. Migration context change characteristics are written into the migration context field of this entry, enabling subsequent detection logic to perceive the differences in the current physical environment when evaluating virtual machine instance behavior, avoiding misjudging normal behavior caused by changes in physical node configuration as abnormal. Simultaneously, the extracted high-risk operation types are appended to the prohibited operation list of this entry, and a shrinking operation is performed on the behavior boundary range: for numerical behavior indicators (such as the maximum number of resource requests per unit time), the upper bound of its allowed range is compressed inward according to a preset shrinking ratio; for categorical behavior indicators (such as the enumerated set of allowed resource types), access categories that semantically overlap with high-risk operation types are removed. After shrinking the behavior boundary, the compliant behavior space of virtual machine instances becomes smaller, and the detection sensitivity is correspondingly improved, enabling the capture of potential lateral penetration attempts at an earlier stage.

[0156] The dependency weights are adjusted based on the anomaly detection contribution rate of each layer in the threat handling results. The number of valid alerts triggered by the virtual machine instance layer, virtualization management layer, and physical resource layer in this round of threat handling is counted. Using the sum of the valid alert counts of the three layers as the denominator and the number of valid alerts of each layer as the numerator, the anomaly detection contribution rate of each layer is calculated and directly used as the dependency weight of the corresponding layer in the three-layer mapping relationship. Let the dependency weight of the virtual machine instance layer be... The dependency weight of the virtualization management layer is The dependency weight of the physical resource layer is All three conditions are met. When the dependency weight of a certain level is lower than a preset weight threshold... At the same time, the priority of the associated index of this level in the three-level mapping relationship is reduced, so that in the reverse tracing process of subsequent resource call chain, the associated nodes of the high-weight level are traversed first, thereby reducing the inefficient index query overhead, and at the same time, the detection resources are concentrated on the level with stronger detection capabilities.

[0157] The monitoring strategy is generated within the framework of the entire lifecycle of a virtual machine instance, outputting corresponding monitoring strategies for the creation, runtime, migration, and destruction phases. The monitoring strategy for the creation phase includes initial behavioral baseline rules and strict anomaly detection trigger conditions: since newly created virtual machine instances have not yet accumulated sufficient historical behavioral data, the behavioral baseline uses the group statistical baseline of similar instances as the initial value, and the anomaly detection trigger threshold is set at a low level to capture configuration injection attacks and image tampering during instance initialization; the priority of the associated index is sorted according to the updated dependency weights, prioritizing the index of the level with the highest contribution rate. The monitoring strategy for the runtime phase uses individualized behavioral baseline rules after migration context updates and behavioral boundary contraction. The anomaly detection trigger conditions combine a prohibited operation list and a behavioral boundary range for dual judgment. When the real-time behavioral indicators of a virtual machine instance exceed the behavioral boundary or touch any operation type in the prohibited operation list, an alarm is immediately triggered and reverse tracing of the resource call chain is initiated.

[0158] The monitoring strategy during the migration phase builds upon the runtime phase strategy by overlaying migration context-aware logic: upon detecting a migration event, it automatically switches to the migration context field corresponding to the target physical node. Within a preset migration adaptation window, it dynamically relaxes behavioral boundaries, allowing temporary behavioral drift caused by physical environment switching to pass detection. Simultaneously, it maintains zero tolerance for cross-tenant resource access operations, unaffected by the relaxed adaptation window strategy. The monitoring strategy during the destruction phase focuses on data residue risks and unauthorized resource holding behaviors during resource release: after a virtual machine instance issues a destruction request, it continuously monitors its resource occupancy status at the physical resource layer. If residual shared memory mappings or unreleased device handles are detected after destruction, this behavior is marked as abnormal, triggering a forced reclamation operation at the physical resource layer. Simultaneously, the corresponding attack phase identifier is updated to post-destruction persistence. These four phase monitoring strategies together constitute a complete monitoring system covering the entire lifecycle of a virtual machine instance. The strategies at each phase are linked through the dependency weights of a three-layer mapping relationship, ensuring that security events occurring at any lifecycle stage can be accurately traced to the corresponding resource call chain and attack path.

[0159] A second aspect of this invention provides an intelligent security monitoring system in a virtualized environment, comprising:

[0160] The polymorphic baseline unit is used to extract physical location switching features and resource access mode change features in the spatiotemporal migration trajectory of virtual machines in cloud computing data centers, and to construct a set of polymorphic behavioral baselines that integrate migration context.

[0161] The three-layer mapping unit is used to construct the three-layer mapping relationship of the virtualization environment by resolving the resource call dependencies between the virtual machine instance layer, the virtualization management layer, and the physical resource layer.

[0162] The attack path marking unit is used to trace the resource call chain triggered by the security event in reverse in the three-layer mapping relationship. When the behavior characteristics of the virtual machine instance layer are detected to violate the polymorphic behavior baseline set, and the violation generates an unauthorized resource scheduling request in the virtualization management layer and is mapped as a cross-tenant shared resource access operation in the physical resource layer, the complete resource call chain from the virtual machine instance layer to the physical resource layer is marked as a cross-layer attack path.

[0163] The threat location and control unit is used to locate the set of virtual machine instances of security threats and their corresponding attack stage identifiers based on the polymorphic behavior baseline set and the cross-layer attack path, generate control instructions, and collect feedback data after executing the control instructions.

[0164] The baseline mapping adjustment unit is used to update the migration context and behavior boundary of the virtual machine instance corresponding to the polymorphic behavior baseline set based on the migration events and threat handling results of the virtual machine instance in the feedback data, and adjust the dependency weight of the three-layer mapping relationship to generate a monitoring strategy covering the entire life cycle of the virtual machine instance.

[0165] A third aspect of the present invention provides an electronic device, comprising:

[0166] processor;

[0167] Memory used to store processor-executable instructions;

[0168] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0169] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0170] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0171] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent security monitoring in a virtualized environment, characterized in that, include: Extract physical location switching features and resource access mode change features from the spatiotemporal migration trajectories of virtual machines in cloud computing data centers, and construct a set of polymorphic behavior baselines that integrate migration contexts; By resolving the resource call dependencies between the virtual machine instance layer, the virtualization management layer, and the physical resource layer, a three-layer mapping relationship for the virtualization environment is constructed. In the three-layer mapping relationship, the resource call chain triggered by the security event is traced in reverse. When the behavior characteristics of the virtual machine instance layer are detected to violate the polymorphic behavior baseline set, and the violation generates an unauthorized resource scheduling request in the virtualization management layer and is mapped as a cross-tenant shared resource access operation in the physical resource layer, the complete resource call chain from the virtual machine instance layer to the physical resource layer is marked as a cross-layer attack path. Based on the polymorphic behavior baseline set and the cross-layer attack path, locate the set of virtual machine instances of security threats and their corresponding attack stage identifiers, generate control instructions, and collect feedback data after executing the control instructions; Based on the migration events and threat handling results of virtual machine instances in the feedback data, the migration context and behavior boundaries of the virtual machine instances corresponding to the polymorphic behavior baseline set are updated, and the dependency weights of the three-layer mapping relationship are adjusted to generate a monitoring strategy covering the entire lifecycle of the virtual machine instance.

2. The method according to claim 1, characterized in that, Extracting physical location switching features and resource access pattern change features from the spatiotemporal migration trajectories of virtual machines in cloud computing data centers, a set of polymorphic behavioral baselines integrating migration contexts is constructed, including: Identify physical host switching events of virtual machine instances from the resource scheduling event sequence of the virtualization management layer; Extract the physical host machine identifier and the timestamp of the switching time before and after the physical host machine switching event to form the physical location switching feature; Extract the resource access operation types and access frequencies of the virtual machine instance within a preset time window before and after the switch, calculate the differences in the distribution of resource access operation types and the rate of change in access frequency before and after the physical host switch event, and constitute the resource access mode change characteristics. The physical location switching feature, the resource access mode change feature, and the virtual machine instance identifier are associated and bound to form a feature tuple carrying the migration context. For each virtual machine instance, based on the resource access mode change characteristics corresponding to different physical host identifiers in the feature tuple, multiple physical location-related behavioral baseline rules are established, and the multiple behavioral baseline rules together constitute the polymorphic behavioral baseline of a single virtual machine instance. The polymorphic behavior baselines of all virtual machine instances within the cloud computing data center are aggregated to form the polymorphic behavior baseline set.

3. The method according to claim 1, characterized in that, By resolving the resource call dependencies between the virtual machine instance layer, the virtualization management layer, and the physical resource layer, the three-layer mapping relationship of the virtualization environment is constructed, including: Record the identifier of the resource request initiated by the virtual machine instance; Extract virtual machine instance layer resource request records and corresponding target resource types, virtualization management layer resource scheduling records and corresponding physical resource location information, and physical resource layer access response records and corresponding physical resource identifiers; Based on the time-series constraints of timestamps, records whose request initiation timestamps, scheduling completion timestamps, and response completion timestamps satisfy a sequentially increasing relationship are initially matched. In the initially matched records, by mapping the requested target resource type to the physical resource location information, and mapping the physical resource location information to the physical resource identifier, a resource semantic mapping path from the virtual machine instance layer to the physical resource layer is established, and the record combination with the complete resource semantic mapping path is marked as a valid association; Using the resource request identifier in the valid association as the cross-layer association key, establish an association index between the virtual machine instance layer resource request record, the virtualization management layer resource scheduling record, and the physical resource layer access response record; Based on the associated index, a resource call dependency chain is constructed from the virtual machine instance layer resource request to the physical resource layer access response, and the resource call dependency chain forms the three-layer mapping relationship.

4. The method according to claim 3, characterized in that, In the three-layer mapping relationship, the resource call chain triggered by the security event is traced backwards. When the behavior characteristics of the virtual machine instance layer violate the polymorphic behavior baseline set, and the violation generates an unauthorized resource scheduling request at the virtualization management layer and maps it to a cross-tenant shared resource access operation at the physical resource layer, the complete resource call chain from the virtual machine instance layer to the physical resource layer is marked as a cross-layer attack path, including: The resource access operations monitored by the virtual machine instance layer are matched with the behavior baseline rules of the corresponding virtual machine instance in the polymorphic behavior baseline set. When the resource access operation exceeds the boundary range defined by the behavior baseline rule, it is marked as a violation and the resource request identifier is recorded. By using the associated index in the three-layer mapping relationship, the resource scheduling record of the virtualization management layer corresponding to the resource request identifier is searched in reverse; When the scheduling instruction permission identifier in the virtualization management layer resource scheduling record does not have a corresponding authorization record in the authorization configuration associated with the virtual machine instance identifier, the virtualization management layer resource scheduling record is marked as an unauthorized resource scheduling request. Based on the associated index, reverse search is performed on the physical resource access response record corresponding to the unauthorized resource scheduling request at the physical resource layer, and the physical resource identifier and the tenant identifier that initiated the access operation are extracted from the physical resource access response record. When the physical resource corresponding to the physical resource identifier is a shared resource shared by multiple tenants, and the tenant identifier is inconsistent with the tenant identifier to which the virtual machine instance identifier belongs, it is marked as a cross-tenant shared resource access operation. The complete resource call chain from the violation at the virtual machine instance layer, the unauthorized resource scheduling request at the virtualization management layer, to the cross-tenant shared resource access operation at the physical resource layer is marked as the cross-layer attack path.

5. The method according to claim 1, characterized in that, Based on the polymorphic behavior baseline set and the cross-layer attack path, the set of virtual machine instances of the security threat and the corresponding attack stage identifier are located, and control instructions are generated, including: Extract the virtual machine instance identifier corresponding to the violation behavior of the virtual machine instance layer from the cross-layer attack path, and use it as the initial threat virtual machine instance; Based on the associated index in the three-layer mapping relationship, find all resource request identifiers initiated by the initial threat virtual machine instance within a preset time window, and trace the associated physical resource layer access response records through the resource request identifiers; Find all physical resource access response records in the physical resource layer access response records that were accessed within the preset time window for the accessed physical resource identifier. From all physical resource access response records, trace back the virtual machine instance identifier that initiated the access and mark the virtual machine instance identifier that initiated the access as an associated threat virtual machine instance. The initial threat virtual machine instance and the associated threat virtual machine instance are merged to form the virtual machine instance set; For each virtual machine instance identifier in the virtual machine instance set, the corresponding attack stage identifier is determined and bound based on the violation behavior, unauthorized resource scheduling request, and cross-tenant shared resource access operation in the cross-layer attack path; When the attack phase is identified as the reconnaissance phase or the penetration phase, a resource access restriction instruction is generated; when the attack phase is identified as the privilege escalation phase or the lateral movement phase, a virtual machine isolation instruction is generated; when the attack phase is identified as the cross-tenant penetration phase or the persistence phase, a virtual machine termination instruction is generated.

6. The method according to claim 5, characterized in that, For each virtual machine instance identifier in the virtual machine instance set, the corresponding attack stage identifier is determined and bound based on the violation behavior, unauthorized resource scheduling request, and cross-tenant shared resource access operation in the cross-layer attack path, including: For each virtual machine instance identifier in the virtual machine instance set, extract the corresponding violation behavior, unauthorized resource scheduling request, and cross-tenant shared resource access operation from the cross-layer attack path; Analyze the resource access operation types that exceed the polymorphic behavior baseline set in the violation behavior. When the resource access operation type is a network scanning operation, mark the attack stage as the reconnaissance stage. When the resource access operation type is a vulnerability exploitation operation, mark the attack stage as the penetration stage. Parse the scheduling instruction permission identifier in the unauthorized resource scheduling request. When the scheduling instruction permission identifier involves virtual machine configuration modification permission, mark the attack phase identifier as the permission escalation phase. When the scheduling instruction permission identifier involves cross-virtual machine communication permission, mark the attack phase identifier as the lateral movement phase. Parse the physical resource identifier in the cross-tenant shared resource access operation. When the physical resource type corresponding to the physical resource identifier is storage resource, mark the attack stage as cross-tenant penetration stage. When the physical resource type corresponding to the physical resource identifier is computing resource, mark the attack stage as persistence stage. The attack phase identifier is bound to the corresponding virtual machine instance identifier.

7. The method according to claim 1, characterized in that, Based on the migration events and threat handling results of virtual machine instances in the feedback data, the migration context and behavior boundaries of the virtual machine instances corresponding to the polymorphic behavior baseline set are updated, and the dependency weights of the three-layer mapping relationship are adjusted to generate a monitoring strategy covering the entire lifecycle of the virtual machine instance, including: Parse the source physical node identifier and target physical node identifier in the migration event record, and query the corresponding physical resource layer configuration differences in the three-layer mapping relationship as migration context change features; The matching degree between the successfully intercepted attack paths in the threat handling results and the cross-layer attack paths is statistically analyzed. When the matching degree exceeds a preset matching threshold, the resource access operation types involved in the attack paths are extracted as high-risk operation types. For the behavior baseline rules of the corresponding virtual machine instance in the polymorphic behavior baseline set, the migration context change feature is added to the migration context field, and the high-risk operation type is added to the prohibited operation list to narrow the behavior boundary range. The anomaly detection contribution rate of the virtual machine instance layer, virtualization management layer and physical resource layer in the threat handling result is calculated in the three-layer mapping relationship. The anomaly detection contribution rate is used as the dependency weight of the corresponding layer. When the dependency weight of one layer is lower than the preset weight threshold, the priority of the association index in the three-layer mapping relationship is reduced. Based on the updated polymorphic behavior baseline set and the adjusted three-layer mapping relationship, corresponding monitoring policies are generated for the creation, operation, migration and destruction phases of virtual machine instances. The monitoring policies include behavior baseline rules, associated index priorities and anomaly detection trigger conditions for each phase.

8. An intelligent security monitoring system in a virtualized environment, used to implement the method as described in any one of claims 1-7, characterized in that, include: The polymorphic baseline unit is used to extract physical location switching features and resource access mode change features in the spatiotemporal migration trajectory of virtual machines in cloud computing data centers, and to construct a set of polymorphic behavioral baselines that integrate migration context. The three-layer mapping unit is used to construct the three-layer mapping relationship of the virtualization environment by resolving the resource call dependencies between the virtual machine instance layer, the virtualization management layer, and the physical resource layer. The attack path marking unit is used to trace the resource call chain triggered by the security event in reverse in the three-layer mapping relationship. When the behavior characteristics of the virtual machine instance layer are detected to violate the polymorphic behavior baseline set, and the violation generates an unauthorized resource scheduling request in the virtualization management layer and is mapped as a cross-tenant shared resource access operation in the physical resource layer, the complete resource call chain from the virtual machine instance layer to the physical resource layer is marked as a cross-layer attack path. The threat location and control unit is used to locate the set of virtual machine instances of security threats and their corresponding attack stage identifiers based on the polymorphic behavior baseline set and the cross-layer attack path, generate control instructions, and collect feedback data after executing the control instructions. The baseline mapping adjustment unit is used to update the migration context and behavior boundary of the virtual machine instance corresponding to the polymorphic behavior baseline set based on the migration events and threat handling results of the virtual machine instance in the feedback data, and adjust the dependency weight of the three-layer mapping relationship to generate a monitoring strategy covering the entire life cycle of the virtual machine instance.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.