Reactive capacity expansion and contraction scheduling method based on cost effectiveness under multi-resource constraint
By introducing reactive scaling strategies in Kubernetes, real-time monitoring of load conditions and adjusting Pod resource configuration, the problems of low resource scheduling efficiency and wasteful cost in the existing technology are solved, and flexible resource management is realized when load changes and service stability is ensured. and resource utilization efficiency.
Patent Information
- Application Number
- CN202410965495.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-18
- Publication Date
- 2024-11-15
AI Technical Summary
Under the multi-resource constraints, Kubernetes is difficult to effectively scheduling when load mutations are loaded, resulting in resource waste or performance problems. The existing vertical Pod automatic scaling mechanism relies on historical data predictions and cannot respond quickly to load changes.
A cost-effective reactive scaling scheduling method based on multi-resource constraints is proposed. By monitoring the real-time load situation, calculating the scaling threshold Score, selecting predictive or reactive scaling resource allocation schemes, adjusting the Pod resource configuration in real time, ensuring that Resource utilization efficiency and cost optimization.
It realizes rapid response when load changes suddenly, ensures service stability and reliability, avoids excessive resource allocation, optimizes resource utilization, reduces operation and maintenance costs, and improves overall system efficiency and reliability.
Smart Images

Figure CN118964002A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of cloud computing container platforms, and in particular relates to a cost-effective reactive capacity expansion and contraction scheduling method under multiple resource constraints. Background Art
[0002] With the development of cloud computing, its advantages and convenience have attracted more and more enterprises and users to the cloud. Container technology has become a key tool for modern application deployment and management. Kubernetes and Docker have become two key technologies for modern cloud computing and containerized application deployment. Their combination provides a powerful container orchestration and management solution, allowing developers to easily deploy, scale and manage containerized applications in different environments. In containerized management systems, Kubernetes has become the de facto standard for managing and scheduling containers, a container orchestration platform suitable for production environments.
[0003] Scalability describes the ability to scale up resources or applications to meet gradually expanding needs. In cloud computing, special attention is paid to the ability of container resources to expand or shrink. This scaling can be done by operators or in an automated way. Autoscaling algorithms are increasingly valued and widely used in the cloud computing industry. The load in a cloud environment is constantly changing and dynamic, which makes it challenging to optimally allocate resources for a given Pod and perform autoscaling. Kubernetes is currently the most widely used and popular cloud-native container orchestration architecture. It introduces two autoscaling mechanisms: horizontal Pod autoscaling (HPA) and vertical Pod autoscaling (VPA), which correspond to the needs of horizontal and vertical scaling, respectively.
[0004] Currently, Kubernetes itself only has one default VPA implementation for vertical scaling of Pods. The process is as follows: First, define the VPA resources that need to be adjusted for the Pod. The VPA Recommender observes the historical usage data of the Pod, evaluates the data, and predicts the amount of resources required by the Pod. The Recommender creates recommended values for resource usage for each Pod, including the requested amount and limit of CPU and memory. If the recommended resource configuration is inconsistent with the current Pod resource configuration, VPA will automatically update the Pod's resource configuration. This usually requires killing the running Pod and recreating a Pod with a new resource configuration. VPA will continue to monitor the Pod's resource usage and make adjustments when necessary to ensure that the Pod always has the right amount of resources.
[0005] However, when adopting the scaling algorithm of default automatic scaling, the only factor it mainly considers is the CPU utilization rate. The single constraint of the container scaling threshold causes some Pods to be unable to obtain sufficient resources due to their different requirements, resulting in resource waste or performance problems. Moreover, due to the lack of research on the energy consumption problem under multiple resource constraints, how to maintain a low-cost state during the operation of the Kubernetes default scaling mechanism in the case of sudden load changes is an unsolved problem.
[0006] For example, when executing tasks, two different types of loads are often faced: one is a relatively stable load, and the other is a suddenly surging load. When using the Vertical Pod Autoscaler (VPA) for vertical Pod scaling, it will adopt a method called sliding window plus half-life histogram for resource scaling prediction in the case of low load. First of all, when the load increases sharply, this prediction method may limit the speed of Pod expansion. Therefore, after multiple scaling adjustments, the resource usage of the Pod can reach the level required under this load. This situation may lead to the inability to meet the Service Level Agreement (SLA), thus increasing the violation rate. Secondly, when the load suddenly decreases, this prediction method may cause restrictions on Pod scaling. In this case, since the prediction model still relies on historical load data for adjustment, it may not be able to adapt to the sharp drop in load in a timely manner. Therefore, the Pod cannot quickly release the excess resources, resulting in the problem of excessive resource waste. Summary of the Invention
[0007] In view of the deficiencies of the prior art, the present invention proposes a reactive scaling scheduling method based on cost-benefit under multiple resource constraints, which includes:
[0008] S1: Based on the Kubernetes container management platform, monitor the metric information of the k8s cluster;
[0009] S2: Submit a YAML task, and initialize the resource allocation amount of the Pod according to the configuration information of the YAML file; use the allocated resource amount for service calls;
[0010] S3: Calculate the scaling threshold Score according to the current resource usage status of the Pod;
[0011] S4: At the current time step, if 0.8 < Score ≤ 2.0 or -2.0 ≤ Score < 0.2, execute the predictive scaling resource allocation scheme; if Score > 2.0 or Score < -2.0, then execute the reactive scaling resource allocation scheme;
[0012] S5: Return to step S3 after an interval of one unit of time until the service call task ends.
[0013] Preferably, the formula for calculating the expansion and contraction threshold Score is:
[0014]
[0015] Among them, N Pod Indicates the number of Pods, N V represents the number of SLA violations, α represents the weight coefficient, and U r (Pod i ,t j ) represents the amount of resources required by the i-th Pod in the case of the j-th SLA breach, U a (Pod i ,t j ) represents the actual amount of resources allocated to the i-th Pod in the case of the j-th SLA breach.
[0016] Preferably, the process of executing the predictive scaling resource allocation scheme includes:
[0017] Fill the Pod allocation resource data into the histogram and calculate the initial size of each bucket in the histogram;
[0018] Calculate the weight index and update the size of each bucket according to the weight index;
[0019] Calculate the recommended resource value, upper limit value, and lower limit value based on the bucket size in the histogram;
[0020] Adjust the Pod allocation resources and make service calls based on the recommended resource values, upper and lower limits.
[0021] Furthermore, the formula for calculating the initial size of each bucket is:
[0022]
[0023] Among them, S n Indicates the initial size of the nth bucket, B first represents the size of the first bucket, and r represents the balance parameter.
[0024] Furthermore, the formula for calculating the weight index and updating the size of each bucket is:
[0025]
[0026] S′ n =ζ×S n
[0027] Among them, ζ represents the weight index, τ represents the current time, τ0 represents the reference time, λ represents the attenuation parameter, and S′ n Indicates the size of the nth bucket after update, S nIndicates the initial size of the nth bucket.
[0028] Furthermore, the formula for calculating the recommended resource value, upper limit value and lower limit value is:
[0029]
[0030]
[0031]
[0032] Among them, T represents the recommended value of resources; L represents the lower limit of resources; U represents the upper limit of resources, b 90 represents the current 90th percentile resource usage; σ represents the safety margin ratio, p represents the minimum CPU usage of the Pod, N represents the number of samples in the prediction window, and b 50 represents the current 50 percentile resource usage, δ represents the forecast error adjustment coefficient, and b 95 Indicates the current 95th percentile resource usage.
[0033] Preferably, the process of executing the reactive scaling resource allocation scheme includes:
[0034] Get the target reference and determine the target namespace, and list the Pods in the target namespace that match the target reference label;
[0035] Calculate the lower limit, unrestricted target value, and upper limit of the Pod's resources;
[0036] Adjust the Pod allocation resources and make service calls based on the lower limit, unrestricted target value, and upper limit of the resources.
[0037] Furthermore, the formula for calculating the unrestricted target value is:
[0038]
[0039] Among them, T fc represents the unrestricted target value, V represents the current resource supply value, NU cur Indicates the current resource utilization of the Pod, NU tar Indicates the target resource utilization of the Pod.
[0040] The beneficial effects of the present invention are as follows: the present invention adds a reactive scaling strategy by studying the default Recommender of VPA. This strategy will pay more attention to the response to the current real-time load, rather than relying entirely on the prediction of historical data. By monitoring the real-time load situation, including the usage of resources such as CPU and memory, the system can respond more quickly and adjust the resource request of the Pod in time to meet the performance requirements under the new load. This reactive strategy can adjust resource allocation more flexibly when the load changes suddenly, ensuring that SLA requirements are met and avoiding over-allocation of resources. For example, when the load suddenly increases, the strategy can quickly adjust the resource request of the Pod to cope with the surge in demand and maintain the stability and reliability of the service; and when the load decreases, the excess resources are released in time to avoid resource waste. In summary, by introducing a reactive scaling strategy, VPA can more effectively cope with the challenges brought by load mutations. The present invention adjusts resource requests more flexibly according to the real-time load situation, thereby optimizing resource utilization to the greatest extent without violating the SLA, reducing cost waste, and improving the efficiency and reliability of the overall system. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] Figure 1 It is a flow chart of the reactive capacity expansion and contraction scheduling method based on cost-effectiveness under multiple resource constraints in the present invention;
[0042] Figure 2 It is a structural schematic diagram of the reactive scaling scheduling method based on cost-effectiveness under multiple resource constraints in the present invention. DETAILED DESCRIPTION
[0043] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0044] The present invention proposes a cost-effective reactive scaling scheduling method under multiple resource constraints. Figure 1 , Figure 2 As shown, the method includes:
[0045] S1: Based on the Kubernetes container management platform, monitor the indicator information of the k8s cluster.
[0046] In the Kubernetes container management platform, monitoring tools such as Cadvisor and Prometheus are used to monitor the information of the k8s cluster, such as CPU utilization, memory utilization, network IO metrics, and external resources (network latency, throughput), etc.; and monitor other resources in addition to the default resources, including custom application-level metrics, such as response time and number of requests.
[0047] S2: Submit a YAML task, initialize the resource allocation amount of the Pod according to the configuration information in the YAML file; use the allocated resource amount for service calls.
[0048] Submit a YAML task. The configuration information in the YAML file contains the Pod resource amount allocation information. Initialize the resource allocation amount of the Pod according to the configuration information in the YAML file, create a Pod, and perform service calls according to the allocated resource amount.
[0049] S3: Calculate the scaling threshold Score based on the current resource usage status of the Pod.
[0050] The formula for calculating the scaling threshold Score is:
[0051]
[0052] Among them, N Pod represents the number of Pods, N V represents the number of SLA violations, and α represents the weight coefficient. U r (Pod i , t j ) represents the resource amount required for the i-th Pod in the j-th SLA violation case, and U a (Pod i , t j ) represents the actually allocated resource amount for the i-th Pod in the j-th SLA violation case.
[0053] S4: At the current time step, if 0.8 < Score ≤ 2.0 or -2.0 ≤ Score < 0.2, execute the predictive scaling resource allocation scheme; if Score > 2.0 or Score < -2.0, then execute the reactive scaling resource allocation scheme.
[0054] If 0.8 < Score ≤ 2.0 or -2.0 ≤ Score < 0.2, execute the predictive scaling resource allocation scheme, and the predictive scaling resource allocation scheme includes:
[0055] Step 1: Fill the Pod allocated resource amount into the histogram and calculate the initial size of each bucket in the histogram. Specifically:
[0056] Assume that the consumption of each resource (such as CPU and memory) is an independent random variable, and its distribution is equal to the distribution of the variable observed in the past N units of time (for example, if the unit is day, calculate the container resource distribution on the 8th day to capture the weekly business container peak). The Recommender component obtains real-time data on resource consumption and stores it in the corresponding resource object CheckPoint. The CheckPoint CRD resource is essentially a histogram.
[0057] The size of each bucket in the histogram increases exponentially. Assume that the size of the first bucket is B. first (the maximum range that a bucket can cover), the initial size of the nth bucket is calculated as follows:
[0058]
[0059] Among them, S n Indicates the initial size of the nth bucket, B first Represents the size of the first bucket, r represents the balance parameter; r is the size ratio of each bucket to the previous bucket (the size of the bucket grows exponentially), and r is a number greater than 1. Usually a reasonable value is selected to balance the number of buckets and the growth rate of the bucket size. For example, if the value of r is 1.1, the accuracy is high and it is suitable for situations where the data distribution is relatively concentrated. If the value of r is 2.0, the number of buckets is small and the computational and memory overhead is small.
[0060] The more buckets there are, the larger the size of the nth bucket is, and the actual number of buckets is B. n for:
[0061]
[0062] Among them, V max It is the maximum value of resources that can be obtained, that is, the maximum value of resources defined in the YAML file configuration information.
[0063] Step 2: Calculate the weight index and update the size of each bucket based on the weight index:
[0064] Weight Index:
[0065]
[0066] Among them, ζ represents the weight index, τ represents the current time, and τ0 represents the reference time; λ represents the decay parameter, which determines the decay rate. Therefore, the closer the data point is to the current time τ, the higher the weight; the farther away from the current time, the lower the weight.
[0067] Recommender obtains weighted sample data from Metrics server or Prometheus, such as CPU and Memory resource usage in container dimensions, and multiplies the weight of each sample data by exponent ζ to ensure that newer samples are given higher weights, while the weights of older samples gradually decay over time. That is, the size of each bucket should be the information of each resource, i.e., each k8s cluster, multiplied by ζ; therefore, the formula for updating the size of each bucket is:
[0068] S′ n =ζ×S n
[0069] Among them, S′ n Indicates the size of the nth bucket after update.
[0070] Step 3: Calculate the recommended resource value, upper limit, and lower limit based on the bucket size in the histogram:
[0071]
[0072]
[0073]
[0074] Among them, T represents the recommended resource value, which ensures that the Pod has sufficient resources; L represents the lower limit of resources, which ensures that the resource request of the Pod will not be lower than a certain safety lower limit; U represents the upper limit of resources, which prevents the Pod from requesting too many resources and causing resource waste; 90 Represents the current 90th percentile resource usage, that is, the resource usage of the Pod is lower than this value in 90% of the time. 90 =90%×S′ n σ represents the safety margin ratio, b 90 (1+σ) represents the predicted value based on historical data, and the error standard deviation is added to provide a conservative estimate. p represents the minimum CPU usage of the Pod, N represents the number of samples in the prediction window, which represents the number of resource usage data points collected during this period, that is, the current number of Pods; b 50 Indicates the current 50 percentile resource usage; δ prediction error adjustment coefficient, used to adjust the range of prediction error; b 95 Indicates the current 95th percentile resource usage.
[0075] Step 4: Adjust the amount of resources allocated to the Pod based on the recommended resource value, upper limit, and lower limit, and make a service call.
[0076] Based on the recommended resource values, upper and lower limits, VPA specifies
[0077] ContainerResourcePolicy further processes to prevent the recommended value from being too large or too small and makes service calls. The ContainerResourcePolicy defined by VPA for each container is used to control the container's vertical scaling mode, the maximum and minimum ranges allowed for resource recommendation values, and adjustable resources.
[0078] If Score>2.0 or Score<-2.0, the reactive scaling resource allocation scheme is executed. The reactive scaling resource allocation scheme includes:
[0079] Step 1: Get the target reference and determine the target namespace, and list the Pods in the target namespace that match the target reference label. Specifically:
[0080] First, get the target reference targetRef, which is usually a Deployment or StatefulSet. Then determine the target namespace and list the Pods in the target namespace that match the target reference label.
[0081] Step 2: Calculate the lower limit, unrestricted target, and upper limit of the Pod's resources.
[0082] The formula for calculating the unconstrained target value is:
[0083]
[0084] Among them, T fc Indicates an unrestricted target value, V indicates the current resource supply value, i.e., the actual supply value of the current CPU or memory or other external resources; NU cur Indicates the current resource utilization of the Pod; NU tar Indicates the target resource utilization of the Pod, where NU tar The default is generally 80%.
[0085] For example, when the current Pod's CPU request value is 200m (milli-cores), the current CPU utilization is 210%, and the target CPU utilization is 80%. Then, the calculation of the unconstrained target value is as follows:
[0086]
[0087] Recommended values for new resources are constrained between minimum and maximum values:
[0088] T′ fc =max(min(T fc ,V max ),V min )
[0089] Likewise, the lower limit:
[0090] L fc =T fc ×(1-Δ)
[0091] The lower value is clamped between the minimum and maximum values:
[0092] L′ fc =max(min(L fc ,V max ),V min )
[0093] Likewise, the upper limit:
[0094] U fc =T fc ×(1+Δ)
[0095] The upper value is clamped between the minimum and maximum values:
[0096] U′ fc =max(min(U fc ,V max ),V min )
[0097] Wherein, Δ is a constant used to determine the offset between the lower limit value and the upper limit value, for example, Δ=0.2 in the present invention. max V is the maximum available resource for the Pod defined in YAML. min The minimum amount of resources available for the Pod defined in YAML.
[0098] Step 3: Adjust the amount of resources allocated to the Pod based on the lower limit, unrestricted target value, and upper limit of the resource and make a service call.
[0099] Based on the unrestricted target value, the recommended resource value can be obtained. Based on the new recommended resource value, the upper limit value of the resource, and the lower limit value of the resource, VPA specifies ContainerResourcePolicy for each container for further processing to prevent the recommended value from being too large or too small, and to make service calls. The ContainerResourcePolicy defined by VPA for each container is used to control the vertical scaling mode of the container, the maximum and minimum ranges allowed for the recommended resource value, and the adjustable resources.
[0100] Through the above formula, the target value and its upper and lower limits can be calculated based on the VPA's optimal request and the system's node resource utilization, which can improve the system's performance under different workloads. Adjust the Pod's resource requirements based on the resource recommendations provided by VPA and the actual operating status of the node. When the load on the working node is very high, such as when the node load Score value exceeds 200%, that is, Score>2.0 or Score<-2.0, a higher target resource value is calculated so that the Pod can obtain more resources to handle the increased load. Conversely, when the node load Score value is low, it means that the load is light, and the Pod can reduce resource usage and save costs. In particular, 0.2≤Score≤0.8, maintain the current recommended value.
[0101] S5: Return to step S3 after a unit of time until the service call task is completed.
[0102] Scale according to the selected scaling algorithm. To prevent oscillation caused by frequent scaling under excessive load changes, the parameter --vertical-pod-autoscaler-downscale-stabilization is introduced. The default value of this parameter is 1 unit time, that is, after the last scaling-down execution, the scaling-down can be performed again after 1 unit time to distribute the scaling results to the Pod. Return to S3 and continue iterating until the task is completed.
[0103] Evaluation of the present invention:
[0104] By normalizing multiple resources as an important indicator for evaluating reactive scaling, a cost-effectiveness model based on multiple resource constraints is constructed. That is, the model is established based on the CPU, memory, storage, IO and other resources of the Pod:
[0105]
[0106] Where n represents different resources, including CPU, memory, storage, etc. θ represents the controllable adjustment of the weight of each resource, with the default initial value of 0.3, U i (t) represents the usage of resource i at time t, P i (t) represents the unit price of resource i at time t. Cost is the total cost, where t0 and t1 are the start and end times of the task. F(t) represents the fixed cost at time t, expressed as:
[0107]
[0108] Where W represents the number of servers, Q i (t) represents the power consumption of server i during time t, S irepresents the resource amount of server i, C p represents the fixed unit price cost of the k8s cluster, S i With C p The multiplication represents the cost price of a server.
[0109] When the load drops sharply, the normalized resource U will decrease faster in reactive scaling than in predictive scaling to adapt to the load change. At the same time, the Cost value will decrease as the resource supply decreases. That is, the reduction of cluster resources effectively saves resources and reduces operating costs.
[0110] The present invention innovatively proposes a Pod cost-effectiveness model that comprehensively considers multiple resource constraints in a microservice architecture. The model not only covers the two traditional computing resources of CPU and memory, but also includes other key resources such as storage and network bandwidth. Through accurate modeling and real-time monitoring of these resource requirements, the present invention can dynamically adjust the resource allocation of the Pod according to predictive and reactive scaling scheduling, and realize the elastic scaling of container resources under different load conditions. This method can effectively meet the requirements of the service level agreement (SLA), while maximizing the reduction of resource waste and reducing operation and maintenance costs. In addition, the present invention expands and optimizes the recommender component of the existing vertical Pod autoscaler (VPA), and adds a reactive scheduling strategy based on current real-time load changes. This strategy abandons the practice of over-reliance on historical data prediction, and instead responds quickly according to real-time load data to adjust the Pod resource configuration to match actual needs. This makes the system more flexible and rapid in the face of sudden changes in load, which can not only ensure the stability and responsiveness of the service, but also avoid unnecessary over-configuration of resources. In summary, by combining the cost-effectiveness model with a reactive scheduling strategy based on real-time data, the present invention can effectively improve resource utilization efficiency and the overall performance of the system, and achieve optimal cost management while ensuring service quality.
[0111] The above embodiments further illustrate the purpose, technical solutions and advantages of the present invention in detail. It should be understood that the above embodiments are only preferred implementation modes of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made to the present invention within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A cost-effective reactive scaling scheduling method under multiple resource constraints, characterized in that: Including: S1: Based on the Kubernetes container management platform, monitor the metric information of the k8s cluster; S2: Submit a YAML task and initialize the resource allocation amount of the Pod according to the configuration information in the YAML file; Use the allocated resource amount for service calls; S3: Calculate the scaling threshold Score based on the resource usage status of the current Pod; S4: At the current time step, if 0.8 < Score ≤ 2.0 or -2.0 ≤ Score < 0.2, execute the predictive scaling resource allocation scheme; if Score > 2.0 or Score < -2.0, then execute the reactive scaling resource allocation scheme; S5: Return to step S3 after an interval of one unit of time until the service call task ends.
2. According to the cost-effective reactive capacity expansion and contraction scheduling method under multiple resource constraints according to claim 1, it is characterized in that: The formula for calculating the scaling threshold Score is: Among them, N Pod Indicates the number of Pods, N V represents the number of SLA violations, α represents the weight coefficient, and U r (Pod i ,t j ) represents the amount of resources required by the i-th Pod in the case of the j-th SLA breach, U a (Pod i ,t j ) represents the actual amount of resources allocated to the i-th Pod in the case of the j-th SLA breach.
3. According to the cost-effective reactive scaling scheduling method under multiple resource constraints of claim 1, it is characterized in that: The process of executing the predictive scaling resource allocation scheme includes: Fill the Pod allocated resource amount data into a histogram and calculate the initial size of each bucket in the histogram; Calculate the weight index and update the size of each bucket according to the weight index; Calculate the resource recommendation value, upper limit value, and lower limit value based on the size of the buckets in the histogram; Adjust the Pod allocated resource amount according to the resource recommendation value, upper limit value, and lower limit value and perform service calls.
4. According to the cost-effective reactive capacity expansion and contraction scheduling method under multiple resource constraints according to claim 3, it is characterized in that: The formula for calculating the initial size of each bucket is: Among them, S n Indicates the initial size of the nth bucket, B first represents the size of the first bucket, and r represents the balance parameter.
5. According to the cost-effective reactive capacity expansion and contraction scheduling method under multiple resource constraints according to claim 3, it is characterized in that: The formula for calculating the weight index and updating the size of each bucket is: S′ n =ζ×S n Among them, ζ represents the weight index, τ represents the current time, τ0 represents the reference time, λ represents the attenuation parameter, and S′ n Indicates the size of the nth bucket after update, S n Indicates the initial size of the nth bucket.
6. The cost-effective reactive capacity expansion and contraction scheduling method under multiple resource constraints according to claim 3 is characterized in that: The formula for calculating the resource recommendation value, upper limit value, and lower limit value is: Among them, T represents the recommended value of resources; L represents the lower limit of resources; U represents the upper limit of resources, b 90 represents the current 90th percentile resource usage; σ represents the safety margin ratio, p represents the minimum CPU usage of the Pod, N represents the number of samples in the prediction window, and b 50 represents the current 50 percentile resource usage, δ represents the forecast error adjustment coefficient, and b 95 Indicates the current 95th percentile resource usage.
7. The method for reactive capacity expansion and contraction scheduling based on cost-effectiveness under multiple resource constraints according to claim 1 is characterized in that: The process of executing the reactive scaling resource allocation scheme includes: Obtain the target reference and determine the target namespace, and list the Pods in the target namespace that match the target reference label; Calculate the resource lower limit value, unrestricted target value, and upper limit value of the Pod; Adjust the Pod allocated resource amount according to the resource lower limit value, unrestricted target value, and upper limit value and perform service calls.
8. The cost-effective reactive capacity expansion and contraction scheduling method under multiple resource constraints according to claim 7 is characterized in that: The formula for calculating the unrestricted target value is: Among them, T fc represents an unrestricted target value, V represents the current resource supply value, BU cur Indicates the current resource utilization of the Pod, NU tar Indicates the target resource utilization of the Pod.
Citation Information
Cited By
Capacity expansion and contraction method and device, computer equipment and storage medium
CN119557056A
