Computing power network container migration method based on probability language and resource similarity

Abstract

The invention discloses a computing power network container migration method based on probability language and resource similarity, which belongs to the field of computing power networks, and comprises the following steps of: calculating a container demand semantic probability according to an average demand quantity of container resources, constructing a multi-dimensional resource judgment matrix and calculating a container demand semantic vector according to an actual demand and a declaration demand of the resources, and calculating a container resource demand probability language quantization vector. Calculating a computing power node resource semantic probability matrix, constructing a computing power node available resource probability language matrix, and calculating a computing power node available resource probability language quantization matrix. And calculating a resource similarity vector according to the container resource demand probability language quantization vector and the computing power node available resource probability language quantization matrix, and selecting the computing power node with the maximum similarity to perform container migration. According to the method, the similarity of multi-dimensional data between the container resource demand and the computing power node resource condition is effectively quantified, the migration result is ensured, the demand of a user container is met, and the stability and the efficiency of computing power network operation are improved.

Description

Technical Field

[0001] The present invention belongs to the field of computing power networks, and specifically relates to a computing power network container migration method based on probabilistic language and resource similarity. Background Art

[0002] With the development of the digital age, cloud computing has concentrated large-scale computing power on single-point devices by establishing large data centers, using virtualization technology to provide computing services to users, and using virtual machine migration technology to efficiently utilize computing resources and save energy. However, many specific application scenarios have specific requirements for the network throughput, concurrent computing and storage of data centers. On the other hand, the emergence of emerging technologies such as 5G / B5G has brought massive data to the edge of the network, accelerating the diffusion of computing power from a few data centers to the edge of the network and even terminal devices. In order to solve the problems of lack of effective coordination mechanism between computing nodes, imperfect allocation and scheduling mechanism of computing tasks, and low resource utilization, the concept of computing power network came into being. The computing power network connects the computing power resources of distributed computing nodes through the network, and provides flexible and high-quality computing services for various applications in the form of a computing power resource pool across the entire network. Therefore, in view of the increase in resource demand dimensions in the computing power network, selecting appropriate computing power nodes for container migration is of great significance to improving the coordination ability between computing power nodes and improving resource utilization.

[0003] The probabilistic language term set is a semantic set composed of semantic sets and the probabilities corresponding to the semantic sets. It can not only express the indecision of experts in program evaluation, but also show the degree of preference of experts. Compared with other semantic evaluation methods, it can avoid the loss of expert preference information in the evaluation process. However, the term set relies on expert scores to extract the importance of each indicator, and it is impossible to obtain indicator information directly in the computing power network. There is a lack of quantitative methods for computing network user preferences and computing power node service capabilities. Among the indicators of the computing power network, the hesitation and uncertainty of different computing network user term sets are different, making it difficult to accurately and quickly measure and compare the matching degree between computing power user demand preferences and computing power service capabilities in the decision-making process. Summary of the invention

[0004] The present invention is a computing power network container migration method based on probabilistic language and resource similarity. The main idea of ​​the method includes: calculating the semantic probability of container demand according to the average demand for container resources, constructing a multi-dimensional resource judgment matrix and calculating the container demand semantic vector according to the actual demand and declared demand of resources, constructing a container resource demand probability language vector, and calculating the container resource demand probability language quantization vector. Calculate the computing power node resource semantic probability matrix, predict the load balancing degree of each computing power node after migration as the computing power node resource semantic vector, construct the computing power node available resource probability language matrix, and calculate the computing power node available resource probability language quantization matrix. Calculate the resource similarity vector based on the container resource demand probability language quantization vector and the computing power node available resource probability language quantization matrix, and select the computing power node with the largest similarity for container migration.

[0005] The technical solution adopted by the present invention is: a computing power network container migration method based on probabilistic language and resource similarity, the steps of the method are as follows:

[0006] Step 1: Get the computing node resource utilization list UV = {U1, U2, ..., U i ,…,U n}, where U i = {u i1 ,u i2 ,u i3 ,u i4},u i1 ,u i2 ,u i3 ,u i4 Respectively represent the utilization of the processor, memory, bandwidth, and disk of the i-th computing power node. Get the total resource list of computing power nodes MV = {M1, M2, ..., M i ,…,M n}, where M i ={m i1 ,m i2 ,m i3 ,m i4},m i1 ,m i2 ,m i3 ,m i4 They respectively represent the total amount of processor, memory, bandwidth, and disk resources of the i-th computing node.

[0007] Get the resource requirement list of the container to be migrated NV = {n1, n2, n3, n4}, where n1, n2, n3, n4 represent the declared requirements of the container processor, memory, bandwidth, and disk respectively. Get the historical requirement data set list of the container DATAS = {DATA1, DATA2, ..., DATA i ,…,DATA n}, where DATA i ={d i1 ,d i2 ,d i3 ,d i4}, where d i1 ,d i2 ,d i3 ,d i4 They represent the actual requirements of the processor, memory, bandwidth, and disk of the container at the i-th moment, respectively. n1 ,d n2 ,d n3 ,d n4 They represent the actual requirements of the container's current processor, memory, bandwidth, and disk respectively.

[0008] Step 2: Calculate the average container resource demand vector HV = {h1,…,h i ,…,h4}, where h i represents the average demand for the i-th type of resource in the container, d ji Indicates the demand for the i-th type of resource of the container at the j-th moment.

[0009]

[0010] The container demand semantic probability vector NSV is calculated according to the following formula: NSV = {ns1,…,ns i ,…,ns4}, where ns i Represents the demand semantic probability of the i-th type of resources.

[0011]

[0012] Step 3: Calculate the container multi-dimensional resource judgment matrix PT = (pt ij ) 4×4 , where pt ij It represents the demand utilization ratio of the i-th resource and the j-th resource.

[0013] pt ij =h i *n j / h j *n i

[0014] The container demand semantic vector RNW is calculated according to the following formula: i ,…,rn4}, where rn i Represents the semantic value of resource demand of the i-th class

[0015]

[0016] Step 4: Calculate the container resource demand probability language vector A = (a1, ..., a i ,…,a4), where the probability language a i The probabilistic language for expressing the i-th resource demand:

[0017]

[0018] where d ni Indicates the actual demand for the i-th resource of the container, ns i represents the semantic probability of the demand for the i-th resource, rn i Represents the semantic value of resource requirement of the i-th class.

[0019] Step 5: Calculate the container resource demand probability language quantization vector NA = (na1, ..., na i ,…,na4), where na i The probability linguistic quantification value representing the i-th type of resource demand:

[0020]

[0021] Step 6: Calculate the average available resource vector V of the computing power node according to the following formula: V = (v1,…,v i ,…,v4), where v i Represents the average available value of the i-th type of resources of the computing power node.

[0022]

[0023] According to the following formula, the computing power node resource semantic probability matrix WT = (wt ij ) n×4 , wt ij Represents the semantic probability of the j-th type of resource of the i-th computing power node.

[0024] wt ij =m ij *(1-u ij ) / v j

[0025] Step 7: Calculate the computing power node resource semantic vector L = (l1, l2, ..., l i ,…,l n ), where l i Represents the semantic value of the i-th computing power node resource.

[0026]

[0027] Step 8: Calculate the probability language matrix of available resources of computing power nodes CN = (CN1, CN2, ..., CNi ,…,CN n ), where CN i =(b i1 , b i2 , b i3 , b i4 ), represents the available resource probability language vector of the i-th computing power node, where b ij The probabilistic language for representing the j-th resource type of the i-th computing node:

[0028]

[0029] Among them l i Indicates the semantic value of the i-th computing power node resource, m ij represents the total amount of resources of the jth type in the i-th computing node, u ij represents the current utilization rate of the j-th resource class of the i-th computing node, wt ij Represents the semantic probability of the j-th type of resource of the i-th computing power node.

[0030] Step 9: According to the computing power node resource semantic probability matrix WT and the computing power node resource semantic vector L, calculate the computing power node available resource probability language quantization matrix NCN = (NCN1, NCN2, ..., NCN i ,…,NCN n ), NCN i =(nb i1 , nb i2 , nb i3 , nb i4 ), represents the available resource probability language quantization vector of the i-th computing power node, where nb ij Represents the probabilistic linguistic quantization value of the j-th available resource of the i-th computing power node:

[0031]

[0032] Step 10: Based on the container resource demand probability language quantization vector NA and the computing node available resource probability language quantization matrix NCN, calculate the similarity vector RES = {res1,res2,…,res i ,…,res n}, where res i Indicates the similarity between the available resources and container requirements of the i-th computing node.

[0033]

[0034] Step 11: Traverse the similarity vector RES = {res1,res2,...,res n}, get the maximum value resmax , output the corresponding computing power node number max as the destination for container migration.

[0035] Compared with the existing method, the present invention has the following beneficial effects:

[0036] 1. The present invention first quantifies the user's sensitivity to resource demand, calculates the stability of container resource demand as the semantic probability of the container, and quantifies the difference between the user's container subjective resource demand and the actual one by calculating the container demand semantic vector. Then, the container resource demand probability language vector is constructed, and the container resource demand probability language quantification vector is calculated, which takes into account the difference between the user's container subjective demand and the actual one, retains the user's container's sensitivity to resources, and effectively quantifies the importance of the user's container resource demand.

[0037] 2. The present invention calculates the semantic probability matrix of computing power node resources, highlights the characteristics of the resource distribution of each computing power node, calculates the semantic vector of each computing power node resource after migration, and quantifies the load balancing situation of each computing power node after migration, so as to construct the probability language matrix of available resources of computing power nodes, and calculates the probability language quantification matrix of available resources of computing power nodes. While considering the current resource situation of each node, it combines the load balancing degree of the node after migration, effectively quantifies the service capabilities of each current computing power node, reduces the probability of subsequent migration of containers, and improves the resource utilization of the computing power network.

[0038] 3. The present invention calculates the resource similarity vector based on the probability language quantization vector of container resource demand and the probability language quantization matrix of available resources of computing power nodes, effectively quantifies the similarity of multi-dimensional data between container resource demand and computing power node resource conditions, and ensures that the migration result not only meets the needs of user containers but also improves the stability and efficiency of computing power network operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a flow chart of the present invention.

[0040] Figure 2 This is an application scenario diagram of the present invention. DETAILED DESCRIPTION

[0041] The present invention will be further explained below in conjunction with the accompanying drawings and specific embodiments.

[0042] Figure 2 The computing power network dispatch center periodically collects the operation information of each computing power node, wherein the overloaded computing power node reports the list of containers to be migrated and the historical operation data, and other computing power nodes report the node load conditions, see step 1 for details.

[0043] For each container in the reported list of containers to be migrated, calculate the corresponding probabilistic linguistic quantization vector. According to the available computing power node load information, calculate the corresponding probabilistic linguistic quantization vector. Find the resource similarity between the container to be migrated and each computing power node, and select the target node, see steps 2-11 for details.

[0044] The technical solution of the present invention is a computing power network container migration method based on probabilistic language and resource similarity, comprising the following steps:

[0045] Step 1: Get the computing node resource utilization list UV = {U1, U2, ..., U i ,…,U n}, where U i = {u i1 ,u i2 ,u i3 ,u i4},u i1 ,u i2 ,u i3 ,u i4 Respectively represent the utilization of the processor, memory, bandwidth, and disk of the i-th computing power node. Get the total resource list of computing power nodes MV = {M1, M2, ..., M i ,…,M n}, where M i ={m i1 ,m i2 ,m i3 ,m i4},m i1 ,m i2 ,m i3 ,m i4 They respectively represent the total amount of processor, memory, bandwidth, and disk resources of the i-th computing node.

[0046] Get the resource requirement list of the container to be migrated NV = {n1, n2, n3, n4}, where n1, n2, n3, n4 represent the declared requirements of the container processor, memory, bandwidth, and disk respectively. Get the historical requirement data set list of the container DATAS = {DATA1, DATA2, ..., DATA i ,…,DATA n}, where DATA i ={d i1 ,d i2 ,d i3 ,d i4}, where d i1 ,d i2 ,d i3 ,d i4 They represent the actual requirements of the processor, memory, bandwidth, and disk of the container at the i-th moment, respectively. n1,d n2 ,d n3 ,d n4 They represent the actual requirements of the container's current processor, memory, bandwidth, and disk respectively.

[0047] Step 2: Calculate the average container resource demand vector HV = {h1,…,h i ,…,h4}, where h i represents the average demand for the i-th type of resource in the container, d ji Indicates the demand for the i-th type of resource of the container at the j-th moment.

[0048]

[0049] The container demand semantic probability vector NSV is calculated according to the following formula: NSV = {ns1,…,ns i ,…,ns4}, where ns i Represents the demand semantic probability of the i-th type of resources.

[0050]

[0051] Step 3: Calculate the container multi-dimensional resource judgment matrix PT = (pt ij ) 4×4 , where pt ij It represents the demand utilization ratio of the i-th resource and the j-th resource.

[0052] pt ij =h i *n j / h j *n i

[0053] The container demand semantic vector RNW is calculated according to the following formula: i ,…,rn4}, where rn i Represents the semantic value of resource requirement of the i-th class.

[0054]

[0055] Step 4: Based on the container resource demand stability and weight vector, use probability language to calculate the container resource demand probability language vector A = (a1, ..., a i ,…,a4), the following formula uses probabilistic language, which can not only express the difference between the user container's subjective resource requirements and the actual ones, but also show the user's sensitivity to different resources. i It is expressed as:

[0056]

[0057] where ai The probability language representing the resource demand of the i-th category, d ni Indicates the actual demand for the i-th resource of the container, ns i represents the semantic probability of the demand for the i-th resource, rn i Represents the semantic value of resource requirement of the i-th class.

[0058] Step 5: Calculate the container resource demand probability language quantization vector NA = (na1, ..., na i ,…, na4), the following formula takes into account the subjectivity and objectivity of the weight, combines the subjective and objective conditions and sensitivity of the user's container demand, and effectively quantifies the importance of the user's container resource demand, where na i The probability linguistic quantification value representing the i-th type of resource demand:

[0059]

[0060] Step 6: Calculate the average available resource vector V of the computing power node according to the following formula: V = (v1,…,v i ,…,v4), where v i Represents the average available value of the i-th type of resources of the computing power node.

[0061]

[0062] According to the following formula, the computing power node resource semantic probability matrix WT = (wt ij ) n×4 , wt ij Represents the semantic probability of the j-th type of resource of the i-th computing power node.

[0063] wt ij =m ij *(1-u ij ) / v j

[0064] Step 7: Calculate the computing power node resource semantic vector L = (l1, l2, ..., l i ,…,l n ), where l i Represents the semantic value of the i-th computing power node resource.

[0065]

[0066] Step 8: According to the load balance of computing nodes and the weight of available resources, use the probability language to calculate the probability language matrix of available resources of computing nodes CN = (CN1, CN2, ..., CN i ,…,CN n ), CNi =(b i1 , b i2 , b i3 , b i4 ), represents the probability language vector of available resources of the i-th computing power node. The following formula uses probability language to express not only the characteristics of resource distribution of each computing power node, but also reflects the load of each computing power node after migration, where:

[0067]

[0068] where b ij The probability language representing the j-th resource type of the i-th computing node, l i Indicates the semantic value of the i-th computing power node resource, m ij represents the total amount of resources of the jth type in the i-th computing node, u ij represents the current utilization rate of the j-th resource class of the i-th computing node, wt ij Represents the semantic probability of the j-th type of resource of the i-th computing power node.

[0069] Step 9: According to the computing power node resource semantic probability matrix WT and the computing power node resource semantic vector L, calculate the computing power node available resource probability language quantization matrix NCN = (NCN1, NCN2, ..., NCN i ,…,NCN n ), NCN i =(nb i1 , nb i2 , nb i3 , nb i4 ), represents the available resource probability language quantification vector of the i-th computing power node. The following formula combines the current resource distribution of each computing power node and the load balancing after migration, and effectively quantifies the service capacity of each current computing power node, where nb ij Represents the probabilistic linguistic quantization value of the j-th available resource of the i-th computing power node:

[0070]

[0071] Step 10: Based on the container resource demand probability language quantization vector NA and the computing node available resource probability language quantization matrix NCN, calculate the similarity vector RES = {res1,res2,…,res i ,…,res n}, where res i Represents the similarity between the available resources and container requirements of the i-th computing node. This vector effectively quantifies the similarity of multi-dimensional data between container resource requirements and computing node resource probability language, simplifies the comparison process,

[0072]

[0073] Step 11: Traverse the similarity vector RES = {res1,res2,...,res n}, get the maximum value res max , output the corresponding computing power node number max as the destination for container migration.

[0074] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A computing power network container migration method based on probabilistic language and resource similarity, characterized in that: The steps include: Step 1: Obtain the computing node resource utilization list UV, the computing node total resource list MV, the container resource demand list NV to be migrated, and the container historical demand data set list DATAS; Step 2: Calculate the container resource average demand vector HV and the container demand semantic probability vector NSV; Step 3: Calculate the container multi-dimensional resource judgment matrix PT and the container demand semantic vector RNW; Step 4: Calculate the container resource demand probability language vector A = (a1, ..., a i ,…,a4), where the probability language where d ni Indicates the actual demand for the i-th resource of the container, ns i represents the semantic probability of the demand for the i-th resource, rn i Represents the semantic value of resource demand of the i-th class; Step 5: Calculate the container resource demand probability language quantization vector NA according to the container demand semantic vector RNW and the container demand semantic probability vector NSV; Step 6: Calculate the average available resource vector V of the computing power node and the semantic probability matrix WT of the computing power node resources; Step 7: Calculate the computing power node resource semantic vector L according to the following formula; Step 8: Calculate the probability language matrix CN of available resources of computing power nodes; Step 9: Calculate the computing node available resource probability language quantization matrix NCN according to the computing node resource semantic probability matrix WT and the computing node resource semantic vector L; Step 10: Calculate the similarity vector RES based on the container resource demand probability language quantization vector NA and the computing node available resource probability language quantization matrix NCN; Step 11: Traverse the similarity vector RES = {res1,res2,...,res n }, get the maximum value res max , output the corresponding computing power node number max as the destination for container migration.

2. According to claim 1, a computing power network container migration method based on probabilistic language and resource similarity is characterized in that: The computing node resource utilization list UV in step 1 is {U1, U2, ..., U i ,…,U n }, where U i = {u i1 ,u i2 ,u i3 ,u i4 },u i1 ,u i2 ,u i3 ,u i4 Represent the utilization of the processor, memory, bandwidth, and disk of the i-th computing node respectively, and obtain the total resource list of computing nodes MV = {M1, M2, ..., M i ,…,M n }, where M i ={m i1 ,m i2 ,m i3 ,m i4 },m i1 ,m i2 ,m i3 ,m i4 Respectively represent the total amount of processor, memory, bandwidth, and disk resources of the i-th computing node; The container resource requirement list NV to be migrated in step 1 is NV={n1,n2,n3,n4}, where n1,n2,n3,n4 represent the declared requirements of the container processor, memory, bandwidth, and disk, respectively. The container historical requirement data set list DATAS={DATA1,DATA2,…,DATA i ,…,DATA n }, DATA i ={d i1 ,d i2 ,d i3 ,d i4 }, where d i1 ,d i2 ,d i3 ,d i4 They represent the actual requirements of the processor, memory, bandwidth, and disk of the container at the i-th moment, respectively. n1 ,d n2 ,d n3 ,d n4 They represent the actual requirements of the container's current processor, memory, bandwidth, and disk respectively.

3. According to claim 1, a computing power network container migration method based on probabilistic language and resource similarity is characterized in that: The average container resource demand vector HV in step 2 is {h1,…,h i ,…,h4}, and its calculation formula is as follows: where h i represents the average demand for the i-th type of resource in the container, d ji represents the demand for the i-th type of resource of the container at the j-th moment; the container demand semantic probability vector NSV in step 2 = {ns1,…,ns i ,…,ns4}, and its calculation formula is as follows: where ns i Represents the semantic probability of demand for the i-th type of resources.

4. According to claim 1, a computing power network container migration method based on probabilistic language and resource similarity is characterized in that: The container multi-dimensional resource judgment matrix PT in step 3 is (pt ij ) 4×4 ,in pt ij =h i *n j / h j *n i where pt ij It represents the demand utilization ratio of the i-th resource and the j-th resource; The container requirement semantic vector RNW in step 3 is as follows: i ,…,rn4}, the calculation formula is as follows: where rn i Represents the semantic value of resource requirement of the i-th class.

5. According to claim 1, a computing power network container migration method based on probabilistic language and resource similarity is characterized in that: The container resource demand probability language quantization vector NA in step 5 is equal to (na1, ..., na i , ..., na4), where: Among them, i Represents the probability linguistic quantification value of the i-th type of resource demand.

6. According to claim 1, a computing power network container migration method based on probabilistic language and resource similarity is characterized in that: The average available resource vector of the computing power nodes in step 6 is V=(v1,…,v i ,…,v4), the calculation formula is as follows: where v i Represents the average available value of the i-th type of resources in the computing power node; The computing power node resource semantic probability matrix WT in step 6 is WT=(wt ij ) n×4 ,in wt ij =m ij *(1-u ij ) / v j where wt ij Represents the semantic probability of the j-th type of resource of the i-th computing power node.

7. According to claim 1, a computing power network container migration method based on probabilistic language and resource similarity is characterized in that: The computing node resource semantic vector L in step 7 is L=(l1, l2, ..., l i ,…,l n ), which is calculated as follows: Among them l i Represents the semantic value of the i-th computing power node resource.

8. According to claim 1, a computing power network container migration method based on probabilistic language and resource similarity is characterized in that: The computing power node available resource probability language matrix CN in step 8 is CN=(CN1, CN2, ..., CN i , …, CN n ), where CN i =(b i1 , b i2 , b i3 , b i4 ), represents the available resource probability language vector of the i-th computing power node, where: where b ij The probability language representing the j-th resource type of the i-th computing node, l i Indicates the semantic value of the i-th computing power node resource, m ij represents the total amount of resources of the jth type in the i-th computing node, u ij represents the current utilization rate of the j-th resource class of the i-th computing node, wt ij Represents the semantic probability of the j-th type of resource of the i-th computing power node.

9. According to claim 1, a computing power network container migration method based on probabilistic language and resource similarity is characterized in that: The computing power node available resource probability language quantization matrix NCN in step 9 is NCN=(NCN1, NCN2, ..., NCN i ,…,NCN n ), where NCN i =(nb i1 , nb i2 , nb i3 , nb i4 ), represents the available resource probability language quantization vector of the i-th computing power node, where, where nb ij Represents the probabilistic linguistic quantization value of the j-th type of available resources of the i-th computing power node.

10. The computing power network container migration method based on probabilistic language and resource similarity according to claim 1 is characterized in that: The similarity vector RES in step 10 is {res1,res2,…,res i ,…,res n },in where res i Indicates the similarity between the available resources and container requirements of the i-th computing node.

Images