A Cloud-Edge Collaborative Optimization and Equilibrium Deployment Method with Delay Guarantee

Abstract

The present invention discloses a cloud-edge collaborative optimization and equilibrium deployment method with delay guarantee. Based on the joint optimization of computing-network-storage, considering the basic resource capacity and the load balance of cluster resources comprehensively, and combining the latency requirements and importance of applications, the present invention optimizes the deployment of applications to achieve the rapid response of applications and the efficient utilization of resources. When deploying business applications, it starts from three aspects: the construction attribute, the business attribute, and the status attribute of the application. After comprehensive consideration, it sorts according to the business volume after introducing the importance weight, and optimizes the deployment according to the differentiated guarantee requirements of business applications. When performing cloud-edge coordinated deployment, on the idea of balancing resource utilization rate, considering the latency of users using applications in the cloud-edge computing environment, it realizes the efficient utilization of resources while ensuring the normal usage experience of users. The present invention can effectively cope with resource constraints and real-time requirements in edge computing.

Description

Technical Field

[0001] The present invention belongs to the technical field of cloud computing and edge computing, and relates to a method for optimizing and balancing the deployment of cloud-edge collaboration with delay guarantee. Background Art

[0002] Cloud-edge collaboration is a combination of cloud computing and edge computing, which gives full play to the advantages of both and makes up for their respective deficiencies. In the cloud-edge collaboration architecture, the cloud center node is generally considered to have powerful computing capabilities, and applications with high requirements for computing resources and insensitive to time delay are deployed to be responsible for processing large-scale data analysis and complex computing tasks; while the edge node has limited computing resources, and applications with low requirements for computing resources and sensitive to time delay requirements are generally deployed to be responsible for processing computing tasks with high real-time requirements and small data volume. The two achieve data sharing and complementary computing capabilities through an efficient network connection and collaboration mechanism. By optimizing resource allocation through cloud-edge collaboration, the IT operation cost is reduced, and the overall benefit of informatization construction is improved.

[0003] How to select appropriate computing nodes to deploy business applications will directly affect the performance, response speed and service quality of the applications. Traditional application deployment strategies will consider factors such as the resource capacity of the nodes and cluster load balancing, but do not comprehensively consider user distribution and the time delay requirements of the applications, reducing the service quality of the applications; do not finely and comprehensively consider the adaptability of basic resources, resulting in uneven utilization of cluster resources; when deploying multiple applications, do not consider the differentiated guarantee requirements of the applications, or cause important applications to be unable to be deployed. Summary of the Invention

[0004] Aiming at the problems existing in the prior art, the purpose of the present invention is to provide a method for optimizing and balancing the deployment of cloud-edge collaboration with delay guarantee, which takes into account delay guarantee while ensuring the balanced deployment of business applications, so as to meet the needs of business operation and achieve the best user experience.

[0005] Based on the joint optimization of computing-network-storage, the present invention comprehensively considers the basic resource capacity (such as CPU, memory and disk storage, etc.) and the load balancing of cluster resources, combines the time delay requirements and importance of the applications, and optimizes the deployment of the applications to achieve the rapid response of the applications and the efficient utilization of resources.

[0006] When deploying business applications, the present invention does not simply deploy them in the default order or simply sort them according to the required resource volume, but starts from the construction attributes, business attributes and status attributes of the applications, and after comprehensive consideration, sorts them according to the business volume after introducing the importance weight, and optimizes the deployment according to the differentiated guarantee requirements of the business applications.

[0007] When the present invention performs cloud-edge coordinated deployment, in the idea of balancing resource utilization, it considers the latency of users using applications in the cloud-edge computing environment, and realizes the efficient utilization of resources while ensuring the normal use experience (i.e., latency) of users.

[0008] The specific flowchart of the solution of the present invention is as Figure 1 shown, and the steps are as follows:

[0009] (1) Obtain the list P of business applications to be deployed and the list K of computing nodes;

[0010] (2) Calculate the latency matrix T of the terminals that are farthest from each edge computing node k in the list K of computing nodes for each business application in the list P of business applications; A terminal refers to the user side using the business application, and the business application is deployed on the computing node; There are multiple terminals using the business application p. If the business application p is deployed on the edge computing node k, the latency of the terminal farthest from the node k can be calculated as t pk , then the value T at the position of the element in the p-th row and k-th column of the latency matrix T pk is 1. If this latency t pk is smaller than the maximum latency of the application p, that is, the latency matrix T selects the computing nodes that meet the latency requirements for all users of this application.

[0011] (3) According to the list of all applications to be deployed and the characteristic information of the applications, calculate the weighted computing power value and sort it in descending order to obtain the new list P_decay of business applications;

[0012] (4) Determine whether there are still undeployed applications in the list P_decay of business applications. If so, go to step (5), otherwise end the deployment process;

[0013] (5) Select the undeployed business application with the highest ranking from the list P_decay of business applications, and obtain the various resources it needs to occupy; that is, the numerical values of the various resources involved in the application characteristic information of this undeployed business application;

[0014] (6) Calculate the remaining computing power resources of all current computing nodes, count their standard computing power, and obtain the computing node matrix Q whose remaining computing power meets the requirements of this business deployment;

[0015] (7) If the currently to-be-deployed application and the already-deployed applications have affinity and the remaining computing power of the computing node where the affinity application is located meets the requirements of the currently to-be-deployed business application, then deploy the currently to-be-deployed application on the computing node with the most affinity services. After the currently to-be-deployed application is deployed, go to step (11), otherwise go to step (8);

[0016] (8) Based on the matrix Q and the matrix T, find the computing nodes among all computing nodes that meet the computing power resources and latency of the current application to be deployed, and then go to step (9). If all computing nodes only meet the computing power but not the latency, then go to step (10);

[0017] (9) Sort the computing nodes that meet the computing power and latency requirements according to the computing power standard values to obtain a list of computing nodes in descending order, and deploy the current application to be deployed on the computing node ranked first. The current application to be deployed is deployed, and the process proceeds to step (11);

[0018] (10) Calculate the average latency from the computing nodes that meet the computing power value to each terminal that uses the current business application to be deployed, sort them according to the average latency, obtain a list of computing nodes in ascending order, and deploy the current application to be deployed on the computing node with the first ranking. The current application to be deployed is deployed, and enter step (11);

[0019] (11) Update the remaining resource information of the computing node, remove the currently deployed application from the business application list P_decay and return to step (4).

[0020] Furthermore, in the step (1), the service application list P to be deployed includes all applications to be deployed and feature information of the applications, as shown in Table 1 below:

[0021] Table 1 is the characteristic information table of the application

[0022]

[0023] At the same time, the computing node list K includes all the computing nodes that can be deployed and their characteristics, as shown in Table 2:

[0024] Table 2 is the node feature table

[0025] Feature Meaning <![CDATA[C k > Quantization value of the CPU resources of the computing node <![CDATA[R k > Quantization value of the RAM resources of the computing node <![CDATA[S k > Quantization value of the hard disk storage resources of the computing node <![CDATA[G k > Quantization value of the GPU resources of the computing node TR Throughput of the computing node

[0026] Furthermore, the latency described in step (2) refers to the time required for a terminal device to make a complete request to a business application deployed on an edge computing node and receive the calculation result. The specific process is: the device initiates a request, reaches the edge management gateway through a network link, is further routed to the edge computing node or cloud center for calculation, and finally returns the data result to the terminal device. Figure 2 Complete request process for the terminal device.

[0027] In this transmission process, the total delay is mainly composed of two parts: request uplink delay and calculation delay. The calculation formula is:

[0028] D tot =D up +Dcal

[0029] Among them, D tot represents the total time delay, D up represents the requested uplink time, D cal represents the computing time delay.

[0030] Furthermore, the requested uplink time in step (2) mainly includes the total duration required for the device terminal to reach the edge nano-network gateway and then be transmitted to the edge computing node through each routing node. Ignoring the transmission delay between the terminal device and the edge nano-network gateway, and regarding the edge nano-network gateway as a routing node, the transmission rate v between nodes with a total number of nodes denoted as X can be recorded as:

[0031]

[0032] Among them,

[0033] Therefore, the routing time required to apply the W traffic from the terminal device is:

[0034]

[0035] Furthermore, the computing time delay in step (2) mainly targets a certain computing node. When the computing resources and storage resources of the node meet the operation requirements of the service application, the computing time is related to the service data volume and the device throughput rate, and is:

[0036]

[0037] In summary, when the l-th service in step (2) is deployed to the k-th computing node, the time delay of the terminal farthest from the k-th computing node is D lk . The time delay matrix T is:

[0038]

[0039] Among them, the value of t lk is: if the application is deployed on this node and causes the time delay D lk of the farthest terminal to be greater than the maximum time delay D_max allowed by the l-th service application, then t lk =0; otherwise, t lk =1. Therefore, for the l-th service, the nodes that meet the time delay requirements are N delayAllowed,l ={k, if t lk ≤D_max},

[0040] Furthermore, the calculation formula for the weighted computing power value of the computing power node in step (3) is:

[0041] ω = ln(Js + B + St + 2) * Ck *R k *S k *G k (1)

[0042] Further, the standard computing power calculation formula for the computing nodes in step (6) is:

[0043] CP = C k *R k *S k *G k

[0044] The computing node matrix Q of the remaining computing power that meets the business deployment is:

[0045]

[0046] where v lk takes the value: if the computing power of the l-th business deployment at the k-th computing node does not meet the application requirements, then v lk = 0; if the computing power of this node meets the requirements, then v lk = 1.

[0047] The computing power of the computing node meeting the requirements specifically means that the CPU, RAM, hard disk storage, and GPU occupied by all applications deployed on this edge node should not exceed the initial available resources of the node. C kl , R kl , S kl , G kl are the available resources of the CPU, RAM, hard disk storage, and GPU respectively allocated by the k-th computing node to the l-th business, then

[0048]

[0049]

[0050]

[0051]

[0052] Further, the affinity mentioned in step (7) means that some applications have frequent interactions and these interactions affect the application latency, then these two applications should be deployed on the same server or in the same computer room; for some applications that rely heavily on a certain data center and are sensitive to latency, then this application should be deployed as close as possible to the data center.

[0053] Further, when sorting the computing power of each computing node in step (9), the central node needs to be considered later, that is, when ranking, the central node is defaulted to the last position.

[0054] Furthermore, the specific operation steps of updating the remaining resource information of the computing node in step (11) are to subtract the quantified value of resources required for the business application of the currently deployed application during calculation from the computing node on which the business application is newly deployed.

[0055] The application and the corresponding deployed nodes obtained through the above steps are the final deployment solution. The present invention introduces the delay situation that has not been considered in the traditional Kubernetes scheduling method, and makes a certain guarantee for the delay while balanced deployment, so as to ensure the user experience during use as much as possible.

[0056] The advantages of the present invention are as follows:

[0057] The cloud-edge optimized balanced deployment strategy proposed in the present invention comprehensively considers differentiated security requirements such as application importance and affinity, and reasonably selects nodes for deploying business applications. It ensures that business latency is met, allocates computing power as fairly as possible, and optimizes cloud-edge resource utilization, which can effectively cope with resource constraints and real-time requirements in edge computing.

[0058] This solution not only takes into account the balance of application deployment, but also guarantees latency and improves user satisfaction. Compared with the deployment solution with minimum latency, this solution solves the problem of concentrated application deployment, which greatly improves the resource utilization overhead of computing nodes and avoids the situation where some computing nodes have too high resource occupancy and other resource nodes have too low occupancy when users make requests. Compared with the Kubernetes native scheduling model, this solution introduces the feature of latency guarantee. When deploying, it gives priority to deploying computing nodes that meet the maximum latency acceptable to users, reducing the average service processing latency and improving user satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 The figure is a flow chart of the method of the present invention.

[0060] Figure 2 Complete request process diagram for terminal equipment.

[0061] Figure 3 This is a schematic diagram of the cloud-edge-end network experimental environment. DETAILED DESCRIPTION

[0062] The present invention is further described in detail below in conjunction with the accompanying drawings. The examples given are only used to explain the present invention but not to limit the scope of the present invention.

[0063] Example 1:

[0064] (1) Obtain a list of business applications P and a list of computing nodes K to be deployed;

[0065] We designed a cloud-edge-terminal network environment as shown in Figure 3 which includes 7 routing nodes, 4 edge gateways, and 10 computing nodes, where N0 is the central node. As an example, three applications and the user distribution of each application are drawn here.

[0066] In the experimental environment, we assume that the network speeds, computing node computing resource information, and service application attribute information among computing power nodes, routing nodes, and edge gateways are as shown in Table 3 and Table 4.

[0067] Table 3 is the computing node list K

[0068]

[0069] Table 4 is the service application list P

[0070]

[0071]

[0072] (2) Calculate the delay matrix T of the terminals that are farthest from each edge computing node k for each service application; the network transmission rate between nodes is shown in Table 5:

[0073] Table 5 is the network transmission rate table

[0074]

[0075]

[0076] The obtained delay matrix T is as follows:

[0077] n0 n1 n2 n3 n4 n5 n6 n7 n8 n9 n10 app1 1 0 0 0 0 0 0 0 0 0 0 app2 1 1 1 1 1 0 1 0 0 1 1 app3 1 0 0 0 1 0 1 0 0 1 1 app4 1 0 0 0 1 0 1 0 0 1 1 app5 1 0 0 0 0 0 0 0 0 0 0 app6 1 1 1 1 1 1 1 1 1 1 1 app7 1 0 0 0 0 0 0 0 0 0 0 app8 1 0 0 0 1 0 1 0 0 1 1 app9 1 0 0 0 0 0 0 0 0 0 0 app10 0 0 0 0 0 0 0 0 0 0 0 app11 1 0 0 0 1 0 1 0 0 1 1 app12 1 1 1 1 1 1 1 1 1 1 1 app13 1 0 0 0 0 0 1 0 0 1 0 app14 1 0 0 0 0 0 1 0 0 1 0 app15 1 1 1 1 1 1 1 1 0 1 1 app16 1 1 1 1 1 1 1 1 1 1 1 app17 1 1 1 1 1 0 1 1 0 1 1 app18 1 0 0 0 1 0 1 0 0 1 1 app19 0 0 0 0 0 0 0 0 0 0 0 app20 1 1 1 1 1 1 1 1 1 1 1 app21 1 1 1 1 1 1 1 1 1 1 1 app22 1 1 1 1 1 1 1 1 1 1 1 app23 1 0 0 0 1 0 1 0 0 1 0 app24 1 0 0 0 1 0 1 0 0 1 1 app25 1 0 0 0 0 0 0 0 0 0 0

[0078] (3) Sort all the service applications to be deployed in descending order according to the weighted computing power value to obtain a new service application list P_decay as shown in Table 6:

[0079] Table 6 is the new service application list P_decay

[0080]

[0081]

[0082] (4) Determine whether there are still undeployed applications in the service application list P_decay. Based on the new service application list P_decay, it can be seen that there are still undeployed service applications, so we can proceed to step (5).

[0083] (5) Select the undeployed application with the highest ranking from the service application list P_decay, which should be app22, and its attributes are shown in Table 7:

[0084] Table 7 shows app22 and its application attributes

[0085] Application attribute app22 Quantization value of CPU resources (C / time slice) 10 Quantization value of RAM resources (R / kb) 359 Quantization value of hard disk storage resources (S / kb) 65 Quantization value of GPU resources (G) 5 Maximum latency allowed by the user (D_max / s) 0.8 Traffic volume (W / kb) 1729 Construction attribute (Js) 2 Service attribute (B) 1 Status attribute (St) 0 Affinity application

[0086] (6) Calculate the remaining computing power resources of all current computing nodes, count their standard computing power, and obtain a list Q of computing nodes whose remaining computing power meets the requirements for this service deployment, as shown in Table 8 below:

[0087] Table 8 shows the list of computing nodes that meet the computing power of app22

[0088] n0 n1 n2 n3 n4 n5 n6 n7 n8 n9 n10 app22 1 1 1 1 1 1 1 1 1 1 1

[0089] (7) If the currently to-be-deployed application has an affinity with the already deployed applications and the remaining computing power of the computing node where the affinity application is located meets the requirements, then deploy the currently to-be-deployed application on the computing node with the most affinity services. Since app22 has no affinity business applications, this step is skipped, and the deployment node is searched from the perspective of latency and computing power.

[0090] (8) Based on matrix Q and matrix T, find the computing nodes that meet the computing power resources and latency of the currently to-be-deployed application among all computing nodes. From matrix Q and matrix T, it can be seen that all current computing power nodes meet the requirements. Therefore, only one computing node with the largest computing power resource needs to be found to deploy the application.

[0091] (9) Sort the computing nodes that meet the computing power and latency according to the standard value of computing power to obtain a list of computing nodes in descending order, as shown in Table 9:

[0092] Table 9 shows the ranking list of computing node computing power values

[0093] Computing node Computing power value n10 1084473000 n8 960000000 n9 881600000 n7 422240000 n3 393750000 n6 276276000 n5 240994000 n4 236544000 n2 231000000 n1 105000000 n0 12000000000

[0094] Since the application is deployed while trying to avoid deploying it on the central node n0, even though the computing power value of n0 is the largest, it is placed at the last position in the queue. Therefore, app22 is deployed on the computing node n10.

[0095] (11) Update the remaining resource information of the computing node, remove the currently deployed application app22 from the business application list P_decay, and return to step (4) to continue deploying the remaining business applications.

[0096] After the above loop steps, the final deployment plan is as shown in Table 10 below:

[0097] Table 10 shows the application deployment results

[0098] Application name Deployment node app1 n4 app2 n3 app3 n6 app4 n6 app5 n0 app6 n8 app7 n6 app8 n4 app9 n0 app10 n0 app11 n10 app12 n8 app13 n9 app14 n9 app15 n5 app16 n7 app17 n7 app18 n9 app19 n0 app20 n2 app21 n8 app22 n10 app23 n4 app24 n10 app25 n0

[0099] The comparison table of the results of other model deployments is shown in Table 11 as follows:

[0100] Table 11 is the evaluation and comparison table of the deployment results of each model

[0101]

[0102]

[0103] It can be found that our deployment method has the highest score in terms of user satisfaction. The satisfaction with the minimum latency is actually lower than that of this model. This is because in areas with a large number of users and tight resources, the minimum latency model cannot balance the resource utilization rate in a small area, resulting in some applications choosing nodes with the minimum latency but still exceeding the user's tolerance range because they cannot be deployed nearby. This drawback is also reflected in the fact that the resource utilization rate overhead of the minimum latency model is the highest. In terms of the average service processing latency and resource utilization rate overhead, our model has a similar resource utilization rate overhead to the balanced deployment model, but the average latency is reduced by 10%. In summary, the comprehensive effect of our deployment method is the best.

[0104] In this solution, the formula for setting the weight in the deployment order of business applications is: ln(Js + B + St + 2), but introducing other weight formulas such as: can also replace this solution.

[0105] Although specific embodiments of the present invention are disclosed for illustrative purposes, which are intended to help understand the content of the present invention and implement it accordingly, those skilled in the art can understand that: within the spirit and scope of the present invention and the appended claims, various substitutions, changes, and modifications are possible. Therefore, the present invention should not be limited to the content disclosed in the best embodiments, and the scope of protection claimed by the present invention shall be subject to the scope defined by the claims.

Claims

1. A cloud-edge collaborative optimization and equilibrium deployment method with guaranteed latency, the steps of which include: (1) Obtain a list P of business applications to be deployed and a list K of computing nodes for deploying the business applications in the list P of business applications; Obtain the location information of the terminals using the business applications; (2) For each edge computing node in the node list K, calculate the latency for the terminal farthest from the edge computing node to access the business application on the edge computing node, obtaining a latency matrix T; (3) According to the characteristic information of each business application in the application list P, calculate the weighted computing power value required for the corresponding business application; sort the business applications in descending order according to the weighted computing power value required for the business application, obtaining a business application list P_decay; (4) Determine whether there are still undeployed business applications in the business application list P_decay. If so, go to step (5); otherwise, end the deployment process; (5) Select the undeployed business application p with the highest ranking from the business application list P_decay, and obtain the various resources it needs to occupy; (6) Calculate the remaining computing power resources of each edge computing node in the current node list K, and calculate its standard computing power value, obtaining an edge computing node matrix Q that meets the requirements for deploying the business application p; (7) If the currently to-be-deployed business application p and the already-deployed business application q are affinity, and the remaining computing power of the edge computing node k where the business application q is located meets the requirements of the business application p, then deploy the business application p on the edge computing node k, and go to step (11); otherwise, go to step (8); (8) Based on the edge computing node matrix Q and the latency matrix T, find the edge computing nodes that meet the computing power resources and latency requirements of the business application p among the edge computing nodes. If found, go to step (9); if only the edge computing nodes that meet the computing power resources required by the business application p but do not meet the latency requirements of the business application p are found, then go to step (10); (9) Sort the edge computing nodes that meet the computing power resources and latency requirements of the business application p in descending order according to the standard computing power value, obtaining a descending-ordered edge computing node list; deploy the business application p on the edge computing node ranked first in the edge computing node list, and go to step (11); (10) Calculate the average latency of each edge computing node that meets the computing power resources required by the business application p but does not meet the latency requirements of the business application p to each terminal using the business application p, and deploy the business application p on the edge computing node with the minimum average latency, and go to step (11); (11) Update the remaining resource information of each edge computing node in the node list K, remove the currently deployed application from the business application list P_decay, and return to step (4).

2. The method according to claim 1, wherein The characteristic information of the business application includes the CPU resource quantization value C, RAM resource quantization value R, hard disk storage resource quantization value S, GPU resource quantization value G required for business application computing, and the maximum latency D allowed for the business application on the terminal max , the size W of the service processing data for one request of the business application calculation, the user attribute U, construction attribute Js, business attribute B, status attribute St of the business application, and the application name Affinity with affinity; The computing power resources of the edge computing node include: the CPU resource quantization value C k , the RAM resource quantization value R k , the hard disk storage resource quantization value S k , the GPU resource quantization value G k and the throughput TR.

3. The method according to claim 2, wherein The weighted computing power value ω required for the business application = ln(Js + B + St + 2) * C k *R k *S k *G k .

4. The method according to claim 2, wherein The standard computing power value CP of the edge computing node is CP = C k *R k *S k *G k .

5. The method according to claim 1, wherein The time delay matrix where the value of the element in the l-th row and k-th column is determined according to D lk D lk is the time delay of the terminal farthest from the edge computing node k when the service application l is deployed to the edge computing node k; if the time delay D lk is greater than the maximum allowable time delay D_max of the service application l, then t lk = 0; Otherwise, t lk = 1.

6. The method according to claim 1, characterized in that, If the currently to-be-deployed business application p and multiple already-deployed business applications are affinity, then deploy the business application p to the edge computing node that has the most affinity with it and whose remaining computing power meets the requirements of the business application p.

7. The method according to claim 1, wherein Determine whether there is an affinity between the two service applications according to the characteristic information of the two service applications.

Images