CN105959412A - Cloud service resource allocation analysis method based on queue mining - Google Patents

Abstract

Translated fromChinese

一种基于队列挖掘的云服务资源分配分析方法，适合处理顾客和资源交互情况下的云服务流程中复杂资源的分配问题。考虑顾客和资源交互情况下的顾客服务日志和资源服务日志，首先直接利用两类服务日志，通过顾客和资源的类别进行资源匹配，并通过损失函数L(d,π(x))进行优化得到资源初次分配方案π(X)；然后基于顾客服务日志提出基于资源服务日志的队列挖掘方法，获得到顾客队列信息，包括顾客队列长度q(t)、顾客时延时间h(t)和放弃队列时间等。以及资源分配的决策信息，包括顾客接受服务时间和资源服务状态F_STA。在此基础上，提出基于服务进程属性的资源分配算法，进行资源再次分配，当队列长度和时延时间达到阈值时，通过优化分析得到最终的云服务资源分配方案。

An analysis method of cloud service resource allocation based on queue mining is suitable for dealing with the complex resource allocation problem in the cloud service process under the interaction between customers and resources. Considering the customer service log and resource service log in the case of interaction between customers and resources, first directly use the two types of service logs, match resources through the categories of customers and resources, and optimize through the loss function L(d,π(x)) to obtain The initial resource allocation scheme π(X); then based on the customer service log, a queue mining method based on the resource service log is proposed, and the customer queue information is obtained, including the customer queue length q(t), the customer delay time h(t) and the abandonment queue time Wait. and resource allocation decision information, including customer service time and resource service status F_STA . On this basis, a resource allocation algorithm based on service process attributes is proposed to re-allocate resources. When the queue length and delay time reach the threshold, the final cloud service resource allocation scheme is obtained through optimization analysis.

Description

Translated fromChinese

一种基于队列挖掘的云服务资源分配方法A method of cloud service resource allocation based on queue mining

技术领域technical field

本发明属于云计算环境下服务资源调度领域，涉及到存在时延的业务过程挖掘和相关资源的分配，特别适用于云计算环境下，存在资源与顾客交互情况下的资源分配问题。The invention belongs to the field of service resource scheduling in a cloud computing environment, relates to time-delayed business process mining and allocation of related resources, and is particularly suitable for resource allocation in the cloud computing environment where resources interact with customers.

背景技术Background technique

在云计算环境下，服务过程是一种特殊的业务过程，它是包括云平台资源和顾客交互的业务过程。随着社会的进步，云服务过程分析被更广泛的应用在越来越多的领域中，同时针对云服务过程的研究在提高服务质量、提高企业服务效率和降低服务成本等方面都发挥重要的作用。目前关于服务过程研究方法大部分是从事件日志入手，主要关注顾客方进行分析，关注云计算平台下的资源方面的研究较少，缺乏顾客和云资源交互后的分析方法，造成服务过程分析不全面，进而不能得到云服务过程中合理的资源分配方案。也有综合顾客和资源进行的分析，但是复杂度较大，同时在公平性和分配效率测评下很难得到合适的分配方案。In the cloud computing environment, the service process is a special business process, which includes the interaction between cloud platform resources and customers. With the progress of society, cloud service process analysis is widely used in more and more fields, and the research on cloud service process plays an important role in improving service quality, improving enterprise service efficiency and reducing service cost. effect. At present, most of the research methods on the service process start from the event log, mainly focusing on the customer side for analysis, and less research on the resources under the cloud computing platform, lack of analysis methods after the interaction between customers and cloud resources, resulting in insufficient service process analysis. Comprehensive, and thus cannot obtain a reasonable resource allocation plan in the cloud service process. There is also a comprehensive analysis of customers and resources, but the complexity is relatively large, and it is difficult to obtain a suitable allocation plan under the evaluation of fairness and allocation efficiency.

因此，面对顾客和资源相交互的云服务过程的研究，有必要在现有过程挖掘技术的基础上，考虑云平台资源方对云服务过程的影响。提出基于资源服务日志的队列挖掘技术，并将其运用到资源分配的研究中。利用通过队列挖掘得到的队列信息和决策信息，对初次分配方案进行补充，得到最终的分配方案。Therefore, in the face of the research on the cloud service process in which customers and resources interact, it is necessary to consider the impact of cloud platform resources on the cloud service process on the basis of existing process mining techniques. A queue mining technology based on resource service logs is proposed and applied to the research of resource allocation. Using the queue information and decision information obtained through queue mining, the initial allocation scheme is supplemented to obtain the final allocation scheme.

发明内容Contents of the invention

本发明所要解决的技术问题是：提供一种先通过云服务系统的领域约束，进行云服务资源的初次分配，保证系统中云资源分配的公平性，然后根据云服务流程中顾客服务日志和资源服务日志进行队列挖掘，得到服务过程的队列信息：顾客队列长度q(t)、顾客时延时间h(t)和资源分配的决策信息：顾客接受服务时间和资源服务状态F_STA，利用这些队列信息和决策信息对资源进行第二次分配，提高资源分配的效率，最后综合两次分配得到最后的资源分配方案的方法。The technical problem to be solved by the present invention is to provide a method to firstly allocate cloud service resources through the domain constraints of the cloud service system to ensure the fairness of cloud resource allocation in the system, and then according to the customer service logs and resources in the cloud service process Queue mining is performed on the service log to obtain the queue information of the service process: customer queue length q(t), customer delay time h(t) and resource allocation decision information: customer service time and the resource service status F_STA , use these queue information and decision information to allocate resources for the second time, improve the efficiency of resource allocation, and finally combine the two allocations to obtain the final resource allocation scheme.

为解决以上技术问题，本发明采用如下的技术方案：In order to solve the above technical problems, the present invention adopts the following technical solutions:

首先，定义云服务流程中顾客服务日志C.S-Log＝(S,G,α_C＝{τ,η,ε})和资源服务日志R.S-Log＝(S,G,α_R＝{τ,η,σ,φ,δ})，其中S表示服务事件(包括顾客和资源)的集合；G表示服务路径的集合；α_C表示顾客的属性集合包括：时间戳属性τ、顾客类别属性η和顾客服务状态属性ε；α_R表示资源的属性集合包括：时间戳属性τ、顾客类别属性η、资源类别属性σ、资源状态属性φ和资源状态转换属性δ。资源服务日志中的顾客的时间戳属性和类别属性区别于顾客日志，它指的是接受资源服务的顾客的属性。First, define customer service log CS-Log=(S,G,α_C ={τ,η,ε}) and resource service log RS-Log=(S,G,α_R ={τ,η) in the cloud service process ,σ,φ,δ}), where S represents the set of service events (including customers and resources); G represents the set of service paths; α_C represents the set of customer attributes including: time stamp attribute τ, customer category attribute η and customer Service state attribute ε; α_R indicates the resource attribute set includes: time stamp attribute τ, customer category attribute η, resource category attribute σ, resource status attribute φ and resource status transition attribute δ. The time stamp attribute and category attribute of the customer in the resource service log are different from the customer log, which refers to the attributes of the customer receiving the resource service.

其次，利用两类服务日志进行基于云服务流程的队列挖掘。基于顾客服务日志得到服务过程的队列信息：顾客队列长度顾客时延其中顾客放弃队列时间基于资源服务日志得到资源分配的决策信息：资源状态信息F_STA＝φ[argmin(t-τ(S))]，其中资源为顾客提供服务的时间最后综合队列信息和决策信息给出资源分配的决策变量：Secondly, two types of service logs are used for queue mining based on cloud service processes. Get the queue information of the service process based on the customer service log: customer queue length customer delay in Customer Abandoned Queue Time The decision information of resource allocation is obtained based on resource service logs: resource status information F_STA =φ[argmin(t-τ(S))], where The time a resource provides service to a customer Finally, the decision variables for resource allocation are given by combining queue information and decision information:

最后，基于队列挖掘技术给出云服务流程中资源分配的办法。首先严格按照顾客和资源的类别属性进行资源的匹配，再利用损失函数L(d,π(x))对资源分配方案进行优化，得到基于领域约束的角度的资源分配方案π(X)。然后利用队列挖掘得到的队列信息和决策信息进行第二次分配，包括第一部分是基于服务日志中顾客固有的属性：根据顾客类别属性η(s)和资源服务状态属性F_STA进行初次分配，第二部分的分配是基于队列挖掘得到的队列信息进行的：顾客时延时间达到阈值时，优先进行资源分配等。最后综合得到资源的分配方案。Finally, based on the queue mining technology, the resource allocation method in the cloud service process is given. Firstly, resources are matched strictly according to the category attributes of customers and resources, and then the resource allocation plan is optimized by using the loss function L(d, π(x)), and the resource allocation plan π(X) based on the perspective of domain constraints is obtained. Then use the queue information and decision information obtained by queue mining to carry out the second assignment, including the first part based on the inherent attributes of customers in the service log: the first assignment is made according to the customer category attribute η(s) and the resource service status attribute F_STA , the second The allocation of the second part is based on the queue information obtained from queue mining: customer delay time When the threshold is reached, priority is given to resource allocation, etc. Finally, a resource allocation plan is synthesized.

附图说明Description of drawings

图1是本发明的流程模型的结构图。Fig. 1 is a structural diagram of the process model of the present invention.

图2是本发明的队列挖掘的流程图。Fig. 2 is a flowchart of queue mining in the present invention.

图3是本发明的资源分配的流程图。Fig. 3 is a flowchart of resource allocation in the present invention.

具体实施方式detailed description

本发明提出基于队列挖掘的云服务流程资源分配方法。首先严格根据顾客类型进行资源的匹配，并用损失函数L(d,π(x))进行检测优化，得到关于领域约束角度的初次分配方案π(X)，它可以保证资源分配的公平性；其次依据服务日志，运用行为轮廓挖掘技术得到云服务流程的队列信息q(t)、h(t)和决策信息顾客接受服务时间和资源服务状态F_STA，运用这些信息进行第二次分配，考虑云服务流程中队列信息对资源分配的影响，提高资源分配的效率。综合两次分配得到最终的资源分配方案。The invention proposes a method for allocating cloud service process resources based on queue mining. Firstly, resources are matched strictly according to customer types, and the loss function L(d, π(x)) is used for detection and optimization to obtain the initial allocation plan π(X) from the perspective of domain constraints, which can ensure the fairness of resource allocation; secondly According to the service log, the queue information q(t) and h(t) of the cloud service process and the decision information customer acceptance service time are obtained by using the behavior profile mining technology and resource service status F_STA , use these information for the second allocation, consider the impact of queue information on resource allocation in the cloud service process, and improve the efficiency of resource allocation. The final resource allocation scheme is obtained by combining the two allocations.

以下结合附图对本发明作进一步的说明。The present invention will be further described below in conjunction with the accompanying drawings.

图1是本发明的流程模型的结构图，包括根据服务系统领域的约束角度进行第一次资源分配，再整理日志序列得到顾客服务日志和资源服务日志，基于队列观点对服务事件进行队列挖掘，得到云服务流程中的队列信息和资源分配的决策信息，在初次资源分配方案的基础上，进行第二次资源分配，得到最终的分配方法。Fig. 1 is a structural diagram of the process model of the present invention, including the resource allocation for the first time according to the constraint angle of the service system field, reorganizing the log sequence to obtain the customer service log and resource service log, and performing queue mining on service events based on the queue view, The queue information and resource allocation decision information in the cloud service process are obtained, and the second resource allocation is performed on the basis of the initial resource allocation plan to obtain the final allocation method.

图2是本发明的队列挖掘的流程图，具体来说包括，根据服务流程二元性定义云服务流程中的顾客事件和资源事件，并整理得到相应的服务日志，再利用队列挖掘技术对事件进行分析得到队列信息：顾客队列长度q(t)、顾客在队列中等待的最长时延h(t)和放弃队列时间决策信息：资源服务时间和资源服务状态F_STA。Fig. 2 is a flow chart of the queue mining of the present invention, which specifically includes defining customer events and resource events in the cloud service process according to the duality of the service process, and sorting out the corresponding service logs, and then using the queue mining technology to analyze the events Perform analysis to obtain queue information: customer queue length q(t), customer waiting in the queue for the longest delay h(t) and abandonment queue time Decision Information: Resource Service Time and resource service status F_STA .

图3是本发明基于队列挖掘的资源分配方法的流程图，它包含分别基于顾客服务日志和资源服务日志的队列挖掘，整理出队列信息和决策信息，在基于领域约束得到的初次分配方案π(X)的基础上，分别根据顾客队列长度q(t)、顾客在队列中等待的最长时延h(t)和资源服务状态F_STA进行二次分配。最后得到该云服务流程的资源分配方案。Fig. 3 is a flowchart of the resource allocation method based on queue mining in the present invention, which includes queue mining based on customer service logs and resource service logs respectively, sorting out queue information and decision information, and obtaining the initial allocation scheme π( On the basis of X), secondary allocation is carried out according to the customer queue length q(t), the longest time delay h(t) of customers waiting in the queue and the resource service status F_STA . Finally, the resource allocation scheme of the cloud service process is obtained.

Claims (3)

1. the cloud service resource allocation methods excavated based on queue, including queue method for digging based on serve log, based on teamThe resource allocation methods that row excavate, it is characterised in that: according to resource in cloud service flow process and client mutual in the case of, based on GuVisitor's serve log obtains about queue length, customer service time delay and the queuing message abandoning Queue time；Based on resource serviceDaily record carries out queue excavation, obtains about the queue letter accepting service time and resource service state in service process about clientBreath.Resource allocation methods includes two steps, the first time distribution carried out based on domain constraint, it is ensured that distributional equity, further according toThe queuing message obtained is excavated in queue, carries out second time resource distribution.Final allocative decision is obtained in conjunction with two sub-distribution, it is ensured that pointAllocative efficiency is improved while joining fairness.

The queue of cloud service daily record the most according to claim 1 is excavated, it is characterised in that: consider client in cloud service flow processThe situation mutual with resource, serve log being arranged is customer service daily record and resource service daily record, then the serve log obtainedCarry out queue excavation, obtain client's queue length q (t), client's decay time h (t) and abandon Queue timeDistribute for resourceQueuing message is provided；Consider resource side's impact on client's queue in cloud service flow process, obtain accepting service time about clientWith resource service state F_STA, provide decision information for resource distribution.

The resource allocation methods excavated based on queue the most according to claim 1, it is characterised in that: it is primarily based on field aboutBundle condition, carries out resource matched according to the classification of client and resource, and recycling loss function L (d, π (x)) is optimized and is providedAllocative decision π (X) that source is first；Excavate the queuing message obtained further according to queue, carry out second time and distribute: utilize queue lengthThe queuing messages such as the threshold value with customer service time delay, are allocated in conjunction with decision informations such as resource service states, obtain optimizationCloud service Resource Allocation Formula.