Intent-based resource matching strategy in cloud

Abstract

Accurate and efficient resource allocation based on user intent is an important issue in a large-scale, distributed environment, such as cloud computing. Although a large number of cloud resource matching models he been proposed, these models do not consider the interests of user and cloud service provider objectively and fairly. Thus, a novel resource matching strategy that regulates multiattribute matching between cloud resources and tasks is proposed in this study. Tasks and resources are initially clustered on the basis of the attribute characteristics to reduce the scope of resource retrieval. Then, the tasks are matched to the appropriate resources in terms of the strict bilateral matching algorithm to improve the satisfaction of both parties. Finally, a series of experiments are reported to show the effectiveness of the algorithm. Experimental results conclusively demonstrate that our proposed methods can availably decrease the scheduling overhead and improve the overall satisfaction of both parties simultaneously.

Introduction

Cloud computing is a type of new network service mode based on virtualization and distributed computing technology it has the characteristics of dynamic expansion, resource sharing, broadband access and powerful computing and storage capabilities [1], [2], [3], [4]. With the rapid development of the mobile Internet, the massive data generated by the proliferation of applications, such as telecommuting, electronic medical care, interactive games, virtual reality, and mobile payment, all rely on the processing power of the cloud platform. Thus, accurate and efficient resource scheduling based on user intent is the key to the powerful processing capabilities of the cloud platform. In the problem of task scheduling and resource allocating (TSRA) under the cloud environment, the scheduling overhead and the quality of service (QoS) have always been the focus of research.

Essentially, the TSRA problem is a difficult NP-hard combinatorial optimization problem [5]. At present, studies on TSRA are mainly based on the improvement or optimization of scheduling algorithms. The efficiency of the algorithm and the service price in dynamic task scheduling were optimized in [6], [7]. In [8], the Pareto based on a fruit fly optimization algorithm was proposed to solve the multiobjective TSRA problem, which showed its effectiveness compared with the improved differential evolution algorithm. Zhang [9] proposed a task scheduling and computing resource allocation algorithm called NG-TSRA based on noncooperative game and showed the existence of its Nash equilibrium. In this manner, the energy efficiency of heterogeneous servers in the cloud computing system can be optimized. Cloud applications [10] are classified into Computing-, I/O-, and bandwidth-intensive models by predicting the demand characteristics of applications in reducing scheduling overhead. In [11], particle swarm optimization was used to generate data sets for training classifiers. Then, a comparative study was performed to determine which classification algorithm is best suited to virtual machines, considering the efficiency and the reliability of the cloud network simultaneously.

In terms of QoS, Li [12] proposed an elastic hybrid mobile cloud resource allocation model, which enabled intelligent scheduling to select appropriate cloud resources and jointly optimized mobile user experience under the constraints of available resources and users’ QoS. Liu [13] proposed a resource management framework model to ensure the QoS of the cloud computing system. Alsarhan [14] presented a new cloud computing service level protocol framework using reinforcement learning to control the rent-seeking strategy of virtual machines. This method can adapt to the changes in the system to ensure the QoS of all clients. By mapping QoS to overload probability, Ran et al. [15] proposed a dynamic instance determination strategy based on large deviation principle. This method calculates the minimum instance of the coming application requirement under the condition that the overload probability is lower than the expected threshold. At the same time, the number of retained instances was calculated by applying an autoregressive model. In view of clustering-based resource allocation, optimal cloudlet resources were obtained in [16] to meet mobile devices$\prime$ operation by clustering the mobile cloudlets. Task scheduling of cloud tasks under partial correlation was studied in [17], given that the clustering of tasks was based on an ideal situation, that is, the tasks were completely related in the past task scheduling. In terms of auction-based resource allocation, a real-time and efficient cloud resource online auction model was designed in [18], where users bid to acquire cloud resources and customize virtual machines that compose their own requirements. The requirements of users and cloud service provider (CSP) were considered in [19], [20], and a cloud resource allocation model based on bilateral auctions was proposed.

However, the resource configuration problem is to fulfil the best mapping of resource and task queues in a specific state to satisfy the intents of users and service providers. The above literature failed to consider the characteristics of different tasks and the comprehensive satisfaction (CS) of user and provider; the satisfaction of the cloud service in [21], [22] was poor. On this basis, studying the characteristics and intents of different tasks and resources in a matching strategy is necessary.

A novel intent-based resource matching strategy is proposed in this study. This approach focuses on the matching between resources and tasks and treats the interests of both parties fairly. The contributions of this study are as follows.

• A bilateral clustering method is proposed in reducing the matching scale of resources and tasks to improve the accuracy and efficiency of matching.

• A strict bilateral matching (SBM) model is introduced between the matching task and resource classes. This model can improve the overall satisfaction of both parties in resource allocation by ensuring to meet the minimum satisfaction degree.

The remainder of this paper is organized as follows. The theoretical basis of TSRA is analyzed and the system model of the proposed approach is introduced in Section 2. Section 3 presents the intent-based resource matching strategy on the basis of certain reasonable assumptions. The performance of the algorithms is shown in Section 4. Section 5 concludes this study.