Efficient auto-scaling scheme for rapid storage service using many-core of desktop storage virtualization based on IoT

Abstract

Following the progressive development of IT technology, on-premise IT resources have been shifted to cloud computing environments. The principle reason for this change in IT resource-composing environments is that cloud computing services allow IT resources to be used as and when necessary, which means without buying hardware equipment. For this reason, studies on diverse aspects are being conducted for better security, rapidity, availability, reliability, and elasticity of cloud computing. Among the virtualization technologies that are basic for cloud computing, desktop storage virtualization (DSV) is composed of distributed legacy desktop personal computers. In DSV environments, clustering by unavailable state time and auto-scaling for storage provision as requested by users are considered very important. In addition, deferred processing for analysis of desktop PC performance states in DSV environments to select an appropriate desktop PC is directly connected to the quality of service (QoS). Although diverse algorithms and schemes for clustering and auto-scaling have been developed to this end, they have limited performance or have been made without considering DSV environments. Consequently, large amounts of deferred processing time are required. In the present paper, an efficient auto-scaling scheme (EAS) is proposed that minimizes deferred processing time in Internet of Things (IoT) environments by using many-cores of the GPU for clustering and auto-scaling in DSV environments. The EAS provides higher QoS to storage users compared to the CPU by mapping the information of numerous distributed desktop PCs on individual threads of the GPU and processing the information in parallel.

Introduction

Following the rapid development of IT technology, on-premise IT resources have shifted to cloud computing environments. Clouds enable accessing IT resources as and when necessary. These clouds are divided into diverse delivery models according to IT resource service provisions and into deployment models depending on cloud access configuration methods. Representative services provided by cloud delivery models include Infrastructure as a Service (IaaS), Platform as a Service (PaaS), and Software as a Service (SaaS). Representative access configurations of cloud deployment models include public clouds, private clouds, hybrid clouds, and community clouds. To configure these clouds, virtualization is utilized as a foundation technology [1], [2], [3], [4], [5], [6]. In the case of these clouds, studies on diverse aspects are being conducted for better security, rapidity, availability, resiliency, and elasticity [7], [8], [9], [10], [11], [12], [13], [14], [15], [16]. Virtualization, which is a basic technology of cloud computing, includes application virtualization, hardware virtualization, desktop virtualization, network virtualization, server virtualization, and storage virtualization. Among them, desktop storage virtualization (DSV) integrates and distributes unavailable IT resources of distributed legacy desktop personal computers (PCs). Due to the capacity of heterogeneous desktop PCs, clustering and auto-scaling of idle and unavailable storage resource is considered very important in DSV [1], [3], [6], [17], [18], [19], [20], [21], [22]. Such clustering and auto-scaling is directly connected to the quality of service (QoS) due to the deferred processing of on-demand storage of many users occurring in unspecified time zones. Therefore, from the viewpoint of clustering, a number of algorithms have been developed, including centroid-based clustering [23], distributed-based clustering [24], density-based clustering [25], and connectivity-based clustering [26]. In addition, although diverse studies and schemes have been developed for auto-scaling, these algorithms and schemes cause large delay times because DSV environments are not considered.

In the present paper, an efficient auto-scaling scheme (EAS) is proposed that minimizes deferred processing time in Internet of Things (IoT) environments using many-cores of the GPU. The EAS provides higher QoS to storage users compared to a CPU by mapping the information of numerous distributed desktop PCs on individual threads of the GPU and processing the information in parallel for clustering and auto-scaling in DSV environments.

The present paper is composed of the following sections:

In Section 2, resource integration schemes for existing cloud storage services, schemes for auto-scaling, and CUDA for using a GPU are examined. In Section 3, an operation scheme for GPU-based auto-scaling of our proposed EAS is explained. In 4 Design of, 5 Implementation of, the design for and the implementation of the EAS application in DSV environments is described, respectively. In Section 6, the processing speed for storage services when auto-scaling is performed in EAS-based DSV environments is evaluated. Finally, a summary of overall conclusions and a suggestion for future studies are presented in Section 7.