Multi-Objective Data Placement for Workflow Management in Cloud Infrastructure Using NSGA-II

The cloud computing paradigm provides massive storage and rich computing resources for workflow deployment and implementation. Nevertheless, workflow applications (e.g., meteorological prediction and financial analysis) are usually data intensive, and substantial data resources with privacy information tend to be accessed during the workflow implementation. Therefore, it remains challenging to design a data placement method for seeking tradeoffs among multiple performance metrics, i.e., resource usage, data acquisition time, and energy cost, while avoiding privacy conflicts of information-overlapping datasets for workflow implementation of the cloud infrastructure. To address this challenge, a multi-objective data placement method for workflow management in the cloud infrastructure with privacy protection is proposed in this paper. Technically, the BCube topology is adopted to establish the resource model in the cloud infrastructure, and the potential privacy conflicts of datasets required for workflow implementation are analyzed. Then, a non-dominated sorting genetic algorithm II is leveraged to promote the resource usage, reduce the data acquisition time, and optimize the energy cost of the cloud infrastructure, while achieving the privacy protection for data placement. Finally, experimental evaluations demonstrate that the performance of the cloud infrastructure is optimized for workflow management.

I. Introduction

IN recent years, workflow technologies are propitious to centralized information management and they automatically process the scientific and industrial applications from various fields, including medicine, education and finance, etc. [1]. Benefit from the efficiency and convenience of workflow technologies, a vast amount of applications modeled by the workflows are tremendously evolved, and the terabytes or petabytes of data resources are required by workflow implementation [2]. The execution performance of the workflow applications depends on the processing of the tasks as well as adequate supply of data resources to a great extent [3], [4]. In the big data era, the workflow applications become more complicate due to the exponentially growing scale of required datasets, the increasing amount of data sources and the complicated executive components of the applications. The execution performance of the workflow applications is seriously affected in the event of the insufficient supplement of processing and storage resources. Accordingly, a large pool of storage and processing resources offered by the cloud infrastructure is engaged to be provisioned for implementing the workflow applications [3], [5].