داده های بزرگ در محاسبات ابری: چشم انداز مدیریت منابع Big Data in Cloud Computing: A Resource Management Perspective

1. Introduction We live in the information age, and an important measurement of present times is the amount of data that is generated anywhere around us. Data is becoming increasingly valuable. Enterprises are aiming at unlocking data’s hidden potential and deliver competitive advantage [1]. Stratistics MRC projected that the data analytics and Hadoop market, which accounted for $8.48 billion in 2015, is expected to reach at $99.31 billion by 2022 [2]. Te global big data market has estimated that it will jump from $14.87 billion in 2013 to $46.34 billion in 2018 [3]. Gartner has predicted that data will grow by 800 percent over the next fve years and 80 percent of the data will be unstructured (e-mails, documents, audio, video, and social media content) and 20 percent will be structured (e-commerce transactions and contact information) [1]. Today’s largest scientifc institution, CERN, produces over 200 PB of data per year in the Large Hadron Collider project (as of 2017). Te amount of generated data on the Internet has already exceeded 2.5 exabytes per day. Within one minute, 400 hours of videos are uploaded on YouTube, 3.6 million Google searches are conducted worldwide each minute of every day, more than 656 million tweets are shared on Twitter, and more than 6.5 million pictures are shared on Instagram each day. When a dataset becomes so large that its storage and processing become challenging due to the constraints of existing tools and resources, the dataset is referred to as big data [4, 5]. It is the frst part of the journey towards delivering decision-making insights. But instead of focusing on people, this process utilizes a much more powerful and evolving technology, given the latest breakthroughs in this feld, to quickly analyze huge streams of data, from a variety of sources, and to produce one single stream of useful knowledge [6]. Big data applications might be viewed as the advancement of parallel computing, but with the important exception of the scale. Te scale is the necessity arising from the nature of the target issues: data dimensions largely exceed conventional storage units, the level of parallelism needed to perform computation within a strict deadline is high, and obtaining fnal results requires the aggregation of large numbers of partial results.Te scale factor, in this case, does not only have the same efect that it has in classical parallel computing, but it surges towards a dimension in which automated resource management and its exploitation are of signifcant value [7].