Parallel Computation For File Metadata Cube In Cloud Computing Environments

As the emergence of cloud computing, the rapid growth of the sizes of storagesystems leads to the performance bottleneck of data management. How to efficientlyretrieve information from large amounts of data and how to store different types of datainto devices are important to the performance of storage systems. File metadatademonstrate the feature of files and can be used to support system management.In the storage community, the related work about metadata focuses on the schemes ofefficient metadata queries, which support efficient metadata analysis and processing.Statistical information of file systems, such as file size distribution, existing time andusers’ behaviors, can directly support the design, metadata query, and storage managementto the file systems. Data cube, a model widely used in database community, is animportant tool to Online Analytical Processing (OLAP), Decision Support System (DSS)and Business Intelligence (BI). Data cube is an abstract multidimensional model and canbe used to summarize and analyze the data from different viewpoints, so it is a suitabletool for system administrators. Traditional data cube computing algorithms are I/O andCPU-intensive and memory consuming, which are challenging. In this paper weimplemented Metadata Cube (MD-CUBE) system, which makes use of a datacompression method to significantly reduce the I/O and memory costs. Moreover, itleverages parallel computing on a multicore platform to accelerate the metadata cubeconstruction.Through collecting the metadata in tens of storage nodes in a cluster and analyze thesemetadata with cube, the results show that metadata cube can accurately and efficientlyexhibit the global state of file systems. So metadata cube can be used to support metadataqueries, metadata mining, storage management as well as system states analysis.