Pipeline-based parallel framework for mass file processing

Currently, there exists billions of files on the Internet, such as pictures, web pages, audio and video files, etc., and the number is still growing rapidly. These huge amount of files need to be processed by some applications as quickly as possible with parallel processing. With the increasing of cores in processors, parallel programming becomes more complex. The behavior that multiple parallel processes/threads access files simultaneously may interfere with each other and cause extra performance loss. Consequently, this paper proposes a pipeline-based parallel framework for mass file processing, in which file processing is divided into multiple stages to compose a pipeline. Files flow through these stages one by one, and the interferences in file-accessing are avoided. Moreover, the parallel programming can be simplified by means of parallel frameworks and programming interfaces. Experiments with one real-world application and some micro-benchmarks show that the framework can efficiently improve system performance.