Investigation of Reconfigurable FPGA Design for Processing Big Data Streams
Usamah Algemili · 2016
Big Data situation has placed a tremendous pressure on the existing computational models. The challenges of Big Data call for a new approach to solve both software and hardware problems. Streaming applications is a form of on-demand software distribution. In streaming scenarios, only essential portions of an application's code need to be installed on the system, while the receiver performs the main operations. The necessary code and files are delivered over the network as, and when, they are required. The hardware architecture plays an important role in improving the efficiency of a streaming system. The variance of hardware performance on different HW architectures is quite interesting. Previous work confirms that the CPUs, GPUs, and FPGAs are performing differently on specific applications. The previous efforts of hardware benchmarking show that GPUs outperformed the other platforms in terms of execution time. CPUs outperformed in overall execution combined with transfer time. FPGAs outperformed for fixed algorithms using streaming [1]. Hence, this paper evaluates the performance of streaming applications on a pipelined FPGA design. In the context of real-time processing, it elects one of the Big Data streaming problems that gets a candidate for majority element on-the-fly, that is Moore's Voting Algorithm. The performance analysis of Moore's algorithm on FPGA highlights a noticeable improvement by using a pipelining architecture.