Big Data Oriented Light-Load Embedded Performance Modeling

With increasing development of big data, the performance assessment and optimization face with a big challenge. The traditional methods widely use delivery-testing-analysis-solving (DTAS) ring. In big data area, big data environment is necessary for the testing phase in DTAS, which results in the big cost in both time and hardware. This paper proposes the big data oriented light-load embedded performance modeling. It ascertains the performance criteria to set the Capacity and Performance (C&P) factors. These factors will be embedded into the software with an on-off switch during the architecture, design and developing phases before DTAS phase. After the software coding done with embedded C&P factors, a small traffic load is run to collect the C&P data. The collected data will be used for the performance bottleneck finding, performance optimization, and forecasting the capacity and performance for various customers' scenarios. Since the data easily help locate the issue, the required running traffic is small, and the problem solving is done before the traditional DTAS, this study is more suitable for the big data application. It can save more than 50% of time, decrease the software development efforts, and reduce the lab resources occupation. Finally, the proposed method is employed in the real prototype of an Internet of Things application, obtains the better capacity and performance, and the experiment data verify its effectiveness.