Programming Model for Non-Volatile Memory toward Extreme Computing

2017 Fiscal Year Final Research Report

• Project/Area Number: 26540050
• Research Category: Grant-in-Aid for Challenging Exploratory Research
• Allocation Type: Multi-year Fund
• Research Field: High performance computing
• Research Institution: National Institute of Advanced Industrial Science and Technology (2017); Tokyo Institute of Technology (2014-2016)
• Principal Investigator: Sato Hitoshi 国立研究開発法人産業技術総合研究所, 情報・人間工学領域, 主任研究員 (00550633)
• Project Period (FY): 2014-04-01 – 2018-03-31
• Keywords: 不揮発性メモリ / GPGPU / 高性能計算 / ビッグデータ

Outline of Final Research Achievements

Emerging NVM (Non-Volitle Memory) devices such as Flash, which have positive aspects of inexpensive cost, high-energy-efficiency, and huge capacity compared with conventional DRAM devices, as well, as negative aspects of low throughput and latency, are widely employed to existing supercomputers and clouds. However, efficient implementation techniques and its productivity to overcome deepening memory hierarchy are open problems, although these NVMs will greatly expand the possibility of processing extremely large-scale datasets that exceed the DRAM capacity of the nodes. In order to address the issues, we investigated the programming model for NVM toward extreme data-intensive computing. Based on our GPU-based MapReduce implementation, we enhanced out-of-core features of the implementation, including various Big Data Kernels such as Sort, PrefixSum, Unique, SetIntersection, and demonstrated efficient performance to datasets that exceed the DRAM capacity of the nodes.

Free Research Field

高性能計算