Succes Stories - Customer Experience

Mitsui Consultants Co. LTD and Kyoto University

Floods are among the worst of natural disasters. From any metric – financial loss to the human toll, floods rank alongside earthquakes, tsunamis, and hurricanes. Being an island country, Japan experiences heavy rains, resulting in subsequent damage throughout the year. To combat the situation and facilitate quick evacuation, Mitsui Consultants Co. Ltd, Disaster Prevention Research Institute-Kyoto University and NEC Corporation have developed ' A National Version of the Real-time Inundation Prediction System' by using Rainfall-Runoff Inundation (RRI) model. The system allows immediate analysis and prediction of simple river channel flow into flood inundation 6 hours ahead in time by recording rainfall data. This system is built on a centralized ecosystem and embraces all rivers (major and minor) in Japan.

Bauman Moscow State Technical University

Dr. Alexander P. Kovtushenko
Dep. Computer Science and Control Systems

Background for evaluating SX-Aurora TSUBASA

Dr. Kovtsuschenko teaches parallel programming at the Moscow State Technical University. He visited the NEC SX-Aurora exhibition booth at the Russian Supercomputing Days 2019 and evaluated SX Aurora TSUBASA vectorization for a couple of months. Dr. Kovtsuschenko experienced the following benefits of using SX-Aurora TSUBASA.

Evaluation results

In order to evaluate the effectiveness of the system in interaction with the compiler, it turned out to be appropriate to use simple code fragments (matrix multiplication and Livermore cycles). It was important to evaluate the sensitivity of the system to the correct nesting of loops, changing the order of traversing the array, and reusing data on vector registers. The following scenarios have been compared

- Manual modification of the code during direct compilation
- Inefficient code with aggressive compiler optimization
- Library routine

“The test results were very impressive” stated Dr. Kovtsuschenko. With direct compilation of a less efficient and more efficient matrix multiplication code, the efficiency varied in the range from 7.2% to 45%. Aggressive optimization of inefficient code yields 55% efficiency and by using the NEC SX-Aurora TSUBASA library routine, the efficiency was 98%.

In addition to the above evaluation, enabling his students use SX-Aurora TSUBASA for their parallel programming learning is highly desirable. Dr. Kovtsuschenko is considering taking SX-Aurora TSUBASA into the curriculum and making it available to students in the future.