11:00 〜 13:00
[AAS05-P06] Verification of the use of Google Cloud Platform for calculating a real-time forecast model
キーワード：クラウドコンピューティング、海氷予測、波浪予測、船舶ナビゲーション
A high-performance computing environment with stability and high-speed operation is used when the forecast model runs in real-time. The calculation time of the real-time forecast model constrains by the schedule because the calculation must be completed earlier than the next forcing data update. Therefore, it is crucial to confirm the start and end times of the model calculation and finish the calculation within that time. This time is called the time quota for the calculation. The schedule defines when the update time of the input data and the calculation result are output to clarify this quota, and the time available for the model is described. This time is used not only for the forecast model's calculation time but also for pre- and post-processing of the model, the extension of the unexpected calculation time, and the free time for the subsequent processing. The demand for computer resources to meet the quota is determined by creating a test case based on the actual model and running a test run while changing the CPU architecture and number of cores. Thus, the real-time forecast model's schedule and the demand for the computing environment are determined, and we can estimate the cost of the production operation of the forecast model.
We delivered sea ice and wave forecast data to R/V Mirai for the Arctic Sea observation cruise MR20-05C (from 19 Sep. to 2 Nov. in 2020) and MR21-05C (31 Aug. to 21 Oct. in 2021). We forecasted every day and used VENUS as a data transfer platform. Sea ice forecast data was output from the sea ice model IcePOM. Wave forecast data was output from the wave model WAVEWATCH-III (WW3). We used ECMWF 10-day weather forecast data as input data for both models.
We used the Google Cloud Platform (GCP) for the calculation environment of this forecast model. We have selected GCP because of the flexibility of computing resources and the ease of constructing a calculation environment. Next, we built the computer environment that demanded the calculation of the forecast model on the GCP. We connected this environment to VENUS, and the data was delivered to R/V Mirai. In this presentation, we would like to introduce the technical results obtained in these construction processes. In particular, we would like to report on the results of comparing on-premises with GCP. This study was a case of using GCP to run the real-time forecast model in the scientific field, and it introduces the design, construction, and operation. For researchers considering cloud services, this case can expect to be beneficial.
We delivered sea ice and wave forecast data to R/V Mirai for the Arctic Sea observation cruise MR20-05C (from 19 Sep. to 2 Nov. in 2020) and MR21-05C (31 Aug. to 21 Oct. in 2021). We forecasted every day and used VENUS as a data transfer platform. Sea ice forecast data was output from the sea ice model IcePOM. Wave forecast data was output from the wave model WAVEWATCH-III (WW3). We used ECMWF 10-day weather forecast data as input data for both models.
We used the Google Cloud Platform (GCP) for the calculation environment of this forecast model. We have selected GCP because of the flexibility of computing resources and the ease of constructing a calculation environment. Next, we built the computer environment that demanded the calculation of the forecast model on the GCP. We connected this environment to VENUS, and the data was delivered to R/V Mirai. In this presentation, we would like to introduce the technical results obtained in these construction processes. In particular, we would like to report on the results of comparing on-premises with GCP. This study was a case of using GCP to run the real-time forecast model in the scientific field, and it introduces the design, construction, and operation. For researchers considering cloud services, this case can expect to be beneficial.