Obayashi Corporation Develops Scalable HPC/HTC for Environmental Analysis (Simulations) in Urban Development, Making HPC Wind Environment Analysis up to 640 Times Faster

Obayashi Corporation’s group vision is “MAKE BEYOND—Transcending the Art and Science of Making Things”. Its goal is to expand beyond its current business boundaries by integrating advanced technology into projects. In urban development, structures don’t just define physical space—they also influence the surrounding environment. For example, buildings can create wind patterns that impact pedestrian spaces, alter views and visibility from nearby structures, and affect how sunlight heats ground surfaces. As part of its urban development projects and high-rise building construction, the company requires these environmental factors to be thoroughly assessed.

To perform these assessments, Obayashi Corporation’s design department relies on 3D models to conduct environmental analysis, which demands significant computational resources. However, completing these analyses quickly has been challenging in its existing PC-based environment. “Our employees solely relied on PCs for analysis,” says Hirotsugu Ueda, Assistant Manager of Advanced Design, Design Solutions Department in Design Division of Obayashi Corporation. “They can secure the resource of few PCs at most. While our R&D team had a small HPC environment, they needed to request assistance from dedicated staff, making frequent use difficult.” Furthermore, the team faced several issues with their analysis workload. Urban environment analysis begins with downloading the PLATEAU 3D city model, an open urban model data overseen by the Ministry of Land, Infrastructure, Transport, and Tourism. They then combine data from PLATEAU with computer-aided design (CAD) data of the planned building to create a visualized model for analysis. The next step involves defining the requirements for analysis based on metrological data and generating a mesh for simulation. Following that, the team conducts analysis, evaluates results, visualizes the data, and compiles a report. Each of these steps was manual, leading to long coordination times between departments.

“The first challenge is obtaining data from PLATEAU,” says Ueda. “We need to download 3D urban models for planned sites using publicly available data from the website. After that, we must convert the data format—CityGML—into one compatible with construction simulation, such as the OBJ format. With around 210 cities in Japan, converting all the data, even when automated, takes over 200 hours. When the 3D model data is combined with analysis data, it amounts to gigabytes or even terabytes. These massive datasets are challenging to process and analyze on PCs, which is why we started considering cloud-based solutions.”

Ueda began developing open-source analysis tools and testing various cloud services in 2020, prioritizing computational efficiency and ease of use. The company ultimately chose Amazon Web Services (AWS) as the most comprehensive and advanced solution. “AWS stood out for the variety and stability of its computational resources,” says Ueda. “We could use up to 30 to 50 nodes per job, which was remarkable back in 2020.” The business implemented Amazon Elastic Cloud Compute (Amazon EC2) Spot Instances to lower costs and used AWS Lambda to instantly scale up to thousands of computational resources, boosting efficiency. The availability of extensive technical documentation also meant Obayashi Corporation could save on training costs.

In 2021, the team developed an analysis tool, initially using it for wind analysis. To process the data, they used AWS Lambda and AWS Batch to convert information from PLATEAU into the OBJ format and stored the results in Amazon Simple Storage Service (Amazon S3), making the converted data easily searchable. This approach significantly reduced the time needed to access 3D city models across Japan.

Each step of the analysis was then modularized using containers to streamline the process. The team leveraged fully managed AWS Batch for HPC analysis, which enabled large-scale parallel processing, while AWS Lambda handled high-throughput computing (HTC) analysis for high-speed serverless distributed processing. Standardizing each analysis component using containers and automating the entire workflow through AWS Step Functions allowed them to seamlessly switch between HPC and HTC processes as required.

“When we considered fully automating the process,” Ueda explains, “we explored AWS Batch and AWS Lambda. Modularizing the process with containers allowed us to flexibly allocate computational resources as needed. While building the workflow for the analysis module, we discovered AWS Step Functions. Six months later, AWS released the GUI-based AWS Step Functions Workflow Studio, which made it easier to build workflows exactly as we wanted. From there, we expanded our simulation products from wind analysis to include insolation analysis and viewshed assessment.”