Improving Automotive Simulation Efficiency with Alibaba Cloud E-HPC
11.11 The Biggest Deals of the Year. 40% OFF on selected cloud servers with a free 100 GB data transfer! Click here to learn more.
By Alibaba Cloud ECS team
According to popular belief in the automotive industry, a car must experience at least 11 crash tests before being produced for the public. Today, in the modern automotive industry, high-performance computing (HPC) makes this practice obsolete through automotive simulation. HPC can be implemented in almost all stages in the design process by integrating advanced simulation software. This relies on stronger parallel computing clusters and flexible data streams, as well as industrial simulation software that implements complex algorithms.
The application of HPC is not limited only to crash tests. At the main forum of the Computing Conference 2018 in Hangzhou, Alibaba Cloud engineers demonstrated a comparative automotive wind tunnel model test and a digital simulated wind tunnel during the Alibaba Cloud Elastic High Performance Computing (E-HPC) release.
In the demonstration, an engineer used a wind tunnel model and a model car to show dynamics of wind flow under different conditions. He explained the general process of a traditional automotive wind tunnel test, and described how a digital simulated wind tunnel could greatly improve test efficiency.
The two core features of this demonstration are Super Computing Cluster (SCC) and Elastic HPC (E-HPC). This demonstration combined Alibaba Cloud SCC with E-HPC operational procedures such as cluster creation and CloudMetrics-based performance analysis to show visualized operational results computed by the ANSYS computational fluid dynamics (CFD) software based on SCC and E-HPC.
Alibaba Cloud Super Computing Cluster (SCC)
The HPC cluster computing node, which powered the ANSYS Fluent CFD simulation of an F1 racing car model during the live show, is composed of SCCH5 instances.
Backed by Alibaba Cloud’s next-generation Elastic Compute Service (ECS) Bare Metal Instance (X-Dragon), SCC not only has the elastic resource advantages of cloud computing, but also has the performance of physical machines. Moreover, SCC supports high-speed remote direct memory access (RDMA) interconnection, which greatly improves network performance and the large-scale cluster acceleration ratio. These are unique features of SCC.
The following formula can be used to describe SCC capabilities: SCC = ECS Bare Metal Instance + RDMA network. According to He Wanqing, a senior technical expert and the director of the Alibaba Cloud HPC division, “SCC has a heart of a virtual machine, muscles of a physical machine, and a nervous system of a high-speed network.”
Compared with traditional physical machines, ECS Bare Metal Instance has no difference in the computing performance. It also features secure physical isolation.
RDMA is a direct memory access technology that transfers data directly from the memory of one computer to that of another without the intervention of both parties’ operating systems. This enables high-throughput and low-latency network communication, especially in large-scale parallel computer clusters. SCC uses the RDMA over Converged Ethernet (RoCE) network scheme.
At the same time, SCC is fully interconnected with mature ECS management and control to ensure consistent user experience. It is also fully integrated with Alibaba Cloud Virtual Private Cloud (VPC), Network Attached Storage (NAS), Cloud Paralleled File System (CPFS), and other products, thereby maximizing the ecological advantages on the cloud and bringing the ultimate computing experience.
Alibaba Cloud Elastic High Performance Computing (E-HPC)
If SCC offers the HPC infrastructure, E-HPC provides users with one-stop and full-process public cloud HPC services.
E-HPC has the following advantages:
- Provides parallel scheduling by integrating HPC schedulers such as PBS Pro and Slurm.
- Enables load-based auto scaling of clusters based on Alibaba Cloud’s control capabilities.
- Supports parallel communication based on the VPC and RoCE network architectures.
- Provides container deployment through the SCC Shifter container.
- Uses CloudMetrics, which has been developed for on-cloud parallel jobs, to provide performance monitoring and analysis as well as a “performance big data” accumulation optimization engine.
In the demonstration, E-HPC supported parallel scheduling, auto scaling, and parallel communication, making large-scale simulation possible.
Compared to traditional HPC centers and self-built IDC clusters, E-HPC is also superior in the following aspects:
- E-HPC can be purchased on demand without scheduling and is ready to use upon purchase. It supports many types of instances so that the most suitable computing resource ratio can be selected based on application needs. E-HPC fully reuses Alibaba Cloud products, greatly improving data security and availability.
- E-HPC saves a great deal of capital expenditure (CAPEX), including hardware system costs, software license costs (such as the Alibaba Cloud elastic license supported by ANSYS), server room construction costs, power and cooling costs, and daily maintenance costs. E-HPC frees users from the worry of hardware upgrades, and uses the public cloud to ensure high scalability.
Currently, Alibaba Cloud HPC is being widely used in manufacturing, scientific computing, and other fields:
- SAIC Motor uses SCC clusters for hybrid simulation of its passenger vehicles, improving the overall efficiency by 25%.
- Pera China provides customers with HPC solutions based on SCC clusters, reducing the overall costs by more than 20%.
- The State Key Laboratory of the Dalian Institute of Chemical Physics uses E-HPC clusters for molecular dynamics simulation computing, increasing the efficiency by 200% compared with the previous generation of GPU clusters.
To learn more about the Alibaba Cloud Elastic High Performance Computing (E-HPC), visit www.alibabacloud.com/product/ehpc