Solution Architecture

HPC applications in the marine domain are wide, crossing physical oceanography, marine forecasting, marine climate, marine biology, and more. Among them, marine-forecast applications are important for addressing climate change and extreme marine weather. Marine forecasting, based on atmospheric and ocean physical and motion characteristics, under certain initial and boundary conditions, uses HPC for numerical computation, solving hydrodynamic and thermodynamic governing equations that describe fluid evolution, predicting marine-motion phenomena for a future period, with high requirements on HPC compute, memory-access, storage, and network-communication capability. Meanwhile, forecast-service departments focus more on business-system product quality and stability and cannot invest much effort in underlying hardware architecture, HPC-software deployment, and business-software implementation, placing high requirements on HPC vendors’ system-integration capability.

Sugon can provide a complete hardware-software integrated solution, minimizing user burden, reducing HPC usage difficulty, and raising usage benefit. Sugon’s marine-forecast HPC platform uses a cluster architecture and can be divided from bottom to top into four layers:

Infrastructure:Based on fourth-generation data-center standards, provide CloudBase computer-room infrastructure overall solutions, including pool-level, row-level, cabinet-level, and other series, supporting the most advanced cold-plate or immersion liquid cooling, providing related hardware and software with a stable, reliable, green, energy-saving runtime;

Hardware-resource layer:Including compute, storage, network, and other systems, providing a hardware environment with high floating-point performance and high-bandwidth, low-latency I/O and network data-exchange performance for marine-application model-system operation.

Management-software layer:Including system-management, system-monitoring, and resource-scheduling software for cluster management, cluster-status monitoring, cluster-resource management, and more, providing specific management and usage interfaces for convenient user use.

Value-added service layer:Including OS, compilers, parallel-program development environments, models and parallel-programming frameworks, math libraries, and more, used to provide a basic application runtime; provide EasyOP application-level HPC butler services covering remote cluster O&M, on-site inspection, and application-level technical support.