Function Architecture

Figure 1 shows the layered architecture of the HPC function framework.

• DonauKit provides the scheduler and platform management functions, including Donau Portal and Donau Scheduler.
  • Donau Portal provides a simple and easy-to-use portal on which you can implement core functions such as job submission, job management, application integration, remote visualization, report representation, and billing.
  • Donau Scheduler is a proprietary scheduler that provides job scheduling with high resource utilization and throughput for large clusters.
• HPCKit, a toolkit designed for developing HPC applications on the Kunpeng architecture, including domain-specific acceleration libraries, Kunpeng Math Library (KML), Hyper Message Passing Interface (Hyper MPI), Hyper IO, Kunpeng Unified Parallel Library (KUPL), and BiSheng Compiler
  • Domain-specific acceleration libraries are libraries deeply optimized on the Kunpeng platform for specific HPC scenarios, such as KuLitho (computational lithography), KuQCD (quantum chromodynamics), and KuRTM (oil and gas seismic exploration). By leveraging the Kunpeng processor architecture, the solution significantly enhances the running efficiency of applications in the corresponding domains through instruction-level optimization, algorithmic reconstruction, and parallel acceleration.
  • The KML is a high-performance mathematical acceleration library deeply optimized on Kunpeng processors. It supports common mathematical functions such as fast Fourier transform (FFT), matrix computation, vectorization, and trigonometric and logarithmic functions.
  • The Hyper MPI library optimizes collective communication algorithms, including Allreduce, Bcast, and Barrier.
  • The Hyper IO library is a type of software that connects system software and application software. It defines storage description formats and provides file operations for scientific data. It is widely used in HPC applications and various post-processing, visualization, and analysis frameworks.
  • KUPL provides foundational parallel acceleration functions optimized for the Kunpeng platform, with all APIs implemented in C/C++ and assembly.
  • BiSheng Compiler, built on the open-source LLVM, incorporates advanced optimizations for mid- and back-end technologies, making it a high-performance compiler tailored for the Kunpeng platform.
  • GCC for openEuler is a compiler toolchain (including the compiler, assembler, and linker) developed based on the open-source GCC. It is released in the openEuler community.
• The solution supports TaiShan 200 2280 servers.
• The solution can be used with the OceanStor Pacific scale-out storage. OceanStor Pacific supports NFS standard protocol scenarios and DPC scenarios.
• The solution supports high-performance RoCE networks.