High Performance Computing
Date: Jun 09, 2021
Research priorities include basic numerical algorithms and parallel programming frameworks to provide common tools across domains, use of numerical algorithms to develop scientific codes for large-scale domain applications, and construction of web-based community portals to provide accessible services. Among all achievements, large-scale phase-field simulation software ScETD-PF was selected as a finalist of 2016 ACM Gordon Bell Prize. The first topological material data catalogue established with CAS Institute of Physics was published in Nature. The computational tool HotSpot3D was published in Nature Genetics and has been widely used in the screening and treatment of tumors.
1. Common Tools across Disciplinary Boundaries
SC_HCAL : A numerical algorithm library
SC_HCAL is a self-developed library of basic parallel algorithms for large-scale numerical computation, which integrates a variety of numerical parallel algorithms and solvers for common problems including eigenvalue problems, preconditions, sparse linear equations, fast algorithms and grid-particle problems. SC_HCAL has been applied in many fields such as computational physics/materials, computational electromagnetism, quantum chemistry and computational mechanics.
SC_Tangram: A parallel application framework
Developed based on the charm++ runtime system, the parallel application framework SC_Tangram provides a highly adaptive and high-performance runtime environment, demonstrating strength in scalability, latency masking, fine-grained parallelism, dynamic load balance and fault tolerance. SC_Tangram has been applied to astronomical fluid simulation, N-body simulation, Phase field simulation and engineering mechanics simulation.
2. Disciplinary Applications
1) Computational Materials Science
Phase Field (PF) is a widely used method to simulate micro-structure evolution and support the study of material properties. ScETD-PF is a large-scale Phase Field (PF) simulation software, which supported the 300 billion degree of freedom PF simulation with a sustained performance of 50 Peta flops (2016 ACM Gordon Bell Prize finalist). The simulation capability of ScETD-PF has been further increased to 1.8 trillion DOFs.
2) Computational Fluid Dynamics
CCFD is a large-scale parallel computational fluid dynamics software based on structured grids, solving the time-dependent conservation law form of the Reynolds-averaged Navier-Stokes equations. With strengths in robustness, efficiency, accuracy and flexibility, CCFD resolves practical problems encountered in the aeronautical and astronautical areas.
3) Earth System Model
Based on CPL7, the coupled integrated software platform SC_ESAP supports the parallel optimization of the main component of the Earth System Model of the CAS. The software platform allows users to optimize the performance and integrate the coupling assembly with different complexity and multiple modes, supporting the collaborative development, integration, efficient calculation and application of the earth system model.
4) Cosmological Simulation
SWFMM: Sunway heterogeneous processor based N-body software
Collaborated with CAS National Astronomical Observatories, the team develops cosmological N-body simulation using fast multipole method (FMM), particle-mesh method (PM) and related parallel algorithms on heterogeneous platforms.
5) Computational Chemistry
MOLY is a software framework developed in-house for high performance molecular dynamics simulations of biosystems and nanomaterials. By implementing flexible MD simulation models and integrating custom force fields, MOLY supports advanced modeling and simulating algorithms adapted to heterogeneous high performance computing platforms.
Heterogeneous ERI library + eMC package
The ERI library is a library for heterogeneous computation of electron repulsion integrals (ERI) that are used in ab initio calculations to support quantum chemistry research. eMC, a Multi-Configurational Self-Consistent Field/Multi-Configurational Linear Response (MCSCF/MCLR) code, is also being developed for multi-configurational systems.
6) Computational Genomics
Hotspot3D is a high-precision software for identifying high-frequency mutation hotspots in pan-cancer somatic cells.
MSIsensor2 is a novel algorithm based on machine learning, featuring a large upgrade in the microsatellite instability (MSI) detection for tumor only sequencing data, including Cell-Free Tumor DNA (ctDNA), Formalin-Fixed Paraffin-Embedded (FFPE) and other sample types. The original MSIsensor is multi-configurational systems designed for tumor/normal paired sequencing data. A typical WES data can be finished within 180 seconds.
3. Domain Portals and Web Services
Biomedical Applications Community based on China National Grid
The Biomedical Applications Community based on CNGrid provides web-based computing services for a considerable number of biomedical applications used in drug development, personalized medicine, bioinformatics and computational chemistry.
Materiae is a data and computing service co-hosted by CNIC and CAS Institute of Physics for topological properties of inorganic materials. It hosts 29000+ records based on high-throughput first-principle calculations and provides automated tools to calculate the topological properties for materials with recently developed methodology based on symmetry analysis.