"Starlight" is a platform that supports the development and operation of supercomputing applications in the context of the convergence of AI， Big Data and HPC applications. It realizes the whole life cycle of application management including development, deployment and operation. Currently, over hundred supercomputing applications have been configured on "Starlight" platform, covering application multiple fields such as industrial simulation, biomedicine, computing materials, artificial intelligence and visualization.
“Starlight” established an Internet-based supercomputing service model, providing high performance computing as a service (HPCaaS). In order to manage users’ computing tasks and corresponding data effectively. “Starlight” integrates resource and data management systems, shields the underlying architectural differences and provided unified interfaces to the upper level. Based on resource and data management systems, the workflow management system realized consequently, users don’t need to care about the underlying resource scheduling.
The platform also implemented a converged scheduling system which makes multiple resource management systems (such as Kubernetes and SLURM) jointly managing a set of hardware resources. Through converged scheduling, different application loads can share the same resource pool, therefore data migration overhead is reduced and resource utilization is improved.
As the largest online HPC education and practice platform in China, EasyHPC, developed by a team from leading universities and companies, provides high-quality educational resources to both undergraduate and graduate students. The platform integrates a wide range of supercomputing educational resources, such as on-line courses, on-line programming environment, online training, case studies, and so on.
The features of this educational platform can be summarized as below.
1. Personalized Adaptive Learning of HPC Knowledge. We provide personalized learning paths for new users to get started easily and learn the basic parallel programming skills.
2. Automatic Deployment of Practice Environments. A large collection of mainstream practice environments covering HPC, AI, and Big data are supported, such as MPI, OpenMP, Pthreads, CUDA, Tensorflow, Caffe, Pytorch and so on.
3. Free High-quality HPC Courses. It provides free access to 40+ HPC courses from several leading universities. The course materials include slides, videos, homework, environments, codes and so on.
4. Feedback Debugging/Testing of Parallel Programs. We provide feedback debugging/testing of parallel programs to help users inspect the program performance in terms of CPU, memory, network and storage utilization. So users can identify bugs and bottlenecks of their code efficiently.
To accelerate the discovery of novel materials, we have developed a new R&D platform Matgen , short for “Material Gene & Generation”, integrating with high-throughput calculation automatic workflow and repository. Matgen adopts an easy-to-access web service. The modules of calculation, workflow and repository are highly coupled.
1. The calculation module contains platform support software and scientific computing software. The former includes repository construction software Matgen-toolkit, data analysis software Matgen-API, and online visualization software 3DStructGen. The latter currently supports VASP, CP2K, Quantum Espresso, RASPA, etc.
2. Workflow module adopts self-developed software, named Matflow, for task submission, monitoring, result analysis and visualization. As a result, the batch data can be obtained with little interaction.
3. The repository module currently contains molten salt, porous, polymer, biological and other common material data. The total data storage capacity exceeds 90 T, and the data volume is still increasing.
Aimed to empower the researchers in the fields of biology, medicine and computer science, Biomedical Computing Platform has integrated a variety of prediction functions and database query services.
1. As a web-based platform, it’s very easy and convenient to access.
2. The platform has integrated a lot of necessary models needed in the field of biomedical research and development like drug research and development, protein-related prediction and medical related prediction.
3. The platform can display the prediction results online, enabling users to view the results intuitively. According to the characteristics of the result file types and tasks, the platform automatically selects the best way to display, such as table, radar chart, histogram, etc.
4.The platform also provides access to some common databases, like tsRFun, a RNA-related database, and NeurodisM, a database of mutations and genes related to nervous system diseases and the relationship between diseases.
Technologies such as cloud computing, big data, AI and blockchain are constantly giving birth to new economical species, new models and new formats of business, and accelerating the development of the financial industry towards mobility, digitalization and intellectualization. The financial technology platform is committed to providing safe, reliable, low-cost, high-performance, highly available cloud computing resources as well as comprehensive solutions for scientific research institutes, financial institutions and regulatory agencies, etc.
The platform can provide the following functions.
1. Data fusion and rapid deployment. It can complete tasks including multi-source heterogeneous data management and access control for the sensitive data, etc.
2. Parallelization for complex models and algorithms. This platform offers optimization of super-computing resource scheduling for big data analysis and parallelization of core algorithms in the financial sector.
3. Risk monitoring and protection. It is capable for deeply exploring market sentiment and risks based on multi-source data and self-developed algorithms.
4. A one-stop platform for college-oriented financial talent cultivation. It has integrated several financial talent cultivation processes like teaching, testing, communication and exchanging.