We are customer-centric and provide customized or one-stop full-stack solutions to empower all industries.
Most computational materials and computational chemistry programs are computationally intensive and memory-intensive applications, which have high requirements for the floating-point computing power and memory performance of the processor. For typical physical and chemical applications, it is recommended to use dual-core servers for computing nodes. Depending on the CPU model, each computing node provides 16 to 40 processor cores, and each core can be configured with 2 to 4 GB of memory on average. For parallel programs such as ADF, VASP, and Gaussian that are mainly based on OpenMP shared memory parallelism, it is recommended to use SMP fat nodes with more computing cores and larger memory to achieve the best performance acceleration ratio. Most physical and chemical software are MPI parallel programs, with good parallel acceleration, which can usually be expanded to more than 32 to 128 processes. The program usually runs in parallel across nodes; from the analysis of parallel algorithms, the inter-process communication of computational materials software is relatively frequent, and it is a communication-intensive application. Therefore, the latency and bandwidth of the inter-node network play a decisive role in the parallel acceleration ratio of the program. It is difficult to give full play to the parallel performance of the program using traditional Gigabit Ethernet. When running across nodes, it will cause serious performance degradation and become the bottleneck of the entire computing process. Therefore, it is recommended that users use high-bandwidth, low-latency InfiniBand networks.
Based on the analysis, most computational materials and computational chemistry programs are computationally intensive and memory-intensive applications, which have high requirements on the floating-point computing capability and memory performance of the processor. A high-density server platform is recommended.
The above figure is a typical configuration of a computational physics and chemistry cluster system. The system has excellent performance and good scalability. The computing nodes use the PowerLeader high-density server platform, 2U dual-channel multi-node server PR2740TP, to meet the needs of general computing software and computing tasks; the computing system can also be configured with several four-way or eight-way SMP fat nodes as needed, specially equipped with ADF and VASP for OpenMP parallel programs such as Gaussian or other large memory applications. The cluster system is configured with two PowerLeader PR2510P2 servers as management login nodes. The two login management nodes are configured with dual-machine hot standby to achieve double insurance for system services such as job scheduling and user management, thereby improving the high availability of the entire cluster; the computing network uses a high-bandwidth, low-latency FDR InfiniBand network to provide a non-blocking high-speed communication channel for parallel programs; in addition, one or two Gigabit networks are configured for cluster management and monitoring, and the Gigabit network can also be used as a backup computing network to improve the high availability of the system; the storage system uses the PowerLeader GS-5312v3 disk array, which provides massive data storage capabilities while its innovative RAID level provides higher data protection functions. Powerful hardware design, flexible expansion operation and friendly management interface provide simple and convenient solutions for customers’ different applications.
PowerLeader's efficient server platform solution provides key virtualization functions through Windows Server Hyper-V, allowing you to integrate roles scattered across multiple servers into different virtual machines running on a single physical server. This solution can help you reduce server sprawl and maximize the use of current physical hardware. Each server can run in its own independent environment, which can achieve higher security and easier management.
The equipment used in the cluster focuses on green energy conservation, selects low-power components, and adopts professional heat dissipation design to avoid unnecessary energy loss; PowerLeader PDCM intelligent power consumption control system can adjust the power consumption of the entire system in real time according to the cluster load, achieving 20% energy saving without affecting the performance of the cluster, truly realizing green and high-efficiency computing.
Supports centralized user management, device management, storage management, and image calling, changing the drawbacks of decentralized management of previous analog systems and greatly improving management efficiency and scheduling capabilities
