Optimization Of Supercomputing Systems Based On Parameter Configuration

High performance computer hardware performance enhancement, needs to have the corresponding software support, to show the performance of the hardware system in actual use. The application of the supercomputer system will be run by users many times on a specific hardware platform, and the optimal parameter configuration of the application on the specific hardware platform can be recommended to the supercom-puter users to improve the application performance. Increasing with the complexity of super computer, high-performance computing applications to get the best performance is a very difficult task, it need to super computer system hardware and software of all levels to maximize the tuning. Compiler optimization options, program runtime param-eters, and hyper threading can have significant effects on program performance. Among them, compiler tuning options, program runtime parameters are program-related param-eter configurations, to optimize it does not need to restart the machine. Hyper-Threading is the hardware system parameter configuration, and Hyper-Threading is the Intel pro-cessor hardware switch, to start the Hyper-Threading need to restart the machine. Ac-cording to the characteristics of the supercomputer system, this paper proposes the opti-mization of the supercomputer system from the perspective of parameter configuration.According to the characteristics of the problem, we optimize the supercomputer system from two levels: program-related parameter configuration optimization and hardware system parameter configuration optimization. The details of this article are as follows:1. Program-related parameter configuration optimization problem can be sum-marized as multi-factor experimental problem. Because the problem is based on the real application scene abstraction, it has a certain practical value. By analyzing the specific characteristics of the problem, an experimental design method is proposed to adapt the specific characteristics of the compiler optimization parameters and the pro-gram runtime parameters, the orthogonal experiment is used to optimize the compiler optimization parameters, and the program runtime parameters are optimized by using the uniform experiment and the orthogonal experiment, and the optimal parameters of the compiler optimization parameters and the program runtime parameters are com-bined to optimize the application, the experimental results of supercomputing platform show the effectiveness of our proposed methods.2. The performance of parallel programs is balances between the effective use of idle cycle and CPU, cache, bandwidth sharing resource conflicts, performance en-hancements of Hyper-Threading depends on the parallel program features. By collect-ing parallel program running characteristics, the relationship between parallel program performance improvement and parallel program running characteristics is studied by regression analysis, the experimental results of supercomputing platform show the ef-fectiveness of our proposed methods.Program-related parameter configuration optimization and hardware system pa-rameter configuration optimization are to tune software and hardware of the supercom-puter system respectively, they are complementary to each other. Program-related pa-rameter configuration optimization can improve the application performance, hardware system parameter configuration optimization can improve throughput rate of the super-computer system, simultaneously tuning of these two levels can maximize the optimiza-tion of supercomputing system.