【摘要】 针对大数据时代CPU或多核服务器中普遍存在的数据处理能力严重不足的问题,文中设计了一种基于GPU通用处理器作为协处理计算单元的平台方案;通过对数据处理的业务流程进行了逐层分解,简化了算法在GPU实现上的复杂度;设计了一套基于双线程池的插件化作业处理平台来提高CPU与GPU的利用率,实现了对数据处理的插件化管理。理论分析和性能测试均验证,GPU完全能够胜任复杂且专项的数据处理业务的实时性工作需求,处理能力表现甚佳。更多还原

【Abstract】 In view of insufficient data processing capability of CPU or multicore server in big data ages, this paper designs a platform solution based on general processing unit(GPU) used as co-processing calculation unit. This solution simplifies the complexity of algorithm by ecomposing the service flow of data processing layer by layer. A plug-in operation processing platform based on double-thread pool is designed to increase utilization ratio of CPU and GPU and achieve plug-in management of data processing. The results of theoretical analysis and performance test show that GPU has excellent processing capability and can meet the real-time requirement of complex and special data processing service.更多还原