摘要
基准测试程序是评估处理器微体系结构设计的重要手段,然而当前的基准测试程序无法有效全面地评估面向高通量应用的处理器微体系结构的设计.基于此,针对高通量应用的特征,提出了用于评估面向高通量应用的处理器微体系结构设计的基准测试程序——HTC-MicroBench.首先,提出一种基于应用特征的高通量应用分类方法,并基于此分类方法对高通量应用中的Workload进行分类.其次,针对高通量应用的特征,提出了一种基于线程的作业处理节点并行化模型,基于此模型完成了HTCMicroBench的设计和实现.最后,从作业并发性、作业之间的耦合性和Cache使用效率等指标对HTCMicroBench进行实验评估;并基于HTC-MicroBench对TILE-Gx和Xeon两种处理器的并行加速能力做了评估,高并发、低耦合和由Workload特征所体现出的不同Cache命中率的评估结果说明了HTCMicroBench能够准确刻画高通量应用的特征,并对面向高通量应用的处理器微体系结构的设计进行有效的测评.
Benchmarks are important means to evaluate processor microarchitecture.The highthroughput application is a kind of application that focuses on throughput efficiency and contents a large number of loosely coupled small-scale jobs.The typical characteristics of high-throughput application are high throughput,hard real-time and high concurrency.The key target of processor microarchitecture design for high-throughput application is how to improve the throughput efficiency of operations.The design of high-throughput processor microarchitecture needs micro benchmark from high-throughput application as evaluation basis for designing high efficient processing architecture.While for now,existing benchmarks can not effectively and comprehensively evaluate the processor microarchitecture design for high-throughput application.In this paper,we propose a suit of new benchmarks—HTC-MicroBench—for the evaluation of designing the processor microarchitecture for high-throughput application. Firstly, we present a classification method for high-throughput applications based on the features of workloads.Secondly,according to the characteristics of highthroughput application,we present a parallelization model based on Pthread model to design and implement HTC-MicroBench.Furthermore,we evaluate HTC-MicroBench from many aspects,such as concurrency,data coupling and cache efficiency.Finally,we use HTC-MicroBench to evaluate the speedup of TILE-Gx and Xeon.The evaluation results show that HTC-MicroBench can effectively evaluate the processor microarchitecture design for high-throughput application.
作者
薛瑞
苗福涛
叶笑春
孙凝晖
徐文星
Xue Rui;Miao Futao;Ye Xiaochun;Sun Ninghui;Xu Wenxing(State Key Laboratory of Computer Architecture（Institute of Computing Technology,Chinese Academy of Sciences）,Beijing 100190;University of Chinese Academy of Sciences,Beijing 100049;Agricultural Bank of China,Beijing 100073,4.Beijing Institute of Petrochemical Technology,Beijing 102617)
出处
《计算机研究与发展》
EI
CSCD
北大核心
2018年第7期1569-1583,共15页
Journal of Computer Research and Development
基金
国家重点研发计划项目(2016YFB0200501)
国家自然科学基金项目(61332009)
国家自然科学基金委员会"创新研究群体科学基金"(61521092)
数学工程与先进计算国家重点实验室开放基金(2016A04)~~
