Dynamic Instrumentation of Large-Scale MPI and OpenMP .ppt

上传人:diecharacter305 文档编号:374389 上传时间:2018-10-06 格式:PPT 页数:11 大小:127.50KB
下载 相关 举报
Dynamic Instrumentation of Large-Scale MPI and OpenMP .ppt_第1页
第1页 / 共11页
Dynamic Instrumentation of Large-Scale MPI and OpenMP .ppt_第2页
第2页 / 共11页
Dynamic Instrumentation of Large-Scale MPI and OpenMP .ppt_第3页
第3页 / 共11页
Dynamic Instrumentation of Large-Scale MPI and OpenMP .ppt_第4页
第4页 / 共11页
Dynamic Instrumentation of Large-Scale MPI and OpenMP .ppt_第5页
第5页 / 共11页
亲,该文档总共11页,到这儿已超出免费预览范围,如果喜欢就下载吧!
资源描述

1、04/16/2003,NC State University,1,Dynamic Instrumentation of Large-Scale MPI and OpenMP Applications,Christian Thiffault, Michael Voss, Steven T. Healey, Seon Wook Kim,Presented By: Jaydeep Marathe,04/16/2003,NC State University,2,The Basic Idea ,Instrumentation needed to collect collect performance

2、data.Static (compile or link time) instrumentationDynamic (run-time) instrumentation,Feasibility study for dynamic instrumentation and dynamically controlled static instrumentation.,Paper asserts: “Dynamic methods better than static instrumentation withrespect to execution overhead & trace data size

3、”,But thats still debatable, as we shall see .,04/16/2003,NC State University,3,Dynamic Instrumentation (with DPCL),JMP _Test,Executing Program,ProbePoint,04/16/2003,NC State University,4,Dynamic Control of Static Instrumentation,Conf_sync(),.,Conf_sync(),Conf_sync(),.,.,.,.,.,.,.,.,Compiler-Instrum

4、ented Program,Conf_sync() return; ,Read Active List,Point Active ?,Instrumentation Library,Instrument compile-time, decide activation at run-time.,04/16/2003,NC State University,5,Instrumentation Tool : Dynprof,For mixed (?) OpenMP + MPI programs,Application Source,KAI Guide Compiler,Executable,Guid

5、eTrace OpenMP Library,MPI Library,Dynprof,VampirTrace Library,Trace Data,Applicable only to the KAI OpenMP compilerInstrumentation : VampirTrace call to log trace,04/16/2003,NC State University,6,Experiments ,Compare execution times for static vs. dynamic instrumentation.4 ASCI benchmarks - Smg98, S

6、ppm, Sweep3d (MPI), Umt98 (OpenMP)IBM Power3 144-node system, each node has 8 Power3 processors. Measure execution times with 5 scenarios:,FULL: Full Static Instrumentation +Trace loggingFULL-OFF: Full Static Instrumentation, no trace loggingSUBSET: Full Static Instrumentation, only selected functio

7、ns active.None: No instrumentation (Baseline)Dynamic: Dynamically inserted instrumentation.,04/16/2003,NC State University,7,Results: Smg98,04/16/2003,NC State University,8,Results: Smg98 (contd),Static approach (SUBSET), more expensive than dynamic insertion (DYNAMIC).,WHY ? - No explanation in pap

8、er.,04/16/2003,NC State University,9,Results: Smg98 Another interpretation,Observe values for FULL, FULL-OFF,SUBSET. Very Close !,Compiler instrumentation inefficiency skews results ! dynamic approach looks better.,04/16/2003,NC State University,10,Time to create and instrument,Significant cost, but

9、 only occurs at startup.May significantly perturb application, when instrumented during run-time.,04/16/2003,NC State University,11,Summary and Conclusions,Tool to instrument MPI and OpenMP programs (Dynprof)Trace logging/visualization through existing VampirTrace library.Compare costs of static,dynamic and dynamically controlled staticinstrumentation.Asserts that dynamically inserted instrumentation is less expensive, as compared to static instrumentation.,

展开阅读全文
相关资源
猜你喜欢
相关搜索

当前位置:首页 > 教学课件 > 大学教育

copyright@ 2008-2019 麦多课文库(www.mydoc123.com)网站版权所有
备案/许可证编号:苏ICP备17064731号-1