instrumenting the mg applicaiton of nas parallel benchmark

Post on 05-Jul-2015

341 Views

Category:

Technology

1 Downloads

Preview:

Click to see full reader

DESCRIPTION

Course: Measurement Tools and Techniques (10-15min): Instrumenting the MG applicaiton of NAS Parallel Benchmark Aim of this presentation: Show the steps followed for instrumenting the application.

TRANSCRIPT

Instrumenting the MG application

of NAS Parallel Benchmark

Maria Stylianoumarsty5@gmail.com

20-APR-2012

34324 - Measurement Tools and Techniques

2

Outline

● Basic Information

● Instrumentation● By observation● Using Performance Counters● Using Histograms

● Conclusions

3

Outline

● Basic Information

● Instrumentation● By observation● Using Performance Counters● Using Histograms

● Conclusions

1 - Basic Info, 2 - Instrumentation (a) by Observation, (b) with Performance Counters, (c) with Histograms, 3 - Conclusions

4

Basic Information

● Personal Laptop● Ubuntu 11.10, 64-bit● Intel Quad Core i5● 4GB RAM

● Boada Server● Intel(R) Xeon(R) CPU E5645 @ 2.40GHz● 12 Cores with HT support● 24 GΒ RAM

Execution Environments

1 - Basic Info, 2 - Instrumentation (a) by Observation, (b) with Performance Counters, (c) with Histograms, 3 - Conclusions

5

Basic Information

● Evaluate the performance of parallel supercomputers

● Several Applications MG – MPI Version● IS, EP, CG, MG Multi-Grid on a sequence ● FT, BT, SP, LU of meshes

● Extrae → Produce traces● Paraver → Analyse traces

NAS Parallel Benchmark

1 - Basic Info, 2 - Instrumentation (a) by Observation, (b) with Performance Counters, (c) with Histograms, 3 - Conclusions

6

Outline

● Basic Information

● Instrumentation● By observation● Using Performance Counters● Using Histograms

● Conclusions

1 - Basic Info, 2 - Instrumentation (a) by Observation, (b) with Performance Counters, (c) with Histograms, 3 - Conclusions

7

Instrumentation by Observation

1 - Basic Info, 2 - Instrumentation (a) by Observation, (b) with Performance Counters, (c) with Histograms, 3 - Conclusions

8

Instrumentation by Observation

1 - Basic Info, 2 - Instrumentation (a) by Observation, (b) with Performance Counters, (c) with Histograms, 3 - Conclusions

9

Instrumentation by Observation

1 - Basic Info, 2 - Instrumentation (a) by Observation, (b) with Performance Counters, (c) with Histograms, 3 - Conclusions

10

Instrumentation by Observation

Initialization

1 - Basic Info, 2 - Instrumentation (a) by Observation, (b) with Performance Counters, (c) with Histograms, 3 - Conclusions

11

Instrumentation by Observation

Execution

1 - Basic Info, 2 - Instrumentation (a) by Observation, (b) with Performance Counters, (c) with Histograms, 3 - Conclusions

12

Instrumentation by Observation

Finalization

1 - Basic Info, 2 - Instrumentation (a) by Observation, (b) with Performance Counters, (c) with Histograms, 3 - Conclusions

13

Outline

● Basic Information

● Instrumentation● By observation

● Using Performance Counters● Using Histograms

● Conclusions

1 - Basic Info, 2 - Instrumentation (a) by Observation, (b) with Performance Counters, (c) with Histograms, 3 - Conclusions

14

Instrumentation using Performance Counters

Instructions

1 - Basic Info, 2 - Instrumentation (a) by Observation, (b) with Performance Counters, (c) with Histograms, 3 - Conclusions

15

Instrumentation using Performance Counters

Cycles

1 - Basic Info, 2 - Instrumentation (a) by Observation, (b) with Performance Counters, (c) with Histograms, 3 - Conclusions

16

Instrumentation using Performance Counters

IPC: Instructions Per Cycle

1 - Basic Info, 2 - Instrumentation (a) by Observation, (b) with Performance Counters, (c) with Histograms, 3 - Conclusions

17

Instrumentation using Performance Counters

L1 Cache Misses

1 - Basic Info, 2 - Instrumentation (a) by Observation, (b) with Performance Counters, (c) with Histograms, 3 - Conclusions

18

Outline

● Basic Information

● Instrumentation● By observation● Using Performance Counters

● Using Histograms

● Conclusions

1 - Basic Info, 2 - Instrumentation (a) by Observation, (b) with Performance Counters, (c) with Histograms, 3 - Conclusions

19

Instrumentation using Histograms

1 - Basic Info, 2 - Instrumentation (a) by Observation, (b) with Performance Counters, (c) with Histograms, 3 - Conclusions

Time Histogram

20

Instrumentation using Histograms

1 - Basic Info, 2 - Instrumentation (a) by Observation, (b) with Performance Counters, (c) with Histograms, 3 - Conclusions

Percentage Histogram

21

Instrumentation using Histograms

1 - Basic Info, 2 - Instrumentation (a) by Observation, (b) with Performance Counters, (c) with Histograms, 3 - Conclusions

Percentage Histogram

22

Instrumentation using Histograms

1 - Basic Info, 2 - Instrumentation (a) by Observation, (b) with Performance Counters, (c) with Histograms, 3 - Conclusions

Percentage Histogram

23

Conclusions

● Scalability● In laptop: No way!● In Boada: Yes!

● #Processors Increase

→ L1 Cache Misses Increase

● Useful information very fast → Histograms!

1 - Basic Info, 2 - Instrumentation (a) by Observation, (b) with Performance Counters, (c) with Histograms, 3 - Conclusions

24

Instrumenting the MG application

of NAS Parallel Benchmark

Maria Stylianoumarsty5@gmail.com

20-APR-2012

34324 - Measurement Tools and Techniques

top related