17
LOGO Parallel computing technique for EM modeling makai 天天天天天天天天天天天天 School of Electronic Information Engineering

LOGO Parallel computing technique for EM modeling makai 天津大学电子信息工程学院 School of Electronic Information Engineering

Embed Size (px)

DESCRIPTION

Company name Background School of Electronic Information Engineering  Arificial neural network (ANN) techniques have been recognized as a powerful tool in electromagnetic(EM)-based modeling and design optimization of microwave passive components. ANN can learn EM responses versus geometrical variables through an automated training process, and the trained ANN can be used as accurate and fast models in the design optimization.

Citation preview

Page 1: LOGO Parallel computing technique for EM modeling makai 天津大学电子信息工程学院 School of Electronic Information Engineering

LOGO

Parallel computing technique for EM modeling

makai

天津大学电子信息工程学院School of Electronic Information Engineering

Page 2: LOGO Parallel computing technique for EM modeling makai 天津大学电子信息工程学院 School of Electronic Information Engineering

Company name

Contents

School of Electronic Information Engineering

1 、 background

2 、 Basic theory of parallel computing

3 、 An example

4 、 plan

Page 3: LOGO Parallel computing technique for EM modeling makai 天津大学电子信息工程学院 School of Electronic Information Engineering

Company name

Background

School of Electronic Information Engineering

Arificial neural network (ANN) techniques have been recognized as a powerful tool in electromagnetic(EM)-based modeling and design optimization of microwave passive components . ANN can learn EM responses versus geometrical variables through an automated training process, and the trained ANN can be used as accurate and fast models in the design optimization.

Page 4: LOGO Parallel computing technique for EM modeling makai 天津大学电子信息工程学院 School of Electronic Information Engineering

Company name

Background

School of Electronic Information Engineering

“Efficient design optimization of microwave circuits using parallel computational methods,” European Microwave Conference, Amsterdam, the Netherlands, Nov. 2012.

“Parallel automatic model generation technique for microwave modeling,” in IEEE MTT-S Int. Microw. Symp. Dig., Honolulu, Hawaii, June 2007.

Page 5: LOGO Parallel computing technique for EM modeling makai 天津大学电子信息工程学院 School of Electronic Information Engineering

Company nameSchool of Electronic Information Engineering

Basic theory of parallel computing

Page 6: LOGO Parallel computing technique for EM modeling makai 天津大学电子信息工程学院 School of Electronic Information Engineering

Company name

OpenMP Programming Model

School of Electronic Information Engineering

Shared Memory Model

Uniform Memory Access Non-Uniform Memory Access

OpenMP is designed for multi processor/core, shared memory machines

Page 7: LOGO Parallel computing technique for EM modeling makai 天津大学电子信息工程学院 School of Electronic Information Engineering

Company name

MPI(message passing interface) Programming Model

School of Electronic Information Engineering

distributed memory architecture hybrid distributed shared memory architecture

Distributed Memory Model

MPI was designed for distributed memory architectures. Basic MPI Concepts:A set of processes executing in parallel. These

processes have separate address space.

Page 8: LOGO Parallel computing technique for EM modeling makai 天津大学电子信息工程学院 School of Electronic Information Engineering

Company name

Parallel computing environment

School of Electronic Information Engineering

Figure 1 A Windows HPC Server cluster of workstations

Page 9: LOGO Parallel computing technique for EM modeling makai 天津大学电子信息工程学院 School of Electronic Information Engineering

Company name

Performance Measurement for Parallel Computing

School of Electronic Information Engineering

p

s

tt

ystemtimeExecutionsystemprocessoronetimeExecutionnS )sparallel()()(

Speedup factor

processors of numberssormultiproce a using timeExecutionprocessor one using time ExecutionE

Efficiency

%n

S(n)E 100 when E is given as a percentage

Page 10: LOGO Parallel computing technique for EM modeling makai 天津大学电子信息工程学院 School of Electronic Information Engineering

Company name

Parallel Algorithm Design process

School of Electronic Information Engineering

Problem

Mapping

Communication

Partitioning

Agglomeration

Figure 2 PCAM process

Page 11: LOGO Parallel computing technique for EM modeling makai 天津大学电子信息工程学院 School of Electronic Information Engineering

Company nameSchool of Electronic Information Engineering

An example of parallel computing

Page 12: LOGO Parallel computing technique for EM modeling makai 天津大学电子信息工程学院 School of Electronic Information Engineering

Company name

An example of parallel computing

School of Electronic Information Engineering

Problem description

359

0

10000000

1

)sin(k i

kk iLWSum =

Page 13: LOGO Parallel computing technique for EM modeling makai 天津大学电子信息工程学院 School of Electronic Information Engineering

Figure 3 Data flow chat

Page 14: LOGO Parallel computing technique for EM modeling makai 天津大学电子信息工程学院 School of Electronic Information Engineering

Company name

Comparison of the running result

School of Electronic Information Engineering

Core Time(s) Speedup Efficiency1 196.724 1.0 100%

2 99.735 1.9 95.0%

3 66.452 2.9 96.7%

4 52.235 3.8 95.0%

5 50.941 3.9 78.0%

6 46.406 4.2 70.0%

8 40.895 4.8 60.0%

Table1 Parallel on one computer

Page 15: LOGO Parallel computing technique for EM modeling makai 天津大学电子信息工程学院 School of Electronic Information Engineering

Company name

Comparison of the running result

School of Electronic Information Engineering

Core Time(s) Speedup Efficiency1 196.724 1.0 100%

2 95.722 2.0 100%

4 48.980 4.0 100%

6 34.680 5.7 95.0%

8 28.468 6.9 86.2%

10 25.409 7.7 77.0%

12 23.053 8.5 70.8%

Table2 Parallel on two computers

Page 16: LOGO Parallel computing technique for EM modeling makai 天津大学电子信息工程学院 School of Electronic Information Engineering

Company name

Comparison of the running result

School of Electronic Information Engineering

Core Time(s) Speedup Efficiency1 196.724 1.0 100%

3 64.221 3.0 100%

6 32.658 6.0 100%

9 22.353 8.8 95.0%

12 18.113 10.9 90.8%

15 16.925 11.6 77.3%

18 15.469 12.7 70.6%

24 13.748 14.3 59.6%

Table3 Parallel on three computers

Page 17: LOGO Parallel computing technique for EM modeling makai 天津大学电子信息工程学院 School of Electronic Information Engineering

LOGO

www.tju.edu.cn/seie/

www.themegallery.com天津大学电子信息工程学院

School of Electronic Information Engineering