Вы находитесь на странице: 1из 16

PARALLEL COMPUTATION IMPLEMENTATION OF CONJUGATE GRADIENT METHOD TO SOLVE THE POISSONS EQUATION IN TWO-DIMENSIONAL POTENTIAL DISTRIBUTION

R. M. Ibrahim Department of Physics Institut Teknologi Bandung

INTRODUCTION
Basically 1 program for 1 computer Algorithm parallelization Utilize resource Minimize time

Parallel Computation

Problem

What factors does affect parallel computation of conjugate gradient performance?

CONJUGATE GRADIENT METHOD


Solve large Linear Equation System Ax = b Often used in Physical Computational System

Iterative Method
Use conjugate of gradient approach

CONJUGATE GRADIENT METHOD: ALGORITHM


r0 b Ax0 p 0 r0 k 0
Yes
Start

End

No

rkT rk
T pk Ap k

k<n

x k 1 x k k p k rk 1 rk k Ap k

rkT11 rk 1 rkT rk
No rk+1rTk+1 < etol.

p k 1 rk 1 k p k k k 1

Yes

PARALLEL COMPUTATION: PERFORMANCE


Parallel computation execution time equation

T parallel
tcomp = computation time tcomm = communication time p = number of processors

t comp t comm p

PERFORMANCE ASPECTS OF PARALLEL COMPUTATION


Speedup
Tserial S ( p) Tparalel

Performance : Problems

Parallel Computation
Granularity

Scalability (Efficiency)

DISTRIBUTED COMPUTATION SYSTEM INFRASTRUCTURE DESIGN


Server Job
All Results
Job All Results

Client

Client

Scheduler or Job Manager

Switch Server

Task Results

Worker Switch Client

Task
Results Task

Worker
Client Worker

Worker

Results

SYSTEM SPECIFICATION
Node Multicore Computer Function Test the performance of parallel programs on multicore computers MATLAB parallel computing Scheduler Store the operating system image for Fat Client configuration Worker Providing networking applications to support parallel computing operations MATLAB Run MDCS applications to form MDCS cluster system Undertake the task of Job Manager Running Client Session Run PCT for parallel computing Connecting Client, Job Manager, and Worker physically Specification Processor: Xeon Quadcore (4 core) 2,9 GHz Memory: 8 GB NIC: 2 Gigabit Ethernet 1000 MBps Hardrive: 1x160 GB Processor: Dual Core 1,6 GHz Memory: 2 GB NIC: Fast Ethernet 100 MBps Hardrive: 1x80 GB, 1x160 GB Processor: Core 2 Duo 2,6 GHz Memory: 2 GB NIC: Fast Ethernet 100 MBps Processor: Centrino Core Duo 2x1,6 GHz Memory: 2,5 GB NIC: Fast Ethernet 100 MBps Quantity

Master Server

Worker

Client

Switch

Fast Ethernet Switch 24 port

RESULT: SPEEDUP ANALYSIS

Multinode

Multicore

RESULT: GRANULARITY ANALYSIS

Multinode

Multicore

RESULT: EFFICIENCY ANALYSIS

Multinode

Multicore

CONCLUSIONS
the performance of parallel execution on multicore computer environment is better than the performance of parallel execution in multinode environment. In this research shows that parallel execution on multicore performance is determined by computational factor than the communication factor. While in the multinode computing environment, parallel execution is strongly influenced by the communication factor.

SUGGESTIONS
Further research is needed to study the performance of conjugate gradient parallel execution with other parallelization strategies, such as using other parallel programming interface, namely OpenMP, MPICH, etc. In addition, the parallelization strategy can also be done using other computational units such as the GPU (Graphics Processing Unit). To improve the execution performance of parallel conjugate gradient on multicore, required an increase in the use of hardware, such as using a computer that has more processor cores, higher clock, and large memory capacity. As for improving the performance of multinode execution, required an increase in the use of network hardware that is used such as the use of Gigabit network interface Ethernet/10 Gigabit Ethernet / Myrinet or use the IPv6 protocol.

ACKNOWLEDGEMENT
The author wishes to thank Mr. Rizal Kurniadi as a lecturer also for his advices for this final assignment.

REFERENCES
Barney, B. (2011, August 17). Introduction to Parallel Computing. Retrieved August 17, 2011, from Lawrence Livermore National Laboratory: https://computing.llnl.gov/tutorials/parallel_comp Grama, A., Gupta, A., Karypis, G., & Kumar, V. (2003). Introduction to Parallel Computing, 2nd Edition. Edinburgh: Addison Wesley. Mursito, E. (2009). Pengembangan Komputer Kinerja Tinggi di Teknik Fisika ITB. Seminar Nasional Teknik Fisika, (pp. 1-8). Bandung. Schewchuk, J. R. (1994). An Introduction to The Conjugate Gradient Method Without The Agonizing Pain, Edition 1 1/4. Pittsburgh: School of Computer Science Carnegie Mellon University. Schmidt, E. (1908). Title Unknown. Rendiconti del Circolo Matematico di Palermo, (pp. 53-77). Palermo.

Вам также может понравиться