Application: TSD-MPI Calculation of Thermal Stress Distribution - PowerPoint Presentation

Slide1

Application:

TSD-MPICalculation of Thermal Stress DistributionBy Using MPI on EumedGrid Abdallah ISSA Mazen TOUMEH Higher Institute for Applied Sciences and Technology-HIAST Damascus – Syria

Africa

6

–Joint CHAIN/EPIKH/EUMEDGRID Support event in

School

on Application Porting- Rabat

, June 6, 2011

TSD-MPI “Purpose”To calculate temperature distribution and thermal stresses distribution around a different combination of heat resources inserted in acrylic cylindrical fin, by using Finite Element Analysis Method.Part of project “Design and build HIAST FEM SOFTWARE: Heat distribution and Thermal Stress Analysis depends on Finite Element Method”

TSD-MPI

“Theory-1”Stress analysis is an important part of engineering science, as failure of most engineering components is usually due to stress. The component under a stress investigation can vary from the legs of an integrated circuit to the legs of an offshore drilling rig, or from a submarine pressure hull to the fuselage of a jumbo jet aircraft..

TSD-MPI

“Theory-2”Thermal stress is a very important factor when examining the cause of failure in components. Thermal stress is created in the object structure mainly under the effect of the difference in the object part’s heat expansion. even at low temperature a large gradient of temperature can produce a fatal stress. Therefore it is necessary to establish a simple and easy method to analyze thermal stress of a body in all material

TSD-MPI

“Theory-3”The finite element method is a numerical technique for finding approximate solutions of partial differential equations (PDE) as well as of integral equations.To increase the accuracy in this methods means: increase in the number of element, decrease in the size of the element, select more complex element configuration and decrease the value of accepted approximation in the conjugate gradient method.

TSD-MPI

“Theory-4”In another words, increasing the accuracy increasing the needs to the more calculation power because now there is more data to analysis more results data to simulate and this mean need more time to do the job.

TSD-MPI

“Heat Distribution 3D-FEM Calculation”

TSD-MPI

“The Concept of Parting Input Geometrical Data”

TSD-MPI

“Geometrical Input Data”

TSD-MPI

“Caustics pattern calculation problem ”

TSD-MPI

“algorithm of work”problem can be solved by pipelined computation algorithm. The solid project in our case the cylinder should be divided into the same number of the processors.During the iteration in conjugate gradient method, in each time the solution become close to the right answer the processor will send the temperate value of the node matrix to the other processor.

TSD-MPI

“algorithm of work-2”In another words each processor is involve in one instance to solve during the time. After send there is a value matching for each node which is similar in coordination. In that case, the results should be more accurate. And more complex problem can be solved.

Solution

:TSD-MPI “Practice”Dividing the cylinder into discs:

Heaters

…

Discs

Each Disc will be a job:

Solution

:TSD-MPI “Implementation”

Searching for real 64-bit platform

Using more memory for data structure.

Platform needed to apply TSD-MPI

SL

5.4 64bit:

intel-fortran-64 bit

MPI package support Fortran language

Data structure

: Most of types must be in 64-bits.Other Matrixes 2D <Elements of nodes & auxiliary variables>.Global constant & variables.TSD-MPI

“

Implementation-2

”

Matrixes 2D

<Object’s definitions & properties>.

<Object’s matrixes # (40,000,000)

2

>.

Timing

:TSD-MPI “Practice”

1 thin disk #25minutes = 1 job.

10 cm as altitude of a cylinder

 1000 jobs:

In serial mode : 25 * 1000 = 25000m

≈

17.36 days

In parallel mode:

1m

: Submission.

10 – 20 minutes : Waiting & Scheduling.

25m : Running.

20m : Retrieving Data.

1 + 20 + 25 + 20 = 66m

Ξ

1Hour only