3DCS and Parallel Works Provide Cloud Computing for FAST Tolerance Analysis

Dimensional Control Systems | 2017 All Rights Reserved
Increase Your Analysis Speed with
Shared and Distributed Computing
Performance Improvement Series
-- With Special Guests --
Donald Gray, Magna Seating & Matthew Shaxted, Parallel Works

Ask Questions Throughout
Survey at the End for Free Trial!
Raise your hand if you have
audio troubles
Send in your questions as
early as you can!

October 12th - Black Box Methodology
-- Learn how to use Black Box Methods in 3DCS to
improve collaboration and reduce model complexity.
October 26th - How To Do Closed Loop Quality
-- How to incorporate real data into your analysis. Find
mean shifts, root cause production issues, test changes.
Plus! Coming Soon
What’s New in 7.4.1.0 – a new version walkthrough
Coming Up!

Presenters
Matthew Shaxted, Founder and President of Parallel Works, has led innovation efforts within several
large world-class architecture / engineering organizations, working at the intersection of parallel
computation, design and the built environment. Matthew now spends much of his time interacting with
Parallel Works customers in the built environment and manufacturing verticals, helping them unlock the
powers of high-performance computing in their business practices.
D.J. Gray, Product Engineering Specialist, has been working at Magna Seating for 18 years in various
engineering roles. In his current role as a DFSS Master Black Belt he works on design, analysis, and
optimization of structural products and mechanisms in addition to training and mentoring DFSS. In his
spare time he enjoys building complex 3DCS models.
Eric Kaphengst, Software Developer at Dimensional Control Systems, has been working at DCS for 5
years. He graduated from Michigan Technological University with a degree in physics.

Two New Solutions for Faster Results
Use Your Existing Hardware -
DCS Shared Memory
On Demand Cloud Computing
- with Parallel Works

● Faster analysis run times with no additional licenses
● Create more accurate models with higher complexity
● Simultaneously run multiple design configurations
● Run more simulation samples for higher accuracy statistics
● Concurrently run multiple analysis methods
● Continue modeling in 3DCS while simulations are running
Benefits of 3DCS Parallel Processing

Model Introduction
Donald Gray, Magna Seating

Some 3DCS Models are taking excessive time to
complete simulations
Models range from 13 - 54 hours
Sensitivity runs (.hlm) are sequential and add 2-3 hours
typically
What’s the Problem?

→ Traditional Model with feature points and DCS points (no mechanical or compliant
moves)
● Part Count - 10
● Tolerances - 54
● Surfaces - 270
● Measures - 295
○ Feature Measures - 67
● 5 different positions (build levels) analyzed
● Move Count - 91
→ Iteration moves - 32
○ With feature measures - 24
○ Nested iteration moves - 5
Model Info (54 Hour Model)

Additional challenges with the current situation:
Prefer to build model with dense feature meshes to create a more accurate model
Need to add in more iterative logic to better mimic physical world
10
Problem

Additional challenges with the current situation:
→ Would ideally like to run 5,000 simulation runs minimum
→ Would like to be able to simultaneously test different design factors (DOE)
→ Would like to have DCS license available for modeling and not tied up in
simulations 11
Problem

Shared Memory
Take Advantage of Multiple Computer Cores

• Shared Memory
– Block of memory that can be used by multiple CPUs
• CPU Core
– Independent processing unit (physical)
• Thread
– Sequence of programmed instructions (software)
Hardware Terminology

● Logical Processor
– Number of physical cores times the number of threads from hyperthreading
• Hyperthread
– Uses threads to split physical CPU core into 2 logical cores
Thread Technology

Shared Memory
Shared -
single computer

• Use System Information
to find # of logical
processors
Logical Processor

• Implemented into same dialog as run
• Thread value currently supports 0-4
How to Run Shared Memory
• 0 is single thread with legacy
seed generator
• 1-4 uses new seed generation
• Computer must have at least as
many logical processors

Shared Memory

Use the Cloud to Quickly Run Large Analyses
Distributed Computing

Parallel Works: A Hub for Big Compute

Distributed -
multiple computers
(on the cloud)
Shared -
single computer
Shared Memory vs Distributed Parallelization

Combine Shared and Distributed
Shared Memory w/ Distributed Computing

Linux
Windows
Enabling the 3DCS Batch Workflow

Runnable Now from Easy-to-Use Web Interface

Flexible system supports the
development of custom
workflow topologies
For example, Design of
Experiment for automated
benchmarking and custom
visualization in matplotlib.
Flexible for Sophisticated Workflows

Output
• Workflows output 2 files
– HTML report
– Raw data (hst, hlm, gf2)
• Easy to download these files
• Load these files into 3DCS

Magna Benchmark Results

Speedup Factor
1 Thread 4 Thread
1 Concurrent Baseline 3-4x
10 Concurrent 7-8x 16-20x
30 Concurrent 12-22x 28-52x

3DCS Analysis Methods
Platform 3DCS 3DCS 3DCS Parallel Works
Interface Graphical Batch Processor Web
Monte Carlo Simulation Yes Yes Yes
HLM Sensitivity Yes Yes Yes
GeoFactor Yes Yes Yes
AAO Geofactor Analyzer Yes Planned Planned
AAO Simulation Based Sensitivity Yes Yes Yes
Worst Case Analysis Yes Planned Planned
Design of Experiments n/a Planned Planned

3DCS now utilizes multiple CPU cores, and can take
advantage of powerful on-demand via the Cloud computing
-- Simple for users to take advantage of modern CPU architecture
-- Scale cloud computers to tens or even hundreds simultaneously
-- Drastically reduces time to solution of analysis - can run jobs concurrently
-- Simulation, Sensitivity, Geofactor and Simulation Based Sensitivity Supported!
31
Solution

Now What? (A Call to Action)

----- NOTES -----
Do NOT Go Beyond This Point

• Shared Memory Performance
– Single vs 4 Thread
• Magna Model
• CM Model
• SBS (reference only)
• Distributed Memory Performance
– Workers: 10,30,50 & Thread: 1*,4
• Magna Model (1 thread complete)
• CM Model
• SBS (reference only)
Results - Before & After

Webinar Agenda
Idea 1 - - Use Conf Presentation
---- Run comparison using shared data
---- Speed up on desktop, then additional with Cloud Computing
- Agenda (explain model) - discuss what goes into it, then showing data
on multi-core runs, then show with additional computing with cloud

Action Items
Formatting - Ben R
SBS - Eric K check - Maria H run simulations
Update Screen Caps of Parallel Works Interface - Eric K
Run Simulations
– Shared: 3DCS Simulations: Single vs 4 Thread
• Magna Model - Eric K
• CM Model - Maria H
– Distributed: Parallel Works Simulations
• Workers: 10,30,50 & Thread: 1*,4
– Magna Model (1 thread complete) - Matthew S
– CM Model - Maria H

Today’s Agenda Sept 21 2017
• Confirm Speakers X
• Update on materials X
• Formatting Discussion X
• Review Webinar Agenda X
• Discuss Schedule X
• Marketing Around Webinar -- PR, Blog, Guest Blog X
• Action Items X

Agenda September 26th - Tuesday
1.Review Updates
2.Check Action Items
3.Review Transitions
4.Practice Run

Survey + poll (Exit Survey Questions)
Would you be interested in a free trial of Cloud Computing
powered by Parallel Works?

• Model File
• Sim Run #
• Interaction #
• Concurrency
• FEA
Simulation-Based Sensitivity

• Model File
• Concurrency
• DoE Spreadsheet
• FEA
• Implementation
requires more
back-end work
• How to show
results?
DOE (Design of Experiments)

User Conf Presentation
https://docs.google.com/presentation/d/1qTR86qK_U1ylhgsns
f_blXwZ08GtU5vYdj4yNyqTg_o/edit#slide=id.g21d2bf70b0_3_
146

Benchmark Test Models
43
Model 1 Model 2 Model 3 Model 4
Runs
Local
Machine
Time (hrs)
10,000 2,000 2,000 2,000
1.8 13.9 25.3 53.7

Benchmark Results
44
# of Computers
Model 1 Model 2 Model 3 Model 4

3DCS and Parallel Works Provide Cloud Computing for FAST Tolerance Analysis

More Related Content

What's hot (17)

Similar to 3DCS and Parallel Works Provide Cloud Computing for FAST Tolerance Analysis (20)

Recently uploaded (20)

3DCS and Parallel Works Provide Cloud Computing for FAST Tolerance Analysis

Editor's Notes