Transcript An Introduction to Parallel Computing With High Performance

Parallel Computing With High Performance
Computing Clusters (HPCs)
By Jeremy Cathey
Super Computing
• Began as early as late 1950’s
• Compute large mathematical calculations
• Large scale task management
Advantages of Super Computing
• Power
• Configuration
• Administration
Disadvantages of Super Computing
• Cost
• Scalability
• Diversity
Alternatives
• Distributed Computing
- Uses queuing system to manage multiple
tasks on several computers serially
• Parallel Computing
- Uses several computers to work on one
task in parallel
Implementations of Alternatives
• Shared Processors (Multiple Processors)
• Shared Memory Pools
• Messaging Systems
Advantages of Parallel Computing
• Cost
• Scalability
• Diversity
Disadvantages of Parallel Computing
• Power
• Configuration
• Administration
Parallelization Techniques
• Manual
- Parallelization implemented by the
programmer at creation of the program
• Automatic
- Parallelization implemented on the fly by
the operating system
Advantages of Manual Parallelization
• Specific Customization
• More Effective Parallelization
Disadvantages of Manual Parallelization
• Difficult to Implement
• Not Very Portable
Advantages of Automatic Parallelization
• Easier to Implement
• Portable Applications
Disadvantages of Automatic Parallelization
• Less Effective Parallelization
Project Goals
• Reduce Disadvantages of Parallel
Computing
• Learn more about Operating System
development and parallel optimization
• Provide a solution for optimized parallel
computing for Dr. Salazar’s research
Goal Techniques
• Reduce power issue of parallel computing
machines by custom building an operating
system kernel for optimization
• Research and Improve Open-Mosix
Parallelization Technique
• Create a cluster that is easy to assemble
and disassemble
• Create and/or implement an interface that
simplifies cluster administration
Operating System Choice
•
•
•
•
•
Bill!!!!
Windows Vs. Linux
Live vs. Install
Community Support
Creation vs. Customization
HARDWARE SUPPORT!!!!!!!!
Implementation
• Dr. Salazar: Professor of Chemistry
• Use cluster to reduce time for
sophisticated mathematical computation
Simulation Details
• 2.6 GHz Computer - 2GB RAM
• 1.2 GHz Computer - 256MB RAM
• 5 Parallel 1.2GHz Computers – 1.28GB RAM
Computer Simulation Results
80.00
70.00
Hours to Completion
60.00
50.00
1.2GHz
Group 2
40.00
30.00
20.00
10.00
0.00
0
5000
10000
15000
20000
Number of Atoms
25000
30000
35000
Computer Simulation Results
80.00
70.00
Hours to Completion
60.00
50.00
Group
2.6GHz1
1.2GHz2
Group
40.00
30.00
20.00
10.00
0.00
0
5000
10000
15000
20000
Number of Atoms
25000
30000
35000
Computer Simulation Results
80.00
70.00
Hours to Completion
60.00
50.00
2.6GHz1
Group
1.2GHz2
Group
Parallel
Parallel3
Group
40.00
30.00
20.00
10.00
0.00
0
5000
10000
15000
20000
Number of Atoms
25000
30000
35000
Hurdles
• HARDWARE DETECTION!!!!!!
• HARDWARE DETECTION!!!!!!
• HARDWARE DETECTION!!!!!!
Did I Mention:
• HARDWARE DETECTION!!!!!!
• HARDWARE DETECTION!!!!!!
• HARDWARE DETECTION!!!!!!
Improvements
• Redesign Operating System Kernel For
More Advanced HARDWARE DETECTION
• Try Simulations on More Advanced
Computers
• Create Custom Installer for Distribution
Special Thanks
• Dr. Kirk: For providing computers for experiment
and for finding a place for me to store my project
• My Wife: For not killing me while I had to store
my project at our home
• Dr. Wilms: for providing older versions of Red
Hat, and for Operating Systems Class
• Dr. Salazar: For time and effort spent helping me
understand a SMALL part of his research
Questions? Comments?
Derogatory Remarks?