11
HPC - High Performance Productivity Computing and Future Computational Systems: A Research Engineer’s Perspective Dr. Robert C. Singleterry Jr. NASA Langley Research Center HPC User Event October 30 th , 2010

HPC - High Performance Productivity Computing and Future Computational Systems: A Research Engineer’s Perspective Dr. Robert C. Singleterry Jr. NASA Langley

Embed Size (px)

Citation preview

Page 1: HPC - High Performance Productivity Computing and Future Computational Systems: A Research Engineer’s Perspective Dr. Robert C. Singleterry Jr. NASA Langley

HPC - High Performance Productivity Computing and

Future Computational Systems: A Research Engineer’s

Perspective

Dr. Robert C. Singleterry Jr.

NASA Langley Research Center

HPC User Event

October 30th, 2010

Page 2: HPC - High Performance Productivity Computing and Future Computational Systems: A Research Engineer’s Perspective Dr. Robert C. Singleterry Jr. NASA Langley

Outline

What is an Engineer’s Job High Performance Computing High Performance Computation? High Productivity Computation Future of Computation Summary

Page 3: HPC - High Performance Productivity Computing and Future Computational Systems: A Research Engineer’s Perspective Dr. Robert C. Singleterry Jr. NASA Langley

What is an Engineer’s Job

Example: Space Radiation Engineer• Keep astronauts safe (<3% excess fatal cancer)• Predict, measure, analyze space exposure• Research what to predict, measure, and analyze

Things to Consider• Materials• Vehicle design• Human biology• Electronic design and operation• Mass, cost, and feasibility• Physics, chemistry, mathematics• Computational aspects

Page 4: HPC - High Performance Productivity Computing and Future Computational Systems: A Research Engineer’s Perspective Dr. Robert C. Singleterry Jr. NASA Langley

High Performance Computing

Utilization of computers to solve a problem• Enable the fastest execution of a computer

program to get an answer (usually LINPACK)• Time spent optimizing computer program to

computer hardware (not counted in above)• Usually one algorithm based somewhat in first

principles science with appropriate mathematical approximations

• Computer hardware and algorithm usually not highly compatible unless written especially for that hardware (CM-2 versus x86 cluster code)

• Can be very labor intensive and expensive but not counted in the performance calculation

Page 5: HPC - High Performance Productivity Computing and Future Computational Systems: A Research Engineer’s Perspective Dr. Robert C. Singleterry Jr. NASA Langley

High Performance Computation?

Analysis of (space radiation example)• Materials (carbon or aluminum)• Vehicle designs (runabout or capsule)• Human biology (shielding or pharmaceuticals) • Electronic designs (shielding or resistance)• Mission (cost and integrated performance)• Basic science parameters

Each analysis is not a single algorithm that fits a single hardware architecture

Can be linked to analyze an astronaut’s trip but what is most important to this process?

Page 6: HPC - High Performance Productivity Computing and Future Computational Systems: A Research Engineer’s Perspective Dr. Robert C. Singleterry Jr. NASA Langley

High Performance Computation?

Creativity, analysis, and humans• Ex: Vehicle design

• Creativity of what it looks like and how it operates• Objectives of mission planners• Requirements generation from engineers• CAD creation of human ideas and objectives• Analysis for space radiation protection (and others!)• Iterate until all requirements are met

High Performance Computation is a small part of this entire task

Page 7: HPC - High Performance Productivity Computing and Future Computational Systems: A Research Engineer’s Perspective Dr. Robert C. Singleterry Jr. NASA Langley

High Productivity Computation

Different than high performance computing! How can the computer, as a tool, increase

productivity in completing tasks? First Question: computer?

• Can 1 computer or computer architecture solve all computational needs? Probably not!

Second Question: Do the varied algorithms needed fit the hardware available? No!

Third Question: Will the computer be operated for needed throughput? Depends!

So what happens next?

Page 8: HPC - High Performance Productivity Computing and Future Computational Systems: A Research Engineer’s Perspective Dr. Robert C. Singleterry Jr. NASA Langley

Future of Computation

Caveat: Predictions more than 3 to 5 years in the computer field are subjective at best

By 2016, NASA should have a million powerful and tightly coupled cores (x86?) in its main computer system

IBM is looking at 10-100 million cell phone cores to get exa-scale computing soon

Floating point operations is what is needed What could be in the future?

• This opinion is highly debatable by the way!!!

Page 9: HPC - High Performance Productivity Computing and Future Computational Systems: A Research Engineer’s Perspective Dr. Robert C. Singleterry Jr. NASA Langley

Future of Computation

Compiler and compiler tools to help generate code that will fit the computer• Better language• Better distributed memory utilization• Better floating point throughput (10-10000000x)

Research algorithms to better fit the computer architecture• Cellular automata to replace finite elements• Hybrid Monte Carlo and deterministic methods

Key word: BETTER and not perfect• Don’t need 100% utilization of the computer• Better utilization than what is available now

Page 10: HPC - High Performance Productivity Computing and Future Computational Systems: A Research Engineer’s Perspective Dr. Robert C. Singleterry Jr. NASA Langley

Future of Computation

Last, but not least: Computer Operations• A time share machine?• A capability machine?• Multiple machines?• Is a desktop is faster than a cluster?

Coherent strategy for computation needed by the organization to fit the work being done by the scientists and engineers

Flexibility of the strategy to meet needs• New computers• New algorithms

Page 11: HPC - High Performance Productivity Computing and Future Computational Systems: A Research Engineer’s Perspective Dr. Robert C. Singleterry Jr. NASA Langley

Summary

PRODUCTIVITY over PERFORMANCE Coherent computational strategy to match

user needs: engineers, operations, research Better

• Language• Distributed memory utilization• Floating point throughput• Algorithms to fit architecture• Computer operations to fit user needs

Ease of use of computers to fit overall tasks• Computers are just one of many tools utilized