End of Year Presentation

An Aspect-Oriented Approach to High Productivity High Performance Computing Chanwit Kaewkasi CNC Supervisor: John R. Gurd Advisor: Chris C. Kirkham

Motivation Scientific problems have become more complex. Performance is always critical for High Performance Computing (HPC) to solve scientific problems. But, Software Engineering techniques is needed to manage this complexity . The goal of the High Productivity Computing Systems programme: By 2010, systems should double productivity for every 18 months Not only hardware speed, but real values of the system [http://www.highproductivity.org], [Kepner, 2004]

An AOP Approach Reduces programs' complexity Introduces programming abstractions for HPC development, but Its weaving process impedes development cycle AOP on VM generates execution overheads Why is AOP important?

Separation of Concerns A real-world example Foods + Drinks + De s serts People can mix them together before they eat . Easier to digest  High Performance Do you prefer to eat them this way? SoC works that way How does AOP work for HPC?

AOP and HPC Separate parallelisation concerns from the algorithm Write an algorithm Apply parallel strategies, aspects, to the algorithm This approach allows simpler development life cycle Develop the serial version of the algorithm Develop a number of parallel strategies for different machines, architectures or platforms. [Harbulot and Gurd 2004, 2006]

Relative Productivity Is measured between any two programming languages (0 and L) [Kennedy et al. 2004]

Relative Productivity (cont'd) T(P) is the total time to solve the problem P I(P) is the development time C(P) is the average compilation time weighted by r 1 E(P) is the average execution time weighted by r 2

Concern Fencing Index concern 1 concern 2 CFI = 5 concern fences

Evaluation Strategy Relative Productivity Metric Development time , I(P) Compilation time, C(P) Execution time, E(P) Concern Fencing Index Program complexity (from AOP point of view) Experimental Environment Distributed Memory Machines, or Computational Grid

Success ? For a newly developed language L, Acceptance Goal Expected Goal

Separate parallelisation concerns Reduce complexity Raise level of abstraction Productivity Measurement Relative Productivity Model Concern Fences Reduce development time with A new AOP weaving technique Reduce execution overheads with A number of optimisation techniques Conclusion

Questions ? Thank you very much !

Behind the scene You shouldn't see this slide

Aspect-Oriented Programming Separates crosscutting concerns Provides abstraction s and constructs Join point abstractions Pointcut designators Advice codes Introduction [Kiczales et al. 1997]

Relative Productivity (cont'd) The proposed model:

Join point Shadows Traditional Approach for AOP

Transparent Join points Our Approach for AOP

Event-based AOP Another Dynamic AOP model Close to the Ideal AOP system Allows Re-defining Aspects at Runtime We take this approach for Block join points Allows fast weaving = Reduce development time =? High Productivity

Motivation Scientific problems become more complex. Performance is always the critical issue for HPC to solve scientific problems. But, we might need software engineering techniques to help manage this complexity . The goal of the High Productivity Computing Systems (HPCS) programme: By 2010, systems should double productivity for every 18 months Not only hardware speed, but real values of the system [http://www.highproductivity.org]

End of Year Presentation

More Related Content

What's hot (20)

Viewers also liked (19)

Similar to End of Year Presentation (20)

Recently uploaded (20)

End of Year Presentation