web computing and scientific software development tung-han hsieh ( 謝東翰 ) assistant technician...

7
Web Computing and Scientific Web Computing and Scientific Software development Software development Tung-Han Hsieh ( Tung-Han Hsieh ( 謝謝謝 謝謝謝 ) ) Assistant Technician Assistant Technician Research Center for Applied Sciences, Research Center for Applied Sciences, Academia Sinica Academia Sinica

Upload: randolph-sullivan

Post on 18-Jan-2016

216 views

Category:

Documents


3 download

TRANSCRIPT

Page 1: Web Computing and Scientific Software development Tung-Han Hsieh ( 謝東翰 )  Assistant Technician Research Center for Applied Sciences, Academia Sinica

Web Computing and Scientific Web Computing and Scientific Software developmentSoftware development

Tung-Han Hsieh (Tung-Han Hsieh ( 謝東翰謝東翰 ))

Assistant TechnicianAssistant Technician

Research Center for Applied Sciences, Research Center for Applied Sciences, Academia SinicaAcademia Sinica

Page 2: Web Computing and Scientific Software development Tung-Han Hsieh ( 謝東翰 )  Assistant Technician Research Center for Applied Sciences, Academia Sinica

Web ComputingWeb Computing

System architecture

- Web interface for scientific computation.- Cluster filesystem for distributed I/O.- Infiniband network for high speed communication- High performance GPU computing

• Ab-initio first principle nano-structure & chemicalcomputation.• Research and development platform for ourresearch fellows and collaborators.

Page 3: Web Computing and Scientific Software development Tung-Han Hsieh ( 謝東翰 )  Assistant Technician Research Center for Applied Sciences, Academia Sinica

The HPC Facilities of RCASThe HPC Facilities of RCASName Machine type CPU # of nodes # of cores Memory Disk space Conn. Year

Warp HP Blade Xeon 5160 48 192 16 - 32 GB 9 TB Gigabit 2006

Quad JCNet 1U Xeon X3220 24 96 8 GB 3.7 TB Gigabit 2007Kohn JCNet 1U Xeon X3220 10 40 8 GB 6.8 TB Gigabit 2007

IBM Blade Xeon E5430 14 112 32 GB Gigabit 2008IBM Blade Xeon E5530 14 112 48 GB Infiniband 2009IBM Blade Xeon X5650 6 48 24 GB Infiniband 2010

IBMYC IBM Blade Xeon E5430 16 128 32 GB 16 TB Gigabit 2008

IBM Blade Xeon X5650 12 96 24 GB Gigabit 2010

SC GPU Tesla C1060 1 2 GPUs 4 GB Gigabit 2010SC GPU Tesla C2050 1 2 GPUs 4 GB Gigabit 2010

Tesla PC GPU GTX 280 2 4 GPUs 1 GB 500 MB Gigabit 2010

HP Blade System (C-class) IBM BladeCenter (H-class) Supermicro GPU server

Page 4: Web Computing and Scientific Software development Tung-Han Hsieh ( 謝東翰 )  Assistant Technician Research Center for Applied Sciences, Academia Sinica

Adaptive Thick Restart Lanczos AlgorithmAdaptive Thick Restart Lanczos Algorithm(for low-lying eigenmode projection)(for low-lying eigenmode projection)

Dedicate for Hermition matrix A. Can attain > 50% better performance than ARPACK

k is optimized for the best performance via maximize of the following object function

Page 5: Web Computing and Scientific Software development Tung-Han Hsieh ( 謝東翰 )  Assistant Technician Research Center for Applied Sciences, Academia Sinica

Scientific package developmentScientific package development(coordinated by Dir. Chang)(coordinated by Dir. Chang)

Motivation Current available packages treat the whole system

altogether, which requires enormous computing resources. Cannot handle large complicated nano-particles, nor the

interaction between nano-particles.

We need a new ab-initio computation method to handle large complicated nono-structures with good accuracy.

IdeaUsing the boundary-integral method to solve the whole Green's function from the ones in sub-domains.

Page 6: Web Computing and Scientific Software development Tung-Han Hsieh ( 謝東翰 )  Assistant Technician Research Center for Applied Sciences, Academia Sinica

Scientific package developmentScientific package development

This is the on-going development project

Page 7: Web Computing and Scientific Software development Tung-Han Hsieh ( 謝東翰 )  Assistant Technician Research Center for Applied Sciences, Academia Sinica

OutlooksOutlooks For HPC development:For HPC development:

• Software porting to GPU platform.Software porting to GPU platform.• Large scale parallel computation.Large scale parallel computation.

For scientific software development:For scientific software development:• Complete the implementation of the boundary Complete the implementation of the boundary

integral method software.integral method software.• Investigate novel numerical algorithms.Investigate novel numerical algorithms.• Get involve into more scientific software Get involve into more scientific software

developments.developments.

For informatics of RCAS:For informatics of RCAS:• Web applications development.Web applications development.• Network security improvement.Network security improvement.