The SG@home SG@home Project: Project: The Experience Sharing Experience Sharing Hing-Yan LEE, Jon LAU Khee-Erng and Nigel TEOW National Grid Office Singapore International Symposium on Grid Computing 2007
PC Grid Computing PC Grid Computing • Stages – 1: Ideas Competition – 2: Selection – 3: Development – 4: Resource Donation • 3 categories – Schools – Institutes of Higher Learning – Open Joint Organizers with Sponsored by
Summary of Entries Summary of Entries • Breakdown of the entries by categories: – Schools 654 – IHLs 27 – Open 30 Total: 711
Background Background • Chairman (National Grid Steering Committee) met Permanent Secretary (Ministry of Education) on 1 September 2005 • Aim is to help rekindle interest in science & engineering, thus encouraging more students to become engineers • NGO will undertake the grid-enabling of the school PCs
Background Background • Grid Computing would provide the opportunity for upper secondary school & JC students to gain stimulating experiences in the science subjects • Singapore students have great appetite for learning . Grid Computing could harness the large number of PCs available in the schools to make a big impact in education
Proposal Proposal • Organize a vacation camp • Build a PC Grid infrastructure using GridMP from United Devices & supported by Singapore Computer Systems • Bring 2 existing grid-enabled applications onto SG@Schools to demonstrate capability • Assign domain-specific mentors to work with students • Provide assistance to 2 schools to grid-enable an application each
Vacation Camp Programme Programme Vacation Camp • Opening Ceremony – GOH: Mr. Richard Lim, CE DSTA – Invited speaker: A/Prof. Tan Tin Wee (NUS) • Introduction to Grid Computing & PC Grids • Demonstration of PC-Grid applications • Hands-on grid-enabling of application • Closing Ceremony
Vacation Camp Programme Programme 2005 2005 Vacation Camp Time Day 1 Day 2 0900 Opening Ceremony Hands-on: Application Development (Part 1) 1000 Break Break 1030 Introduction to Grid Computing & Hands-on: PC Grid Computing Application Development (Part 2) 1200 Lunch Lunch 1300 Demonstration of PC Grids & Hands-on: Installation of Clients (Part 1) Application Development (Part 3) 1530 Break Break 1600 Demonstration of PC Grids & Hands-on: Installation of Clients (Part 2) Application Development (Part 4) 1730 Closing Ceremony 1800 End of Day 1 End of Camp
TALKS INVITED TALK OPENING ADDRESS BY TAN TIN WEE BY RICHARD LIM NATIONAL UNIVERSITY OF SINGAPORE CHIEF EXECUTIVE, DSTA
Vacation Camp Programme Programme 2006 2006 Vacation Camp Time Day 1 Day 2 0900 Opening Ceremony Hands-on: • Speeches Application Development (Part 1) • Invited Talk by CTO, United Devices (BOINC, Grid MP, or XGrid) • Presentation on Grid-enabled projects 1000 Break Break 1030 Introduction to Grid Computing & Hands-on: PC Grid Computing Application Development (Part 2) (BOINC, Grid MP, or XGrid) 1200 Lunch Lunch 1300 Demonstration of PC Grids & Hands-on: Installation of Clients (Part 1) Application Development (Part 3) (BOINC, Grid MP, or XGrid) (BOINC, Grid MP, or XGrid) 1530 Break Break 1600 Demonstration of PC Grids & Hands-on: Installation of Clients (Part 2) Application Development (Part 4) (BOINC, Grid MP, or XGrid) (BOINC, Grid MP, or XGrid) 1700 Closing Ceremony 1800 End of Day 1 End of Camp
TALKS INTRODUCTION TO GRID INVITED TALK COMPUTING & PC-GRIDS BY JIKKU VENDAT BY BERTIL SCHMIDT CTO, UNITED DEVICES INC NANYANG TECHNOLOGICAL UNIVERSITY
PC- -Grid Middleware Grid Middleware PC Presentation (3 Tracks) Presentation (3 Tracks) • GridMP by United Devices – Supported by Singapore Computer Systems • BOINC (Open-source) – Supported by ST Electronics • XGrid by Apple – Supported by Elchemi Education & Apple
BOI NC BOI NC • Open source PC Grid software – Powering SETI@home & many others • BOINC efforts – BOINC SG@Schools seminar on 15 May 2006 – BOINC tutorial on 16 May 2006 – Presented by Dr. David Anderson (founder of SETI@home & UC Berkeley scientist) • ST Electronics provides systems support & technical support for grid-enabling of applications
Grid TRACKS TRACKS -Grid PC- PC BOINC BOINC
-Grid TRACKS Grid TRACKS PC- PC GridMP GridMP
XGRID XGRID Grid TRACKS TRACKS -Grid PC- PC
Vacation Camp 2005 & 2006 – About 70 students & teachers per year Jointly organized by
• Infrastructure – A PC grid linking Hwa Chong Institution and Raffles Institution with 200 United Devices clients • Participating Schools in vacation camp – Hwa Chong Institution – National Junior College – Raffles Institution/Junior College – Temasek Junior College – Victoria Junior College – Dunman High School – Anglo-Chinese School (Independent) – Nanyang Girls High School – Raffles Girls School – Maris Stella High School – River Valley High School – CHIJ St Nicholas
• Beyond vacation camp … – Raffles Institution • Mentors: Zhang Xinhuai (NUS) & Dr. Bryan Lee (BII) • Project: Auto-docking – Hwa Chong Institution • Mentor: Dr. Li Haizhou (I 2 R) • Project: Speech modelling
Enabling Autodock for a Grid Enabling Autodock for a Grid Environment Environment • Computational prediction of ligands binding to protein target • Well-established docking tools/algorithms for protein-ligands docking from The Scripps Research Institute • A computational technique used for – Drugs/Potential-Drugs discovery & development – Protein-Drugs interaction studies • Source code available for academic development
[Courtesy of Raffles Institution team]
Why Grid- -enable Autodock? enable Autodock? Why Grid • Issue: – Docking of chemical library against a protein target – Chemical library consists of thousands to millions of ligands – Docking of 1 ligand will take approx 6 to 12 hours – A screening job of 1000 ligands may take 6000 – 12000 hours on single computer • Objective: – Shorten the time to obtain docking results [Courtesy of Raffles Institution team]
What Was Done What Was Done • Approach – Break down each job into smaller execution units. • Identify which parameters have to be changed – Each execution unit is distributed to different machines on the Grid which process them separately. – Retrieve the results from each computer and compile them to get the final result. • Each result is concatenated to form the final result. [Courtesy of Raffles Institution team]
Benefits Benefits • Speed increase of over 50% – Test job (5 workunits of 50 results) took approximately 23 hours – Computational time on a single dedicated machine would have taken more than 50 hours – Limited increase due to lack of computers. There will be a larger speed increase on a bigger grid. Grid computers may not be dedicated, restricting the speed increase • Running job on multiple computers – In case of failure on one computer, the server can re-submit to another computer [Courtesy of Raffles Institution team]
RI - - Conclusion Conclusion RI • Good experience • Exposure to grid environment coding • Learn to work together better • Successful completion of 1 st grid enabling attempt • Reflections – Grid-enabling an application is challenging (for 1st- timers) – Grid computing has great potential in computationally intensive tasks [Courtesy of Raffles Institution team]
Grid- -Enabling Speech Processing Enabling Speech Processing Grid Separate the Detect single Determine if there pitch tracks Speech speaker are 0, 1 or 2 (fundamental Recordings segments people speaking frequency) [Courtesy of Hwa Chong Institution team]
What I t I s For What I t I s For • When input voice consists of more than 1 speaker, it is hard to transcribe the extra one voice • Identify & differentiate the voices • Multipitch - part of speech recognition system used for e.g. meeting transcription • Why grid-enable the application? – 36GB data to be processed! – On a single PC, estimated time required to process all data is 285 days! – Impractical to run on single PC for 285 days – No dedicated PC to process the data – Grid-enabling can speed up processing [Courtesy of Hwa Chong Institution team]
Benchmark Results Benchmark Results ~32X speedup~ ~32X speedup~ 300 250 200 150 100 50 0 Single PC - Theoretical Grid Network (Elapsed Time) Time taken (in days) [Courtesy of Hwa Chong Institution team]
HCI - - Conclusion Conclusion HCI • Grid-enabling can reduce the processing time for large amounts of data • Users can work on even larger amounts of data with increasing parameters • Learning Points – Gained knowledge on Speech Recognition technology – Understand better Grid technology • Opens up a new realm of computing • Grid has immense potential [Courtesy of Hwa Chong Institution team]
2007 School Projects 2007 School Projects • RI has started this year ’ s grid-enabling projects 1) Continuation work on Autodock • Windows client version • Web submission interface 2) POV-Ray Animation • HCI will start on their project in April.
Recommend
More recommend
