The Grid the united computing power Jian He Amit Karnik
Outline History and vision Motivation Application scenarios Architecture Challenges Approaches –Language-related –Object-based –Toolkit: Globus Future directions References
History and Vision From History – Late 1980s, “metacomputing” was coined. – Early 1990s, “Gigabit Testbeds”, from bandwidth- oriented to application-oriented. – 1995, “network computing” at SC95 – 1998, NCSA, NSF, NASA, DOE, Al Gore … Vision (analogy to “Electric Power Grid”) – Dependable: performance guarantees. – Consistent: uniform interfaces. – Pervasive: “plug-in” from everywhere.
Motivation Resource sharing – Episodic – Low utilization Coordinated PSEs Emerging tools and techniques From
Application Scenarios Distributed supercomputing High-throughput computing On-demand computing Data-intensive computing Collaborative computing Grid: high-performance sharing computational power Web: sharing documents
Application Scenarios (cont.) High-throughput computing “A computing environment that delivers large amounts of computational power over a long period of time.” (example: NEOS & Condor) Internet Client NEOS Server Submit jobs Return results NEOS Solver library Condor GW matchmaker scheduler monitor protector UNIX COW
Architecture From
Challenges The nature of applications Programming models and tools System architecture Problem-solving methods Resource management Security End systems Instrumentation and performance analysis Network protocols and infrastructure
Approaches Languages, compilers and libraries –MPICH-G2 (a grid-enabled MPI - message passing interface) Object-based approaches –Legion –Commodity Computing (Three-tier, CORBA, Java/Jini) Toolkit –Application-specified toolkit (NetSolve) –Service-oriented toolkit (Globus)
Approaches (cont.) Language-related: MPICH-G2=Globus services + MPI (message passing interface) –Distributed memory message passing –Portability –Heterogeneity Communication Network M PP M P M Application MPI IBM SP2 Cray T3D/E SUN. FORTRAN or C
Approaches (cont.) Everything is an object Classes manages their instances Users can provide their own classes Core objects implement common services Object-based: Legion - Architecture & object model From
Approaches (cont.) Object-based: Commodity computing Example: CORBA fits within the Grid Architecture From
Approaches (cont.) Application-specified toolkit: NetSolve CFortran MatlabCustom PSEs and Applications Globus proxy NinfLegion Ninf proxy Legion proxy NetSolve proxy Middleware Resource Discovery System Management Resource Scheduling Fault Tolerance From Metacomputing resources
References Book I. Foster and C. Kesselman, “The Grid: Blueprint for a New Comuputing Infrastructure”, Morgan Kaufmann, Papers Michael C. Ferris, Michael P. Mesnier, Jorge J. More, “NEOS and Condor: Solving optimization problems over the Internet”, April, 1998 David Henty, “The Grid - A Critical Review of Current Status and Future Directions in Grid Technology”, Oct.,