September 5, 2002
CMCS 818S - Alan Sussman 1
CMSC 818S
Computational Grids
Alan Sussman
September 5, 2002
CMSC 818S - Alan Sussman 2
Administrivia
• Class introductions
•Project
– MPI project coming soon
• it will count, but only for a small part of your grade
– sign up for a Linux cluster account if you don’t have
one already (redleader.umiacs.umd.edu)
• Next class will be led by Henrique Andrade, on
Globus toolkit
– chapter/paper available by tomorrow – see Readings
web page
• We’ll talk about choosing topics to present next
Thursday
CMSC 818S - Alan Sussman 3
What is a Grid?
• The hardware and software infrastructure that
provides computing resources that are:
– dependable – performance guarantees
– consistent – standard services/interfaces
– pervasive – available everywhere supported
– inexpensive – economic argument
• “Coordinated resource sharing and problem
solving in dynamic, multi-institutional virtual
organizations” – from The Anatomy of the Grid
(Foster and Tuecke, Argonne/UI-Chicago)
CMSC 818S - Alan Sussman 4
Why Grids?
• To provide more computing power
– from new technology – take advantage of new
hardware, wherever it is located
– to share available resources on demand (like time-
sharing)
– to increase utilization of underused resources (cycle
stealing)
– to share computational results – collaboratories
– take advantage of new tools and techniques
• e.g., network enabled solvers (Netsolve @UTK, Ninf in Japan)
• teleimmersion – collaborative data exploration/ analysis
(Discover @ Rutgers)
CMSC 818S - Alan Sussman 5
Analogy to the Electric Power Grid
• Salient characteristics
– heterogeneity
• generators/outlets vs. machines/networks
– consumers have different requirements
• power consumption/service guarantees/$$$ vs.
computational requirements/QOS/$$$
– want to take advantage of economies of scale
– politics
• local control, but interfaced to uncontrolled
environment – need standards
CMSC 818S - Alan Sussman 6
Applications
• Distributed supercomputing
– to maximize available resources for large
problems
• within an organization, or across organizations
– examples include distributed interactive
simulation, simulating complex physical
processes
– challenges include co-scheduling, scalability of
services to large numbers of nodes, tolerating
latency, and obtaining high performance across
heterogeneous systems
