Globus

1
Globus

• Ian Foster and Carl Kesselman
• Argonne National Laboratory and University of
  Southern California
• http//www.globus.org

2
Computational GridThe Definition

• A distributed computing infrastructure for
  coordinated resource sharing and problem solving
  in dynamic multi-institutional virtual
  organizations

3
Computational GridExamples

• Financial forecasting with ASP and SSP
• Industry consortium for feasibility study using
  multidisciplinary simulation
• Crisis management responding to oil spill
• Multi-institutional high-energy physics
  collaboration for analyzing petabytes of data

4
(No Transcript)
5
Computational GridCharacteristics

• Heterogeneous and dynamic environment
• Diverse and dynamic resource-sharing
  relationships across multiple administrative
  domains
• Performance critical

6
Computational GridA New Challenge?

• Differs from traditional distributed systems
• Resource sharing vs. Information sharing
• Peer-to-peer vs. Client/Server
• Computation vs. Communication
• Differs from traditional parallel computing
• Loosely coupled, heterogeneous, and dynamic
  systems
• Spans over multiple administrative domains

7
Design Issues

• Interoperability vs. Flexibility
• Performance vs. Convenience
• Local control vs. Global coordination

8
Design Philosophies

• A bag of services
• The hourglass principle a balance between
  interoperability and flexibility
• Translucency a balance between performance and
  convenience
• Layered design Enabling global coordination
  while maintaining local control

9
Design PhilosophiesThe Hourglass Principle
10
Design PhilosophiesTranslucency

• Managing heterogeneity, not simply hiding it
• Transparency for convenience
• Exposing certain low-level details to facilitate
  performance optimization

11
Design PhilosophiesTranslucency Examples

• Provide ways to discover and control aspects of
  the underlying system
• Reliability or Low latency
• Security or No security
• Message passing, Shared memory, or IP

12
Design PhilosophiesLayered Design
13
Design PhilosophiesLayered Design Layers

• Fabric Interfaces to local control
• Connectivity Communicating easily and securely
• Resource Sharing local resources
• Collective Coordinating multiple resources
• Applications

14
Design PhilosophiesLayered Design Examples
15
Globus Layers
16
Conclusion

• Computational grid poses challenges that are
  beyond the current state of art in distributed
  systems and parallel computing
• Globus provides an infrastructure to addressing
  these issues with interesting design philosophies