|Volume 7, Issue 1 -
Globus: Building a Computational Grid
Ian Foster, Argonne National Laboratory; Carl Kesselman, University of Southern California Information Sciences Institute
The Globus project is investigating and developing the technologies required to support the construction of a national-scale computational grid. If successful, this grid will use high-speed networks to connect supercomputers, databases, scientific instruments, and advanced display devices across the country. Availability of this grid will enable new classes of high-performance applications that require unique capabilities not available in a single computer or at a single site. It will also increase accessibility to supercomputing capabilities and the impact of supercomputing technologies.
Experience with high-speed networking testbeds has convincingly demonstrated that there are applications of considerable scientific and economic importance that can benefit from networked supercomputing capabilities. For example, the Information Wide Area Year (I-WAY) network experiment connected supercomputers and other resources at 17 different sites across North America. Deployed at Supercomputing '95, the I-WAY enabled 60 groups to develop applications in areas as diverse as large-scale scientific simulation, collaborative engineering, and supercomputer-enhanced scientific instruments. (See Winter 1996 Parallel Computing Research, page 9.)
The I-WAY effort also demonstrated that while physical infrastructure to build such systems is becoming widespread, the heterogeneous and dynamic nature of high-performance networking environments poses new challenges for developers of system software, parallel tools, and applications. Like a distributed system, a networked supercomputer must integrate resources of widely varying capabilities connected by potentially unreliable networks and often located in different administrative domains. However, the need for high performance can require programming models and interfaces radically different from those used in distributed systems. As in parallel computing, networked supercomputing applications often need to schedule communications carefully to meet performance requirements. However, the heterogeneous and dynamic nature of networked computing limits the applicability of current parallel computing tools and techniques.
These considerations suggest that while networked supercomputing can build on distributed and parallel software technologies, it also requires significant advances in mechanisms, techniques, and tools. The Globus project is intended to accelerate these advances. In a first phase, the researchers are developing and deploying a networked supercomputing infrastructure toolkit that provides basic capabilities and interfaces in areas such as communications, information access, resource location, resource scheduling, authentication, and data access. Together, these toolkit components define an abstract machine on which a range of alternative infrastructures, servers, and applications can be constructed. The group is building parallel programming tools and resource discovery and scheduling services, and other groups are working in other areas.
The long-term goal of the Globus project is to address the problems of configuration and performance optimization in networked supercomputing environments. These are challenging issues because of the inherent complexity of networked systems, the fact that resources are often only identified at runtime, and the dynamic nature of resource characteristics. Successful applications must be able to configure themselves to fit the execution environment delivered by a metacomputing system, and then adapt their behavior to subsequent changes in resource characteristics. The group is investigating the design of higher-level services layered on the Globus toolkit that enable the construction of such adaptive applications. These services are collectively forming an Adaptive Wide Area Resource Environment (AWARE).
Globus development is based at Argonne National Laboratory and the University of Southern California Information Sciences Institute. The Aerospace Corporation, Trusted Information Systems, and Northern Illinois University also play significant roles. Globus research is supported by DARPA, DOE, and the NSF. The two awardees of the NSF's new Partnerships for Advanced Computational Infrastructure program, the National Computational Science Alliance (NCSA) and the National Partnership for Advanced Computational Infrastructure (NPACI), are expected to make significant use of Globus components. (See "CRPC Partners with National Supercomputing Competition Winners," page 1.)
In support of Globus research and development, the researchers are developing a large testbed called GUSTO. This testbed currently incorporates computers at four sites and will expand over the summer to include about 15 sites, some international. Basic connectivity is via the Internet, although some machines are also connected via OC3 (155 Mb/sec) ATM networks. Eventually, most GUSTO sites will be accessible over the NSF's OC3 vBNS network. Globus researchers are also making use of the CAIRN networking testbed for experiments with network quality-of- service issues.
For more information on the Globus project, see http://www.globus.org/.
Table of Contents