Special Report: Network Provisioning

1. Introduction

The large-scale U. S. Department of Energy (DOE) science projects of the next generation will increasingly depend on close collaborations of multi-disciplinary researchers dispersed across the country or around the globe. Such collaborations collectively represent capabilities unavailable at any single national laboratory or university. Furthermore, these projects span a wide spectrum of disciplines including high energy physics, climate computations, fusion energy, genomics, astrophysics, and others, which are of large interest to DOE. These collaborations invariably involve geographically distributed resources such as supercomputers and clusters that offer massive computational speeds, user facilities that offer unique experimental capabilities, repositories of experimental and computational data, and human experts with deep and broad knowledge in technical areas. Of particular importance are the new experimental facilities coming on-line such as the spallation neutron source (SNS), and the relativistic heavy ion collider (RHIC), which present unprecedented opportunities and challenges for distributed and collaborative remote experimentation and data analysis. The ability to remotely perform the experiments and then transfer the large measurement datasets can significantly enhance the productivity of scientists and facilities. In general, a seamless access to the distributed resources by the researcher teams is essential to carry out the DOE large-scale science missions: Indeed, the "network" has become a critical component of the modern infrastructure for large-scale science, much like the supercomputers or experimental facilities. The above networking capabilities add a whole new dimension to the access of these computers and user facilities, thereby eliminating the "single location, single time zone" bottlenecks that plague these valuable resources.

Advances in high-performance networks hold an unprecedented potential in realizing these network capabilities, thereby expanding the impact of a number of DOE large-science computations and experiments. Such networking opportunities together with the potential benefits to various science areas have been identified in the DOE network planning workshop that took place in August 2002 [1], and have been repeatedly highlighted in other DOE workshops and conferences [2,3]. In June 2003, a roadmap has been formulated for the DOE networks, which envisions a seamless, high-performance network infrastructure to facilitate collaborations among the researchers and their access to remote experimental and computational resources [2]. The next generation DOE large-scale science projects and programs have requirements that will drive extreme networking. Some of these requirements involve massive (Petabyte sized) data transfers across the country and around the world. In other cases they involve distributed collaborative visualization, remote computational steering, and remote instrument control. These requirements place different, possibly mutually exclusive, demands on the network. The network capabilities required to support this scale and range of networking activities surpass, by several orders of magnitude, the performances achieved by today's leading-edge high-bandwidth networks. In summary, a main conclusion of the workshop is that:

An ultra high-performance network with powerful and flexible provisioning and transport modalities is needed to meet the demands of the DOE large-scale science applications.

The field of ultra high-speed networking is currently at a critical crossroads with no clear evolutionary path to eliminate the performance gap that exists between the link speeds and application throughputs. While the optical technologies promise links at Terabps (Tbps) the corresponding provisioning and transport technologies needed to deliver this performance on- demand to the applications are severely lacking. The widely deployed Transmission Control Protocol (TCP) transport mechanisms do not scale to these unprecedented optical bandwidths in terms of application throughputs. While the commercial demand for faster backbone networks will continue to improve the link speeds based on optical networking technologies, the lack of such demand at the application-level will prevent the development of the required mechanisms including protocols and components. Consequently, with the advent of multiple Gigabps (Gbps) routers and switches, the end-to-end bottleneck has moved from the core network to host systems and end components, which are often outside the priorities of service providers. This workshop is a foundational step in identifying the critical networking technologies for DOE large-scale science projects and programs. To keep it manageable, this workshop concentrated only on two key areas

Previous    Next    Table of Content for report: Network Provisioning    Home

Network Provisioning

 

 

 

Photuris.com - Optical Data Networking