Initial T3 Deployment in Place on the NSFNET

By Ken Horning

Operational deployment of NSFNET's new T3 backbone reached the halfway point in the final hours of 1990. T3 installations are now complete and ready for operational traffic at the backbone end nodes in Ann Arbor, MI, Urbana-Champaign, IL, San Diego, CA, and Palo Alto, CA.

Installations at the remaining four backbone end nodes, Pittsburgh, PA, Ithaca, NY, Cambridge, MA, and Argonne, IL will be completed during the coming weeks.

"This upgrade again reflects the National Science Foundation's (NSF) commitment to keep NSFNET the world's leading computer network for the support of research and education," said Dr. Stephen S. Wolff, Division Director, Division of Networking and Communications Research and Infrastructure, at the NSF. "New applications that were not feasible on slower networks will be possible with the availability of T3 bandwidth."

Production ready

T3 connections at the four completed backbone end nodes have undergone thorough pre-production testing during recent weeks. Testing procedures have included continued verification of hardware, software and circuits to evaluate reliability.

A suite of testing tools and procedures has also been created which will facilitate installations at the remaining T3 production backbone end node sites.

The model developed for high-speed backbone transmission involves a new generation of Nodal Switching Subsystem technology developed by IBM. Advanced circuit technology for the T3 upgrade is being provided by MCI.

Operational traffic is being routed at the initial four T3 end nodes and the installed T3 backbone is considered to be production ready with more production traffic being phased in at press time according to Eric M. Aupperle, President of Merit Network, Inc.

"Performance levels at these initial end node installations will be improving with planned technology enhancements," said Aupperle.

Future improvements

The architecture for the T3 network is utilizing a collection of IBM Core Nodal Switching Subsystems (C-NSS) within the MCI infrastructure, forming a cloud of co-located packet switching capability. Exterior Nodal Switching Subsystems (E-NSS) are located at client sites and connect into the C-NSS cloud.

With the deployment of the new T3 architecture, the node packet switching performance will improve significantly. The initial T3 deployment employs an Ethernet interface to the local area network, providing material performance improvement compared to T1 NSS performance.

As the NSFNET partnership completes the FDDI interoperability testing and deploys FDDI with the new technology, even more significant performance improvement will be realized.

Additional new technology due in '91

Later in 1991, the partnership plans to deploy new technology which will use intelligent subsystems for the extended interfaces. These subsystems or powerful RISC-based adapters utilize bus master and slave capabilities on high bandwidth implementations of the microchannel to achieve very high-speed card-to-card forwarding with no system intervention.

Coupled with optimized distributed protocol code, these systems can achieve very high throughput rates. IBM's RISC on RISC architecture utilizes RS/6000 RISC chipsets for the control processor and a 25 MHz superscalar, RISC embedded controller with on-chip cache and data RAM for the adapter engines. The new technology with on card packet forwarding will dramatically improve the performance on the T3 network.

"NSFNET is significantly expanding the network capability of our researchers today with this T3 implementation," said Aupperle.


Taken from The Link Letter, Vol. 3 No. 5, December 1990.