NSFNET Reaches a Billion Packets in First Year
By Ellen Hoffman
July 1989 marks the first anniversary of the re-engineered National Science Foundation Network backbone. Just one year ago Merit, IBM, MCI, and the NSFNET mid-level networks were scurrying to bring up the new thirteen node, cross-country T-1 backbone. Through the cooperative efforts of individuals at all these organizations, the new NSFNET became a reality on schedule and has been operational ever since.
Within a month of the NSFNET becoming operational, packet counts doubled those recorded for the previous six-node backbone. Almost 195 million packets passed across the backbone in the first full month of operation (August 1988). And less than one year later (June 1989), the backbone marked another first with a billion packet month for an increased over 500% in the first year!
Growth has also occurred in the number of local access networks which are connected to the backbone via the mid-levels. From an initial 170 local networks last year, the total has now reached over 600 campus, government, and research nets. In addition, international connections include links to Australia, Canada, Denmark, Finland, France, Iceland, the Netherlands, New Zealand, Mexico, Norway, and Sweden.
More recently, Merit completed a major reconfiguration establishing a new phase of the NSFNET backbone during early July 1989. The backbone has undergone a total redesign and expansion within the first year of operation. The cross-country circuits have increased from an original fourteen to nineteen, all running at T1 speeds from the NSSs.
Continued growth is assured with the National Science Board's June approval of Merit's expansion proposal which will authorize additional NSF funding for the backbone by approximately $6 million over the remaining years of the agreement and make possible new nodes in the coming year (see article, page 2).
"We are proud of the accomplishments of our first year of operation," said Eric Aupperle, President of Merit. "Our successes have only been possible because of the cooperative efforts of everyone involved. With the committed staffs and ongoing research programs of Merit, IBM, and MCI, we expect the coming years to be equally exciting."
A Retrospective: Getting Started
The growth of NSFNET is a sign of the importance of national networking for the research community as well as the dynamic nature of communications technology. Many advances have occurred since Merit first submitted a proposal in August 1987 responding to the National Science Foundation's solicitation request for re-engineering and management of the NSFNET backbone.
Merit's original proposal outlined a strategy for creating a hybrid circuit and packet switched national network with strong centralized network management. An ongoing research and development program, much of which is undertaken on a dedicated separate research test network, was also proposed for implementing advanced technologies to improve backbone performance.
As part of the strategy for implementing this proposal, Merit developed a partnership with IBM, MCI, and the State of Michigan to incorporate the best of the academic, corporate, and government institutions in the network's development. Merit's strategy not only made possible the implementation of the new backbone on schedule and within budget in July 1988, but also provided for a dynamic environment where new technologies will be tested and implemented over the five years of the grant period.
The National Science Foundation signed a cooperative agreement with Merit in November 1987 for the proposed project, initiating an eight month period of planning, development, and testing for the new backbone. This included developing the hardware and software platforms for the new packet switches, known as Nodal Switching Subsystems (NSS), in cooperation with IBM and installing and testing the 14 T1 circuits provided by MCI which linked the thirteen mid-level and supercomputer consortium networks.
At a news conference in Detroit announcing the award, Michigan's governor compared the planned network development to a super highway.
"[NSFNET] will create a fast, reliable communications 'highway' linking the nation's seven regional research computer networks and NSF's six supercomputer centers," said the Governor of Michigan, James J. Blanchard. "The NSFNET award will enable Merit to provide leadership in national networking and to enhance research and development productivity."
By spring of 1988, the Merit facilities on the University of Michigan campus were a flurry of activity during which a new network operations facility was constructed and the thirteen NSSs were assembled and tested before being shipped to the sites under the guidance of IBM. The mid-level networks also participated in this activity, with personnel attending meetings in Ann Arbor to learn about the new equipment they would be receiving and many taking the opportunity to help "build" their own NSS with Merit and IBM.
At each site, staff were required to implement power and cooling requirements and work out the details for telephone cabling running from their site's building demarcation points to the location of their new equipment. IBM and MCI local service personnel were regulars at the node sites, helping mid-level staff with the planning needed to install the NSSs.
By mid-June, all sites had equipment in place and tests were running to ensure the new on- site equipment performed as planned. Then came July 1988 and actual operational status. Over that first month of operation, mid-level traffic was phased in to make a smooth transition from the older 56 kbps backbone to the re-engineered T-1 technology. Daily operational traffic was actually moving across the 8,000 miles of new NSFNET circuits, routed through the newly installed NSSs.
"An undertaking of this scope would be a challenge under any circumstances, a true stretch of our collective capabilities," said Robert V. Mazza, director of development, IBM Technical Computing Systems, announcing the newly functioning network. "To have done it in eight months is extraordinary."
Changes and Improvements
This re-engineering is only a first step in NSFNET evolution, and the past year has seen many changes and improvements. Yet through all the changes, NSFNET has shown exemplary performance. While individual nodes have experienced short outages, the robustness of the deployed technology has prevented extended interruptions in service and enabled researchers to take advantage of the high reliability to accomplish their tasks.
The following is a list of some of the other achievements in the last year:
New Monitoring Technologies
Implementation of new monitoring technologies in the network, including Simple Network Management Protocol (SNMP), and NSF Net STATistics (NNStat).
These have been accompanied by the development of better network management tools for use by Merit's Network Operations Center. Among these are a Netview interface for alerts from MCI, IBM's Electronic Customer Communications Option (ECCO) system for placing service calls to IBM automatically from the NOC, IBM's Information Management product for trouble ticket generation and reporting, and IBM's X Windows-based Graphic Monitor (XGMON) system for monitoring the NSSs using Simple Gateway Monitoring Protocol (SGMP), SNMP, and Packet Internet Groper (PING) with feeds to Netview.
Deployment of improved Exterior Gateway Protocol (EGP) implementations for mid- level/backbone reachability information exchange, the growth of policy based routing, and experimentation with Border Gateway Protocol (BGP) for further improvements in the routing interface.
An advanced routing workshop was sponsored by Merit in Ann Arbor in March to consider these topics with another scheduled for later this summer.
Redesign of Network Topology
Redesign of the entire network topology which increases speed and capacity for the entire backbone. This includes increasing the number of circuits for a total of 19 covering over 13,000 circuit miles.
In particular, all nodes now have multiple T1 connections, a change from the earlier design in which six nodes were spurs with a single, non redundant connection. With this change also came the use of higher speed interfaces making possible T1 capability for each Packet Switching Processor (PSP) in the NSSs. All circuits now are linked to MCI's Digital Reconfiguration sites which will enable new circuit switching methods using automated reconfiguration at Merit's control in the future.
Growth of Research and Test Network
Growth of the Merit/IBM/MCI research network facility to test new technologies before deploying them on the operational NSFNET and to provide a platform for studying load and traffic patterns without disrupting researchers.
Development of Information Services
Development of Merit's NSFNET Information Services to improve communications among the networking community. This included implementing an 800 number for trouble reports and network consulting, establishing a monthly newsletter to keep interested subscribers informed of network developments, and developing an online query system for information on such areas as member networks, topology, outages, and site contacts. Two major seminars were held in Hilton Head, South Carolina and Denver, Colorado to help new campuses and research organizations improve their internetworking expertise.
Completion of NSF Review
Completion of a successful review by NSF of the project in May 1989, with an evaluation that noted Merit and its partners performance had been excellent, "based on outstanding examples of engineering implementation and resource management."
Networking into the future
In addition to these changes, Merit and its partners are actively looking at new technologies for the NSFNET. Experiments with applications such as packet video or high performance LAN interfaces using Fiber Distributed Data Interface (FDDI) technologies are ongoing and will bring further changes to the network. Plans for the next year include the development and testing of T3 technologies along with continued work exploring Open Systems Interconnect (OSI) support.
"In some ways, today we are at a stage similar to that in the 1890's when the voice telephone system began to be interconnected," said William A. Wulf, NSF assistant director, Computer and Information Science and Engineering (CISE) in July 1988 at the backbone's inauguration. "In 1890 it was impossible to predict the central role of the telephone in modern society."
"It is equally difficult to predict exactly what will emerge as researchers take advantage of increased access to more capable, less expensive computing and communication technology," Wulf continued. "However, I foresee the development of a single, integrated but geographically dispersed community of researchers, collaborating much as only those that are physically adjacent can do today. Such a 'network laboratory' will amplify the talents of our researchers enormously."
Taken from The Link Letter, 15 August 1989, Vol. 2 No. 3.