NSF/NCRI; George Strawn, Mark Luker
Introduction
In 1985, the NSF decided to support Internetworking for research and education. By 1995, the Internet had become a world-wide system of more than 100,000 networks; and NSF retired the NSFNET backbone network service, transferring the provision of commodity Internet services to the emerging Internet industry. Also in 1995, NSF authorized the creation of the vBNS, a next-generation network for research and education, with initial connections to the NSF-supported supercomputing centers and NAPs. In 1996, NSF updated its Connections to the Internet Program (NSF 96-64), including a high-performance connection option for research and education institutions with significant successful Internet experience who have new application requirements that can not be met on the current generation commercialized Internet.
This article will discuss the likely evolution of the vBNS in the presence of the high-performance connections program. First, the definition of and need for high-performance networks for research and education will be discussed. Then NSF's projected policies relating to vBNS usage and interoperation with other networks will be laid out. Finally, several possible strategies for containing the high current costs of high-performance Internetworking will be considered.
What is a High-performance Research Network?
As stated at the vBNS workshop held in Washington, D.C., in the summer of 1995, a high-performance network is more than a high-bandwidth network. Additional characteristics, such as the capability to reserve network resources in advance and to guarantee performance to match specific qualities of service are needed. With such services, the high-performance Internet will be able to support delay-sensitive, bandwidth-intensive applications such as distributed computing, real-time access and control of remote instrumentation, and efficient multicasting to support video and multimedia collaboration. Moreover, the most demanding new applications will require automated methods of scheduling and allocating major network resources, much as has been done with supercomputers over the years.
There is a need for a High Performance Research Network for R&E
In the early development of the Internet, the research and education community was able to deploy new protocols, network services, and applications on the NSFNET that expanded their capabilities for collaboration and accelerated the pace of research. These creative efforts rapidly spilled over to the commercial Internet as a key ingredient of its phenomenal growth. Now that the Internet is commercialized (and heavily loaded), most experimentation and introduction of new protocols and services must be done on separate testbed networks.
A high-performance research and education network is critically needed to support today's advanced applications for R&E and to provide a platform for the early deployment of tomorrowOs protocols, network services, and applications. Such a network would promote both continued leadership for U.S. research and education and the accelerated availability of new services and applications on the commercial Internet.
Expanding the Role of NSFOs vBNS with High Performance Connections
NSF's very high-performance backbone network service, the vBNS, is now operating between five supercomputer centers at 155 mbps (soon to be upgraded to 622 mbps). It is intended to support new applications that require emerging technologies to guarantee multiple qualities of service. NSFOs recently announced high-performance connections program is anticipated to provide some 100 additional R&E institutions with high-performance access to the vBNS over the next several years, defining a significant and growing Ohigh-performance networking communityO within R&E. New policies arising from the recent program review of NSFNET will shift the balance of the vBNS more toward a "leading-edge but stable" platform to support the advanced applications of R&E and away from a volatile network for temporary experiments in networking. Together these developments will make the vBNS an important high-performance network for the R&E community.
Access and Acceptable Use
The vBNS must support extensive development and experimentation with high-performance applications in R&E without undue competition with the commercial Internet. Its access and acceptable use policies can be described in terms of communications within the group of vBNS-authorized institutions (VAIs) who have received an NSF high-performance connection award, and between the VAIs and NSF-designated collaborating vBNS-partner institutions (VPIs) from other federal agencies, industry labs, foreign universities, etc.
Institutions with High Performance Connections to the vBNS must have commercial Internet connections as well
Since most of the traffic from a typical R&E campus is to sites outside the R&E community, each must maintain a robust commercial Internet connection in addition to its vBNS access. NSF's high-performance connections program is designed to help campuses upgrade their off-campus connection (and indirectly their internal networks) to efficiently support both the high-performance applications of the vBNS and the commodity networking needs of the campus at large. Key to this strategy is the use of emerging technology to share expensive high-bandwidth links wherever possible while guaranteeing appropriate levels of service for both types of traffic.
The vBNS will interconnect with other U.S. research networks
The vBNS will seek to interconnect with other U.S. research networks to support R&E projects of interest to vBNS-authorized institutions. Examples of such research networks include federal agency networks and other public and private sector networks organized to support open research.
The vBNS will interconnect with International research networks
The vBNS also will seek to interconnect with international research networks to support R&E projects of interest to vBNS-authorized institutions. Examples currently include R&E networks in Canada, Europe, Asia-Pacific, and consortia of such networks.
The vBNS will not offer transit services
The vBNS is an expensive, limited resource intended primarily to support high-performance applications for R&E and to drive the implementation of new protocols and network services that can better support R&E in the future. For this reason it cannot be used simply to transport commodity "transit" traffic between a VAI and a non-vBNS institution, or any transit traffic between two non-VAI institutions. Exceptions may be granted in appropriate cases if, for example, the other networks involved also support the experimental protocols and qualities of service associated with vBNS.
The vBNS will allow general traffic between vBNS-authorized institutions
Under a proposed policy change, vBNS-authorized institutions initially would be allowed to exchange general traffic on the vBNS with another VAI or a VPI. (Similarly, VPIs will be allowed to use the vBNS for general traffic to VAIs, but not for transit to another VPI.) This approach would greatly simplify policy and routing, would accelerate the migration of advanced technologies for quality of service into the commercial Internet, and would have relatively minor impact on overall traffic patterns.
"gigaPoP" strategies for controlling costs and policies for high performance networking
The cost to a campus of maintaining a robust commercial Internet connection in addition to a high-performance connection to the vBNS depends heavily on local circumstances, but may range as high as $500,000 per year or more. The high-performance connections program is designed to minimize the cost of the off-campus link though sharing this bandwidth. Interconnecting the vBNS with other U.S. and international research networks can greatly increase the access to these resources from vBNS institutions while reducing the costs of such connections. Beyond this, it will be important to consider how R&E institutions might take advantage of geography or other special circumstances to share costs.
As an example of such a strategy, institutions that are near to each other (measured in network interconnection costs) might share connections to the vBNS and commercial network providers at a common access point, here called a OgigaPoPO to suggest Ohigh-performanceO and Oconnection pointO. In some cases, these institutions might share a single last-mile connection from their institutions to a gigaPoP to carry the high-performance and the commodity traffic of each. To implement this strategy, the gigaPoP would need to have connections to both the vBNS and to their vendorOs commercial network. (It might have connections to other networks as well.) The gigaPoP connections to the vBNS and the commercial network would be shared by the participating high-performance connection institutions, thus reducing overall costs.
What is a gigaPoP? It could be implemented at an existing NAP, a metropolitan or regional network, an existing vBNS institution, or in an entirely new manner. One key feature is the ability to make high-performance connections (i.e., with the current state of quality of service to the vBNS) and the ability to make commodity connections to the commercial Internet. It may also be desirable to establish some form of management or collaboration between gigaPoPs to promote the coordinated rollout of new technologies for the high-performance connections.
The following examples illustrate several of the approaches that could be considered for determining the policies, procedures, and operations of gigaPoPs: First, a consortium of connecting institutions could determine its own policies and procedures and also arrange for the operation of a gigaPoP. Second, a gigaPoP oversight board could be formed to establish general policies and procedures for all gigaPoPs and each consortium could arrange for the operation of its own gigaPoP. Third, a gigaPoP organization could be formed to establish policies and procedures and to operate all gigaPoPs. For each of these approaches, the costs of the gigaPoPs could be recovered by fees paid by the connecting institutions. Such fees initially could be supported by high-performance connection awards from NSF. The need for special gigaPoPs should be reduced over time as high-performance capabilities become available in the commercial Internet.
Conclusions
This article has discussed the possible evolution of the vBNS from a nine-node network to one with possibly 100 nodes. This expansion would be accomplished by means of NSFOs recently-announced high-performance connections program. The need for high performance research networks to develop the next generation Internet services and applications was discussed, as was NSFOs projected policies relating to vBNS usage and interoperation with other networks. Finally, several strategies for containing the costs of high performance connections were outlined for consideration. NSF welcomes public discussion and advice regarding these and related matters.