NSF should recognize the unique and critical contributions made by various research communities active in creating, using, and occasionally abusing advanced networks. NSF has a responsibility for nurturing and supporting the alternative environments necessary for these groups to work independently and jointly toward common goals of advancing the art of networking.

Introduction

In 1995, the National Science Foundation (NSF) transitioned the NSFnet Backbone Service to the private sector, officially ending NSF's stewardship of "the Internet". NSF also initiated the NSFnet Program's collaboration with MCI to provide the very-high-performance Backbone Network Service (vBNS). Now, as we near the new millenium, NSF's Advance Networking Infrastructure and Research division (ANIR) is seeking inputs from the research and education community concerning the priorities for NSF's future support of networking infrastructures for research.

On March 1-2, 1999, 32 individuals from the computing research community met in La Jolla, California for a workshop on " Future Scenarios - NSF Networking Research and Associated Infrastructure Support ". The Computing Research Association (CRA) sponsored the workshop with Dr. Mary Jane Irwin, Pennsylvania State University, and Dr. Sidney Karin, University of California's San Diego Supercomputer Center, chairing the meeting.  

Throughout the two days various subgroups presented their concerns about current Federal networking initiatives and alternative opinions about what "infrastructures" the NSF has a role to support. By the second day, participants settled on three distinct types of networking infrastructures that are characterized by what one attendee called the network creators, users, and abusers. Together, these three groups constitute the "research community" that drives the evolution of our advanced networks. Creators develop and test new protocols and technologies for driving the Net's advancement; users require ever more advanced, yet stable infrastructures to support research in various scientific disciplines; and abusers continually push the envelope on available technology and the capacity of the networks to support their advanced applications and technologies.

Participants urged NSF to assume the leadership role defined for it in the President's Information Technology Advisory Committee (PITAC). The PITAC report writes that…

An important means of following-up the PITAC recommendations would be for NSF/ANIR to take proactive steps to stimulate new programs targeting the needs of the various communities dependent on networking infrastructures. In defining specific alternatives, workshop participants identified three critical goals that should guide any future efforts by NSF with respect to networking infrastructures. NSF/ANIR should use its programmatic influence to

• effect improvements in the last mile and in campus infrastructure through financial, policy, and other means

• promote multi-disciplinary, cross-disciplinary research relating to networking and use of advanced networks

• challenge the community to come up with creative solutions to help solve the growing people-power shortage in networking and, indeed, in all of the Information Technology (IT) sector

Workshop participants suggested that the NSF adopt a programmatic thrust targeted at supporting three inter-related but distinct research infrastructures:

1. Network Research Infrastructure - an infrastructure for pure network research, including small focused testbeds and possibly large, flexible research-oriented networks.
2. Community Infrastructure - an infrastructure characterized by the non-hardware elements that promote advancements in networking technology and use, including middleware and education.
3. Advanced Production (Service) Infrastructure - an infrastructure serving the needs of the research and education community -- in particular, an advanced network designed to serve the advanced networking requirements of the various disciplines supported by NSF Directorates.

The sections that follow highlight some specific actions associated with these three infrastructures that NSF may consider as it implements its new programs.

The Three Communities

One of the major discussion topics at the meeting was the distinct needs of the three different communities: the creators, users, and abusers. The communities are very different and one of the challenges the meeting saw for ANIR going forward was determining when the different communities' needs overlapped and when their needs were in conflict and required separate infrastructures to support.

Creators seek to develop and test new protocols and technologies that can serve to drive the Net's advancement. The needs of individual creators or teams of creators vary wildly, depending on the type of research being done. However, one broad theme is that the process of creating a new technology inherently implies that the infrastructure is fragile. A critical part of creating something new is making mistakes and analyzing and learning from those mistakes. So the optimum infrastructure for creators is usually one that they can break, and once broken, keep off-line for necessary analysis.

Users, in contrast, seek a stable infrastructure that supports their, typically non-networking, research. Researchers in a wide range of scientific fields have come to depend on the Net to do their research. At the same time, these users, being leading edge researchers in their own fields, often benefit from the presence of leading edge networking features in their networks. Participants heard from several self described users about their reliance on prototype networks such as the MBONE.

Abusers, as some discussants jokingly described themselves, are extreme users. Their research so dramatically benefits from advanced networking features that they seek an infrastructure that contains the latest and greatest technologies, even if they are not quite stable. Abusers will accept periodic network failures (swiftly repaired) if that is the price of getting the advanced features they need.

The three types of communities define a spectrum. Users seek a stable infrastructure with as many advanced features as are consistent with good network service. Abusers seek advanced features and will accept occasional loss of network service as the price of being advanced. Creators seek to develop advanced features and routinely will shut the network down for extended periods to assist in development.

National demonstration efforts such as Information Highway (IWAY) demos during Supercomputing '95 both demonstrated opportunities associated with high performance networking and the estrangement between the different groups. The desirability of closing the gulf separating these communities has been articulated in IWAY debriefings and in numerous workshops and community discussions over the past years. Better integration of these and other related communities is viewed by participants as an important challenge for NSF and other Federal agencies.

Ensuring Infrastructure(s) for Network Research

Members of the Network Research Community are the creators in this world.

Historically, the Defense Advanced Research Project Agency (DARPA) has had a reputation for supporting long-term, experimental research endeavors with demonstratable prototype systems (e.g., quantum computing). NSF, on the other hand, was viewed as supportive of more theoretical, nearer-term research. The distinctions between these funding agencies have blurred recently, with DARPA emphasizing mission-critical research endeavors and NSF putting more focus on experimental systems. NSF also acknowledges a role in promoting an advanced networking infrastructure as support for users in the research and education community.

The requirements of the Network Research community include:

• a focus on scalability - The single most important issue facing researchers and networks today is scalability. We now need to think in terms of the "billions of nodes" promised by emerging network appliances, sensors, and other Internet technologies. All aspects of network research, future architectures, and hardware, software, and "peopleware" must be fundamentally rethought in light of the potential uses of the Internet for communication, business, and scientific research. A major challenge is that we cannot build a billion node network to experiment on - rather we need to find creative ways to use infrastructure and simulation to study pieces of the larger problem.
• open and flexible infrastructures - Because the Network Research is about change and creation, any limitations on the ability of researchers to change their infrastructure or share the results of their work typically inhibit or completely prohibit use of infrastructure for research. Research infrastructures need to be open: open hardware, open software, and open results. And, of course, it must be acceptable to completely disable the infrastructure in support of experiments.
• flexible programs - Past experience with the very-high-performance Backbone Network Service (vBNS) illustrates that researchers (and NSF) rarely appreciate the full range of questions needing exploration when the network infrastructure is being formed. It is therefore critical that the agreements and technology being employed in research infrastructures be as conducive to meeting these evolving needs as possible, including supporting incorporation of new technologies and network media, as well as supporting partitioning into "virtual" networks to support alternative test environments.
• alternative networks / open networks - The pace of technological advancements is increasing, as are requirements for testing new protocols and technologies over these media (both individually and in combination). Of greatest importance are research infrastructures incorporating optical fiber, wireless, and satellite infrastructures.
• research on testbed technology - Historically, network researchers were able to set up simulation facilities in laboratories, then incorporate the results into standard models of network behavior. With the complexity and scaling issues inherent in today's multi-provider commercial Internet, traditional testbed approaches and simulation software supporting networking research are increasingly obsolete, with few new alternatives on the horizon. New simulation tools and testbed technologies reflecting real-world traffic conditions are desperately needed by the network research community.

Building a Community Infrastructure

Since the NSFnet was decommissioned, the U.S. research community (the users) has struggled to define its networking infrastructure needs.

NSF can encourage communities in two ways. The first is simply through revitalizing the provision of working infrastructure. NSF needs to take steps to make sure that any NSF-supported infrastructure designed for users has outstanding access and cutting edge features. In particular NSF should:

• Ensure services are effectively delivered to all appropriate users in the campus - Participants heard multiple stories of infrastructures connected to campuses that lacked plans to deliver the infrastructure's services to the desktop of the researchers who needed the services. To promote the usefulness of its programs, NSF should resume the practice of requiring campuses, as a condition of funding for connection to new network services, to build out the infrastructure necessary to deliver those services to relevant users.
• Assist in the creation of collaboratories and virtual environments - These new environments for research hold significant promise for altering the manner in which researchers interact with one another. Key points made by participants include the fact that a small amount of financial support in this field may have significant payback for the disciplines using these techniques. While there is considerable commercial activity underway currently, these efforts may not directly meet the needs of researchers due to their high costs and limited focus. Open APIs and protocols are fundamental to advancements in this field, middleware is needed, emphasizing transparency and usability.
• Promote communities - An important extension of the collaboratory and virtual environments thrusts includes demand for conferencing and whiteboard tools; MBONE and multicast tools; shared file systems and code control software. Many of these tools are nearing maturity. Researchers report considerable benefits from using the tools when available. Yet they are available only sporadically.

NSF can also encourage communities by supporting research that benefits multiple communities and providing support in ways that encourage cross-community interactions. Examples of such research include:

• Research and demonstration projects focused on scaling issues - Scaling in this instance refers to the sheer growth of the Internet and the ubiquity of basic Internet services. The scaling affects the uses of the network itself and the applications that users develop for the network.
• Research support for overlays - Participants observed that one way for creators and users to share networking resource is to build infrastructures that allow virtual overlay networks (i.e., building a more complex network on top of a basic network). Participants cited several compelling research endeavors that could enable or use overlays, ranging from micronet testbeds, to switchlets and other enablers of virtual research networks, to computational grids involving supercomputing and other distributed computational resources. Tools facilitating access, scheduling, and use of research resources over the Internet are an important priority for NSF support.
• Passive and active instrumentation and measurement - Measurement should be both end-to-end and integrated within backbones, with a tight coupling between the entities responsible for the instrumentation and measurements, and the researchers, end-users and networks that might utilize the data. The goal should be to provide large amounts of freely available data to a wide range of researchers. In addition, there is a pressing need for complete and non-anonymized packet traces in several fields and NSF, as the owner of research testbeds, should be collecting such data on its testbeds and providing it to qualified researchers. Engagement of Internet Service Providers (ISPs) and vendors in the design, conduct and analysis of the measurements is critical to this process and to eventual adoption of these technologies by the private sector.
• Advanced simulation environments - Simulation techniques are evolving, consistent with the evolution of the Internet. They must be flexible, multi-level, and distributed with a tight coupling with the real world. Incorporation of real measurement data and involvement of ISPs and vendors in these processes is important to their relevance and success.
• Rethinking technology transfer - Traditional models of technology transfer evolving from pure research, to demonstration and testing, to commercialization are less relevant in a dynamic information technology environment, where the research sector is finding it increasingly difficult to keep up with new commercial deployments. New means of partnering and cooperation among academia, government, and the private sector are critical for the advancement of the networking sector and for keeping the research community focused on issues relevant to an increasingly complex commercial environment. Similarly, partnerships involving the campus chief information officer (CIO), applications users, and networking researchers can also have spinoff benefits throughout the campus environment and related communities.

Advanced Production (Service) Infrastructure

ANIR provides a critical service for the U.S. research community through its support of an advanced production (service-oriented) infrastructure. Alternative advanced research and education (R&E) networks, such as the vBNS and Internet2's Abilene network, offer backbone services to the R&E community. Other commercial networks have also announced their intention to seek certification as an NSF-approved high performance connections network. Competition among these networks helps to drive the introduction of advanced backbone services. However, this pressure is insufficient to keep pace with the growing needs of the research community for more advanced services and for attention to the critical problems inherent in the last mile connections.

Support of Experimental Applications: NSF needs to provide support for (1) interdisciplinary efforts involving networking technologies and (2) transitioning projects from preliminary demonstration projects to large-scale experiments. In particular, participants urge NSF to consider supporting efforts involving:

• Distributed collaboration & visualization
• Distributed data mining
• Distance learning
• Digital libraries
• Remote instrumentation

Funding of Required Services: An array of advanced services are necessary to support and extend the capabilities of advanced networks in order to make them readily usable by researchers. These services include:

• Local infrastructure - Just as users need services to their desktop, so too do advanced users. NSF must ensure that the hardware, software and general support are available from the border router to the researchers' desktops; in some cases, this program may require off-campus support of remote researchers and facilities.
• Global resource/knowledge discovery (finding aids) - Assist in navigating and analyzing the massive volumes of data present throughout the web.
• Caching services - Provide national, regional and local temporary storage of text, audio, visual, and other data; as appropriate, these services may include intelligent retrieval and prefetching of select datasets, particularly those relating to large scientific datasets.
• Multicast services - Provide a fundamental service supporting interactive communications over the Net in a sound, infrastructure-friendly manner.
• Security, privacy and authentication services - Recognize the need for balancing the interests and concerns at the local level with requirements for interoperability across the R&E infrastructure.
• Instrumentation and monitoring - Support monitoring at the level of the local and backbone network and the application - with appropriate feedback loops for the various groups, e.g., user, applications developer, campus NOC, ISP/backbone NOC, as well as networking researchers.
• Bounded delay (up to real-time) transport services - Ensure that users and applications developers are capable of adjusting their requirements to coincide with performance parameters -- if expectations have been appropriately communicated.

Specific Research/Funding Needs: Technologies specifically identified as worthy of consideration for NSF support include:

• Software aimed at
  - promoting ease-of-use of the Internet for researchers, e.g., making the infrastructure transparent for users;
  - providing toolkits and services to enable use of experimental applications
  
• Consulting and operational support for services such as the ones described in this report;
• Research to identify and develop future required services to make the Internet more directly relevant and useful for researchers in various disciplines.

Conclusion

The participants of this workshop feel that this meeting provided an important opportunity to discuss their concerns with the NSF Program Officials and with other individuals from a cross-section of the computing research community. As part of NSF's community-building activities, similar workshops should be held quarterly to follow up on the topics described in this report and to identify emerging research priorities and evolving community needs.

Future Scenarios - NSF Networking Research and Associated Infrastructure Support, was held at the Sea Lodge Hotel, La Jolla, CA on March 1-2, 1999. It was hosted by the Computer Research Association (CRA) in collaboration with the University of California's San Diego Supercomputer Center (SDSC). Dr. Mary Jane Irwin, Pennsylvania State University, and Dr. Sidney Karin, UCSD/SDSC, co-chaired the meeting, with funding provided by NSF's ANIR division.

Thirty-two (32) individuals from 21 institutions participated in the discussions (identified below). Several attendees contributed survey answers in advanced of the workshop; these submissions are available at http://www.npaci.ed u/post-vBNS/CRA/scenarios.html.

For comments or questions, contact tmonk@caida.org