Princeton University seeks support under the high performance connections portion of NCRI's "Connections to the Internet announcement, NSF 96-64." The award would provide partial support for two years for an OC-3c connection.
For more than a decade, Princeton University has been committed to supporting meritorious researchers and educators with appropriate high performance networking infrastructure in support of world class research and instruction.
The proposal details many of the research and educational activities that are constrained by latency, limited bandwidth, and the lack of prioritization in the current Internet. Meritorious applications include Cosmological simulations, Remote Telescope Operations, the Sloan Digital Sky Survey, Radio Observations of Pulsars, an experimentally designed video signal processor architecture, coding images with statistics, research conducted by the Elementary Particles Group, research in networked sound synthesis control, and a multi-university live video course in Human Computer Interface Technology. These projects involve high performance applications such as large, distributed simulations, remote observation, visualization, and imaging, collaborative hypermedia, distributed computation and image processing, remote experimental control, collaborative peer review and classrooms, work with supercomputing facilities on numerical and visualization problems, and remote access to image and video archives.
The University will assist these researchers to establish connections to the vBNS with sufficient Quality of Service to support their research and instructional activities.
Princeton University is a private, coeducational university located in Princeton, New Jersey. Princeton's 600-acre main campus in Princeton Borough and Princeton Township consists of more than 5.5 million square feet of space in 140 buildings. Nearly all undergraduates live on the main campus. The University's James Forrestal Campus in Plainsboro consists of one million square feet of space in four complexes set on 340 acres.
Founded in 1746 as the College of New Jersey, Princeton has an undergraduate population of approximately 4500 students working towards the Bachelor of Arts (A.B.) or Bachelor of Science in Engineering (B.S.E.) degree. Fully coeducational since 1969, the University is home to students from all 50 states and from more than 50 foreign countries. In the last academic year, 1,810 students enrolled in the Graduate School. A single full time faculty of 750 teaches both undergraduate and graduate students.
At Princeton and throughout higher education, excellence in education and research has come to depend not only on a great faculty and a superior library, but also upon access to computing and information technology. Hardware, software, and networking have become indispensable aids for faculty, students, and staff, who look to the University to provide a computing environment capable of supporting world-class instruction and research. Computing and Information Technology (CIT) at Princeton has been responding to these needs by developing a world-class telecommunications infrastructure; by developing computer laboratories for undergraduates and graduate students; by identifying, obtaining, and then supporting appropriate academic and administrative applications; by providing a wider variety of educational and research services; and by pursuing a variety of technologies and pilot projects that are expected to enhance the University's computing environment and use of information technology.
Over the course of the past decade, the University has built a campus-wide network to connect all central and distributed computing facilities as well as to provide access to the Internet. Dormitory connections permit all students who own computers to reach campus and network resources at Ethernet speed from their on-campus rooms. Large scale computing is provided by three Sun Ultra Enterprise 2/2170s, a Sun SparcServer 20/71, and three Silicon Graphics Origin200 systems as well as an IBM 9672 running VM/CMS and MVS.
Princeton is currently building the required infrastructure to support high-speed networking on its campus. New cabling will permit high-speed connectivity to user machines. By integrating this infrastructure into the life of the campus, the University will promote the growth of new and expanded applications.
Princeton University is committed to the creation of a high performance local network infrastructure and is responsible for funding the local infrastructure without NSF/NCRI assistance. The high performance local network infrastructure will be made available as appropriate to all qualified users and will not be restricted to the applications described in this proposal that require the high performance network.
Princeton University and its computing support staff continue to play an important role in the development of the national telecommunications infrastructure. The University was among the earliest network users and was a founding member and strong supporter of the John von Neumann National Supercomputer Center and JvNCnet, the first T-1 based regional network. Upon the closing of the Supercomputer Center in 1990, the University provided a home and administrative assistance for JvNCnet for two years. The PI was the co-founder of BITNET and now serves as President of CREN. Upon coming to Princeton from CUNY, he led the BITNET II project that moved BITNET to TCP/IP. The Manager of the Princeton University Network Group, Peter Olenick, authored VMNET which permitted BITNET sites to connect to the Internet. The University is a founding member of the Internet2 Project. Over the course of the past year, the University has collaborated with Rutgers University, the University of Delaware, the University of Pennsylvania, and Lehigh University in the planning for the deployment of the distributed, Mid-Atlantic GigaPop. In addition, the PI is Chief Scientist for JSTOR, an initiative of the Andrew W. Mellon Foundation that now provides distributed access to an image archive of scholarly literature. He served on the board of Trustees for the Internet Society from 1992-4.
The connection of the University to the vBNS will significantly benefit numerous campus researchers. The following are examples of ongoing research that would benefit directly from the connection.
Cosmological Simulations (Jeremiah P. Ostriker, Young Professor of Astronomy, Provost)
The Princeton University sub-group of the Grand Challenge Cosmology Consortium (GC3) has developed a scheme for making out-of-core, very large cosmological hydrodynamic simulations. [NSF grant AST-9318185, "The Formation of Galaxies and Large-Scale Structure," PI: J.P. Ostriker; NASA grant NAG5-2759, "The Universe at Moderate Redshift" PI: J.P. Ostriker] The basic ingredient in this scheme is to have fast I/O speed (between RAM and disk), approximately 10 times faster than usual disk I/O speed. This method prevents us from being limited by the (usually relatively small) size of expensive RAM so that we can now make very large simulations. A very interesting and useful application of this scheme is to integrate different computers and disks over the network to utilize different facilities at different locations, thus to increase the overall computing power. In order to make this possible, a high speed network connection is needed, and vBNS fits ideally here. Our primary computing partner for this application is the National Center for Supercomputing Application (NCSA) at Urbana-Champaign, Illinois.
Remote Telescope Operations (Edwin L.Turner, PhD., Professor of Astrophysical Sciences)
The Department of Astrophysical Sciences, along with sister departments at five other US institutions (Johns Hopkins University, New Mexico State University, University of Chicago, University of Washington and Washington State University), operates a modern 3.5-meter optical/infrared telescope located at Apache Point Observatory in New Mexico's Sacramento Mountains, near Alamogordo. This important research facility began routine operation in late 1994 and was designed from the beginning to utilize remote observing and fast instrument change technologies. Remote observations are carried out via Internet links to the site by astronomers located at their home institutions scattered around the country. The remote computer interfaces allow the users to examine sky and weather conditions at the Observatory, operate the telescope and its several modern instruments, take scientific data and transfer it back to local computers for evaluation and reduction. In effect, the tasks which used to require astronomers to travel to distant and isolated telescope sites can be accomplished from local university offices and laboratories. The data which need examination and evaluation are typically very large images (with sizes characteristically in the range of 1 to 100 Mbytes, and increasing as instrument technology improves) that can be generated at high rates by the telescope. It is, of course, essential that these tasks be carried out on a real-time and interactive basis due to the constantly changing sky and weather conditions (that are often the limiting factor determining data quality) and due to the need to examine data quality and modify observing procedures immediately in order to maximize the effective use of astronomy's most precious resource, telescope time.
Remote observing, combined with the APO 3.5-meter's capability for fast (20 minute) instrumentation changes, has important scientific consequences far beyond convenience and low cost. In particular, they allow modes of telescope operation which are simply not possible with conventional telescopes and thus enable unique science. One example is "shared night" observing in which observing time is not assigned to astronomers in blocks of one or more nights, as at conventional telescopes, but rather in only the parts of the night during which target objects are particularly well placed in the sky for observations. This alone is believed to increase the scientific effectiveness of the telescope per unit time by a factor of two or more. Another example is that monitoring programs which require a small amount of observing time per night on a frequent basis, perhaps several nights per week, with a consistent instrument available can be easily carried out, although they are effectively impossible with a conventionally operated telescope.
At present, remote observations with the APO 3.5-meter are carried out via a dedicated T1 link from the Observatory to the New Mexico State University campus in Las Cruces and then beyond via the commercial Internet. This arrangement is only marginally adequate at present in the sense that the rate at which data is acquired is often limited by available bandwidth of the connection. It is clear that it will be entirely inadequate in the near future due to the introduction of new, much higher data rate instrumentation at the Observatory. Improvements in astronomical electronic detectors and other features of modern instruments will increase data rates by factors of ten to one hundred within the next few years. Unless a new source of network bandwidth, matching the requirements of these new instruments, is found, remote observations will become practically impossible and much of the Observatory's unique scientific capabilities will be lost.
The NSF vBNS network offers an elegant solution to this problem. Given Princeton's share of the 3.5-meter's time (15.6%) and a normal rate of losses to bad weather at the Observatory, high bandwidth connectivity between the Princeton campus and the site will be required for only about 400 hours per year (at night, naturally), so it is definitely not a constant demand. However, when the link is in use, high bandwidth will directly result in increased scientific productivity and better data quality.
Sloan Digital Sky Survey Science
The Sloan Digital Sky Survey (SDSS) [AP1] is a collaboration of a number of institutions, in which Princeton University plays a leading role. It will use a dedicated 2.5 meter optical telescope at Apache Point Observatory to make a digital map of 1/4 of the celestial sphere in five photometric bands, followed by a spectroscopic survey of the 1,000,000 brightest galaxies and the 100,000 brightest quasars. The construction phase of the project is nearing completion, and we expect the first imaging data in roughly six months from this writing. We foresee three main uses for vBNS connectivity in this project.
1) Access to the SDSS data stored at FNAL.
1a. The raw imaging data. When the telescope is operational, it will be taking data at 4.6 Mb/s. This data is stored at Fermilab National Accelerator Laboratory (FNAL), one of the partner institutions in this project, but we are developing the image processing codes here in Princeton. The ability to work with this data over the net will greatly ease and speed up the debugging phases of the project. The image processing code is described by Lupton. [AP2]
1b. The catalogues of reduced data. The databases containing the SDSS catalogues will be stored at FNAL; by the end of the project they will contain information on a few hundred million objects. We expect heavy use of these databases by Princeton scientists, as we start to exploit the rich potential of the SDSS data.
2) Access to our telescopes at Apache Point, NM. The system on the mountain is extremely complex; the main imaging camera, built in Princeton, contains 54 CCD detectors, and will need to be carefully monitored and checked for scientific consistency. The primary responsibility for this lies with personnel at Apache Point, but we foresee many times when we will want to access the imaging data as it comes off the camera to diagnose problems. As noted above, the data rate is 4.6 Mb/s; of which we would like to look at some fraction. This is not possible using our current Internet connectivity.
3) Access to our colleagues at JHU, who are responsible for the science database described in 1b above. We expect to collaborate with them in the refinement of the access methods, and the graphical display of the resulting data.
These uses of vBNS are based on two pieces of information:
Multimedia Research (Bede Liu, Professor of Electrical Engineering, Chair, Department of Electrical Engineering)
The availability of vBNS will greatly enhance several major research program in the Department of Electrical Engineering.
The State of New Jersey has recently funded a multi-year Center for Multimedia Research at Princeton and NJIT. Research to be conducted in this multi-year program has two major components
Research to be conducted under Multimedia Technology [NSF grant # MIP-9408462 "An Experimentally-Designed Video Signal Processor Architecture," PI's: A. Wolfe, B. Liu, W. Wolf PIs,"] includes: video library [EE1-EE3], algorithm and implementation of image and video compression [EE4-EE7], testbed for investigation of perceptual quality, knowledge-based medical imaging [EE-EE11], and video communication [EE12-EE15]. Under Collaborative Hypermedia, research will be conducted on asynchronous distributed group support system, collaborative peer review system, virtual classroom, automating hypermedia for decision support systems, and multimedia supported collaboration. The department also has a DARPA-funded project on Web-based video libraries, which has produced the first video library on the World Wide Web.
The emphasis of both the Center for Multimedia Research in general and video library research in particular is Internet access to video material. Video transfers require a great deal of bandwidth---the largest video in our collection is currently 1.4 GB. [NSF grant # MIP-9616716, "Coding Images with Statistics: An Exploratory Study"] Some transfers will require guaranteed bandwidth for streaming. Even transfers which are non-streaming require fast transfer of data which, given the large files involved, requires high bandwidth. We are already conducting initial experiments with remote sites for transferring files and we expect a growing number of experiments which require the transferal of large amounts of video data, both non-streaming and streaming, over the Internet; examples include video library interoperability experiments with Carnegie Mellon University. These experiments will be greatly facilitated by access to vBNS.
Radio Observations of Pulsars (Joseph H. Taylor, James S. McDonnell Distinguished University Professs-or of Physics, Dean of the Faculty)
The Princeton University Physics Department is involved in a continuing research effort requiring radio astronomical observations of pulsars with very high time resolution [NSF Grant: "Radio Observations of Pulsars", PI: Joseph H. Taylor, Number: AST-91-15103]. The results are used for studies in gravitational physics, cosmology, stellar evolution, fundamental astrometry, and time-keeping metrology. Data acquisition takes place at the National Astronomy and Ionosphere Center, an NSF funded facility located in Arecibo, Puerto Rico and operated by Cornell University. Starting in mid-1997, a new data acquisition system will be used for the project. It records digitized data at a rate of 10 MB/s and is likely to be used for as much as 50 hours per week. The first stage of data analysis is carried out with special-purpose computing equipment located on site, reducing the volume of data from terabytes to mere gigabytes. Nevertheless large numbers of gigabytes still remain, and they need to be moved quickly to Princeton for further analysis, combination with previously archived data, etc. The Cornell Theory Center is working on providing a high-speed link to the Arecibo Observatory. With a fast connection into the Princeton Physics Department, it will become possible to provide quick feedback of results to Arecibo, thus optimizing the telescope observing schedule and making the best possible use of all resources.
The Elementary Particles (EP) Group (Daniel R. Marlow, Professor of Physics)
The Elementary Particles (EP) Group of the Physics Department carries out experiments in laboratories both in the United States and abroad. It comprises ten faculty, seven research staff, a technical staff of ten, as well as several graduate and undergraduate students. At present, the group is active in experiments at CERN (Geneva, Switzerland), DESY (Hamburg, Germany), Brookhaven National Laboratory (Upton NY), the Stanford Linear Accelerator Center (Palo Alto, CA) and the Japanese National Laboratory for High Energy Physics (Tsukuba City, Japan). Several members of the group are permanently stationed at the aforementioned laboratories.
Computer networks have long played an essential role in the group's research activities. The group makes extensive use of the Department of Energy's "Energy Sciences Network" (ESnet). This network is used for remote access to laboratory-based computers, transfer of raw and reduced data, exchange of code with collaborators, and general communication via e-mail and the World Wide Web. The EP Group's present connection to ESnet is marginally adequate. In view of the historical trend toward increased usage, which shows no signs of abating, it is clear that improvement is needed. A connection between the proposed vBNS and ESnet would serve that purpose (provided, of course, Princeton is connected).
Research in Networked Sound Synthesis and Control, and a Multi-University Course in Human Computer Interface Technology (Perry R. Cook, Assistant Professor, Computer Science)
We have constructed audio synthesis server software which runs on multiple platforms, including various Unix and Windows-based systems. The software allows for scaleable sound quality and processor load, and supports a variety of control sources including MIDI, scorefile scripts, and a number of Graphical User Interfaces (GUI's) such as TCL/Tk and JAVA. This allows for high quality sound synthesis on a remote computer with low network control bandwidth. For truly interactive control of sound, however, control bandwidth must be guaranteed, and of guaranteed low latency. Further, transmission of actual audio sample streams is required for a number of applications. A vBNS connection would allow a large number of control channels to be combined with a few channels of CD-quality audio samples, allowing research to be conducted in areas such as auditory display of algorithms and network traffic, and remote artistic collaborations. Princeton CS has experience and ongoing research in all of these areas, and has recently collaborated on networked audio issues with other institutions such as Stanford University, U.C. Berkeley, and Columbia University.
A vBNS connection would provide for improved course quality and research in pedagogy related to our new course, "Human Computer Interface Technology, a Multiple University Course Taught via Remote Videoconferenced Network." The original offering of this course was in the Fall Semester of 1996, with 38 students enrolled and 3 auditors at Princeton. There were 14 students taking the course at San Jose State University, and 12 students taking the course at Stanford University. The course planning and development was awarded NSF grant #9527459 to San Jose State University in 1995. The course utilizes collaborative and remote teaching via Networked Video Conferencing, EMail, and the World Wide Web. Expert researchers in each sub-area present interactive lectures from locations around the United States. All lecturers contribute laboratory exercises, exam questions, and homework questions, and advise students on projects.
The course introduces basic technologies available today for enhancing computer input and output, specifically in real time to control sound and graphics. A secondary course topic, which is the primary focus of one of the other participating schools, is the use of novel input and output devices to provide computer access to handicapped users. Students first look at standard I/O devices such as the mouse, keyboard, and video monitors, then move on to sound, haptic (touch) display, and various position detection systems. Labs include work in pattern and gesture recognition, 3D graphics rendering, and sound and voice input and output. The students end the semester by executing projects in small groups.
So far, the course has been offered using ISDN (324kBaud) video conferencing equipment granted to Princeton CS through the AT&T/Lucent program for Special Purpose Grants in Science in Engineering, with matching from the Gordon Wu (Princeton Internal Gift Matching) program. ISDN connect time and multicast bridge time were paid from departmental funds. The course has received support for lab computers (Pentium Computers) from the Hewlett Packard University Instructional Philanthropy Program. The course has received support for laboratory test and fabrication equipment from the Princeton 250th Fund for Innovation in Undergraduate Education.
In offering the course using ISDN, it became clear that the quality of the compressed video and audio was too low for adequate presentation of the central topics in many modern engineering courses. Lectures on audio synthesis and compression, and on high quality graphics rendering and display, were severely compromised by video conferencing compression algorithms which are targeted to faces, simple graphics, and speech. Also, in order to make the video classroom experience as rich as a live seminar or lecture, and to avoid the impression to students that they are just "watching TV," more active interaction must be supported by the technology and planned into the sessions. The switching time of the bridge which arbitrates between all sites is too slow, the delays introduced by the compression algorithms and transmission network are too long, and these delays impose a social penalty for interrupting a lecture to ask questions.
The solution to these problems is a full bandwidth media "multi-bus" which connects all sites at ATM rates. Cross-country ATM connections are not yet available to all sites, and the vBNS connection will allow us to offer the course now, and exploit the pedagogical experience later when cheaper connections are available. The Princeton Computer Science department already has high speed fiber to all offices and classrooms.
Interactive Multimedia World Wide Web Czech Language Course (Charles E. Townsend, Chair)
The Slavic Department at Princeton University is developing an interactive multimedia World Wide Web Czech language and culture course. The course will contain nearly 100 hours of custom audio recordings, likely the largest Internet-based language course. The initial plans to include video have remained undeveloped, owing in large part to the inability of existing networks to deliver an acceptable product. The availability of the vBNS connection would permit the project to overcome this limitation, thereby greatly enriching the language learning environment.
Princeton's WWW Czech course, like any foreign language course, requires that the audio be of the highest possible quality. This does not mean simply that the sound must be delivered without the all too frequent break-ups and distortions to which many distance learners have become accustomed. It means, rather, that the learner must be able to depend on CD-quality sound delivered in real time. Even with recent advances in the delivery of streaming audio (and video), only a significant increase in bandwidth, which the vBNS connection would make available, would provide such an assurance and permit students and researchers at other vBNS sites to take full advantage of the course materials.
While the current project is limited to creating a course in Czech language and culture, the long-range plan is to make the basic software easily adaptable for use with any other foreign language. The implementation of the vBNS connection at Princeton would thus permit scholars at other vBNS-connected institutions to view a working 'model' upon which they could build an interactive French (or German or Chinese, etc.) language course that can be shared effectively over the network.
The JSTOR project (Journal Storage Project), for which the PI is the Chief Scientist, is an experimental, Andrew W. Mellon Foundation initiated project which is seeking to make current and legacy journals available to researchers world-wide through the medium of the World Wide Web. JSTOR is unique in that it provides researchers with on-line access to high-quality images of the actual journal text pages, as well as providing searching capability through background OCRing of the journal text. JSTOR can enormously facilitate the dissemination of critical scientific information, while providing instantaneous access to a vast volume of existing peer-reviewed literature.
Currently, JSTOR is providing comprehensive on-line access to 30 journals in the areas of Economics, Ecology, History, Mathematics, Political Science and Population Studies (see http://www.jstor.org for details). An additional 20 journals will be made available this year in the fields of Education, Mathematics, Finance, and Asian Studies. To handle the anticipated access load, JSTOR is running at two synchronized sites -- one at the University of Michigan (already a vBNS grantee), the other at Princeton University. The challenge for JSTOR is to maintain close synchrony between these two sites in the face of currently planned expansion in the list of available journals and journal issues.
For 1998, JSTOR will be adding journal pages at the rate of 100,000 pages/month. Each page is saved as both a high-quality TIFF image, and an OCR scanned image. The TIFF images average 150 KB per page, for a total of 15 GB/month of TIFF images, plus an additional 600 MB/month of ASCII data. Addition of journal pages will not be spread evenly through a month, however, and JSTOR will routinely need to exchange multiple Gigabytes of data on a nightly basis. Currently, this exchange occurs via DLT tapes, which necessitate at least a 4-day build/ship/load transfer time. JSTOR also conducts a continuous quality-control review of its current offerings; these quality control runs can generate 500 MBs to 1 GB of changed images.
The vBNS will provide JSTOR with the ability to maintain near-absolute synchrony in its journal offerings, an interesting experimental use of the network for data synchronization. This will become particularly critical as JSTOR moves from this experimental stage into being a primary provider of peer-reviewed literature in the academic community.
Rather than creating a "one size fits all" network, the University has designed a physical network structure that can cost-effectively accommodate the wide range of requirements that exist within the University community. Over the course of the past decade, the University has successfully distributed computing resources by creating a telecommunications infrastructure that now reaches every academic and administrative site on campus. Nearly 12,000 computing devices are now connected to the campus network at speeds that permit easy access to important information sources. DormNet brings Ethernet and Internet service into University dorm rooms for more than 3,000 students.
|
|||
|
Faculty /Staff connections |
DormNet connections |
Remote Access accounts |
|
|
|
6,672 |
3,005 |
2,311 |
|
|
6,384 |
2,496 |
1,466 |
|
|
6,243 |
1,772 |
728 |
|
|
6,073 |
1,033 |
320 |
The University has designed its campus network to meet current and future telecommunications needs. The design aims to reduce congestion points in the flow of traffic to central campus resource and to the external networks.
At the center (core) of the campus network, a Cisco router of the 7500 series, Whittaker ATM/Ethernet switches, and a Cabletron MMAC Plus are used to connect Fast Ethernet to the FDDI backbone. A Cisco Catalyst 1200 switch connects Ethernet to FDDI. Details of the core can be found in the diagram: Princeton University Core Configuration.
In the construction of the fiber backbone, provision was made by laying extra fiber to each building so that future high-speed networks could be built in parallel to the existing production network. In this way, essential services can continue undisturbed while new network infrastructures are put into service. The campus's fiber optic backbone includes both multi-mode and single mode fiber as well as empty conduit space for future growth of the backbone infrastructure. Every one of the campus's 140 buildings is connected to one of eleven local hubsites via a fiber bundle consisting of 16 multimode and 16 single mode fiber. Each of the hubsites, itself with space and environment for housing active electronics and passive optic equipment, is connected to the core of the campus network via a fiber bundle of 48 multimode and 48 single mode fibers.
Today, every campus building uses between two and four pairs of fiber to support its connections to the campus core and for specialized services such as IBM fiber optic channel extenders. The remaining fibers are intended for future growth in the production network infrastructure and for providing a way to construct networks for specific purposes or projects. By using spare backbone fiber optics cables, a single building can be divided into what logically appears to be separate buildings in order to provide increased network capacity to locations that require additional bandwidth. The use of VLan technology permits those logically separated buildings to be combined into a single IP subnet. Thus, the University has been able to achieve an increase in capacity without having to face the difficulties associated with renumbering hosts or services.
The campus-wide production Ethernets (one for the academic and administrative buildings and one for the dormitory network) provide connections to approximately 10,000 attached devices. The University offers a variety of services including shared 10 megabit Ethernet, switched 10 megabit Ethernet and 100 megabit Ethernet. The intra-building wiring within most buildings, Category 3 twisted-pair to the desktop, was installed almost 10 years ago. As part of the Category 3 cabling plant, a limited amount of fiber optics were run between wiring closets within a building. These fiber optics were installed to support the intra-building network infrastructure. As of two years ago, all new buildings and major renovations are cabled with Category 5 twisted-pair and fiber optics to the desktop. All wiring closets in these buildings are connected via fiber optics and have enough capacity to "home-run" each desktop to a single location within the building. At this concentration point, building electronics and/or fiber patch panels provide connections to the campus fiber backbone.
The University plans to re-cable all campus building with Category 5 - fiber optic cable to the desktop. The old Category 3 cabling is now being replaced with the new Category 5 - fiber optic cable on an as-needed basis to support new applications and facilities. The University has also designed a master plan for re-wiring all campus buildings over the next five years with the new Category 5 - fiber optic cable.
Today's core electronics consist of switched Ethernet (both 10 and 100 megabit) and high performance routers [see Princeton University Core Configuration in Figure 1]. The core electronics are connected via FDDI rings. The loading of these FDDI rings is carefully monitored to insure that data rate thresholds are not exceeded. The core provides access to campus-wide services such as compute and file servers, printing, and electronic mail.
Princeton uses multiple ISPs, GES (16 Mbs) and TCG (15 Mbs), to provide redundant Internet access for the campus. These service providers are accessed via the core network.
The University can easily and rapidly add additional capacity (such as additional FDDI rings or the deployment of new technologies) because the core network exists within a small physical space. For example, many of the University's central servers have been transparently moved to switched 100 megabit Ethernet.
The University is deploying switched Ethernet to replace the shared Ethernet hubs that are used to connect desktop systems within campus buildings. Use of switched Ethernet to the desktop permits the existing Category 3 intra-building wiring to provide greater throughput. Within a building, switches are linked via the existing closet fiber optic cable to create a full duplex 100 megabit high-speed building backbone. By using the campus fiber optic backbone to extend the building backbone to the core electronics, campus buildings have a bandwidth growth path. The bandwidth requirements vary greatly among campus buildings. The University is deploying the new Category 5 - fiber optic cabling to desktops or departmental servers that require more than 10 megabit service.
The vBNS connection will be integrated into the overall design of the campus network. A new parallel ATM network and the existing campus Ethernet network will be fully interconnected and be able to access the vBNS through the OC-3c link.
The University's master plan for the campus network calls for the creation of a parallel campus ATM network with an enterprise ATM switch supporting both wide area connections and on-campus connections. The proposed vBNS OC-3c connection will attach to the new enterprise ATM switch [see Princeton University Core Configuration in Figure 1]. The Princeton campus termination for the vBNS link will be at the ATM Enterprise Switch located in the networking alcove of the main University Computing Center machine room. This machine room is a secure facility with card access control, redundant air-conditioning and a motor-generator UPS system.
An on-campus OC-3c connection will link the enterprise ATM switch with the campus core. By adding an OC-3c connection between the existing campus production network and the new ATM network, the campus community as a whole will immediately gain access to vBNS resources. The central routers will enforce the vBNS acceptable use policy by means of policy based routing.
The parallel ATM-based network will provide QoS for the research detailed within this proposal and other researchers and projects that require high capacity access to the vBNS network. Within a building, connections to the ATM network will be implemented in a number of different configurations including, but not limited to, direct server connections, ATM edge devices, and other ATM switches. By using this design, the vBNS connection will be fully integrated into the campus network with sufficiently flexibility to accommodate and support researchers and projects with specific needs.
The construction of this plan has involved CIT's Network Systems and Desktop Support groups who are responsible for the operation of the campus network, faculty and/or their representatives from the proposed applications areas, other faculty with possible future interests in using the vBNS, and the University's wide-area network service providers.
The University staff will work closely with its current ISPs and MCI for the vBNS. The goal will be to insure robust and usable external access for all members of the University community. The University Network Operations Center is the locus of network monitoring for the campus and will coordinate with the MCI vBNS Network Operations Center to resolve any connectivity or QoS concerns on behalf of campus vBNS users.
Initial efforts to guarantee QoS at Princeton will rely upon the segregation of meritorious research traffic in order to avoid congestion. The network design detailed in section 4 of this proposal and illustrated in the Princeton University Core Configuration (see Figure 1) works to insulate meritorious research traffic from general campus and commodity Internet traffic. Initially, use of PVCs over ATM will permit reservation of vBNS bandwidth for meritorious researchers on the campus ATM network.
The University will investigate and test promising technologies and approaches such as use of SVCs, RSVP, tag switching, and MPOA. It is clear that there will be active research and development in this area over the course of the next few years. The University expects to aggressively watch and participate in this activity as well as tests and trials at other vBNS sites and will deploy appropriate technologies as they are made available to the production vBNS community.
Princeton has had a long history of participation in networking and the Internet. VP Fuchs was the co-founder of the highly successful network BITNET and was responsible for the development of VMNET which permitted the participation of BITNET in the Internet. Princeton was a charter member of JvNCnet, the first high-speed (T1 network) NSFnet consortium, and a home for JvNCnet for two years.
Princeton's Desktop Support Group (7 FTEs) has responsibility for the installation, maintenance, and daily operation of the campus networks that include the fiber network, broadband data network and CATV network. A wide range of equipment exists on the campus including private microwave for data and satellite dishes for video reception. This group is certified for Category 5 twisted pair and fiber optic installation.
Princeton's networking group (5 FTEs) is responsible for the overall planning and design of the logical network. This includes the configuration of routers and switches, external connectivity to the Internet (working with ISPs), and exploration of future technologies such as ATM.
The Princeton network has grown to 10,000 Ethernet connections and includes providing Ethernet access to each student's dormitory room. The campus network is multiprotocol, supporting not only TCP/IP but also IPX for the administrative Novell access. In addition, Princeton provides dial-in remote access services via traditional analog modem and ISDN and is about to deploy ADSL.
Peter Olenick is the manager of the Princeton University Networking Group. He has been involved with computing at Princeton for more than 30 years. In the early days, he was active in the development of communication packages for support of Remote Job Entry (RJE) and terminal access. As data communications progressed, he was an early supporter of BITNET, the world-wide academic network founded by Ira Fuchs.
When Ira Fuchs assumed the position of Vice President for Computing and Information Technology at Princeton, he headed a project (BITNET II) which was moving BITNET to using TCP/IP. As part of this project, Peter Olenick authored a program, VMNET, which created the ability for IBM protocols such as NJE (Network Job Entry) to be transported via TCP/IP. In turn, VMNET allowed BITNET sites to make use of the Internet. In addition to the programming, Olenick and Michael Gettes (another member of CIT at Princeton) were responsible for the reorganization of BITNET and EARN (the European equivalent of BITNET) into a tier structure based on availability of TCP/IP network access.
As the Princeton campus network has expanded, first to a broadband (CATV) network which has now been upgraded to a fiber optic network, Olenick and the Network Group have been responsible for the overall design of the network's layout. The Network Group continues to explore new technologies for deployment on the campus.
Given the physical structure of the campus network with its fiber optic backbone, central core, and upgraded intrabuilding cabling (see section 4.7), access to/from the vBNS will be available to any location on the campus which requires it.
The University has always charged network users directly for the on-going costs of their connections. By so doing, the University has been able to recover the costs associated with the operation of the campus network. The University intends to apply the same model for the continuing support of the vBNS connection.
The University will establish a High Performance Network Committee that will include the Principal Investigator as well as involved faculty. The Committee will publicize the benefits of the vBNS and encourage others to integrate such use in their research efforts.
The University ATM network will require an Enterprise ATM switch to act as the focus of ATM services on the campus. A cisco 2020 Enterprise switch will be configured for this purpose. The initial configuration will include eight OC3 ports. A cisco ATM interface board will be needed for the central campus router.
ATM workgroup switches and edge devices will be deployed within campus buildings to support researchers' access to the vBNS ATM network. Initially, the University will deploy three ATM workgroup switches equipped with eight OC3 ports per workgroup switch and two Ethernet/ATM edge devices equipped with four 100 megabit and four 10 megabit connections.
In addition to these infrastructural equipment costs, there will be cabling costs associated with extending fiber or Cat 5 to the researchers' machines and interface boards (ATM or 100 megabit Ethernet) for the machines themselves.
The University is actively working with the other universities (Rutgers University, the University of Pennsylvania, the University of Delaware, and Lehigh University) in MAGPI [Mid-Atlantic gigaPOP for Internet2] to plan for cost-effective, shared Gigapop access. As these plans take root in the near future, the University will review its options for vBNS connectivity.
After six months of operation and every six months thereafter, an evaluation will be conducted to review how the vBNS has benefited each researcher with access to the vBNS. The evaluation will focus upon two areas:
The PI attends and is a regular speaker at all major conferences and meetings related to networking in higher education. The PI will actively disseminate detailed information regarding the infrastructural effort and its impact on research and instruction. Dissemination of information related to the project will occur at conferences, in published articles, and on the web.
The web site: http://www.princeton.edu/cit/vbns.html contains a copy of the proposal as well as relevant links to information about the University's telecommunications infrastructure as well as links relevant to Internet2 and vBNS activities. The site will be used to track project milestones as well as the results of the above evaluations.
The nature of the multi-university HCI coursework and research will make the benefits of the new network capabilities known to the other participating universities (Stanford, San Jose State University, and U.C.
Santa Cruz). Course instructors and researchers from these institutions have presented first-year reports on the 1996-7 course at the Acoustical Society of America Conference, the American Society for Engineering
Education Conference, and the Salzburg Symposium on Networked Music Applications. Future conference presentations and publications are planned for dissemination of experiences and technology gained from the research and course offerings.
Finally, the University will host demonstrations that will detail its experiences with the infrastructure as well as the impact in the various research areas.