NSF Form 1207 - Cover Sheet - page 1
NSF Form 1207 - Cover Sheet - page 2
Florida's senior institutions of higher education have a long history of leading edge research, diversified and high quality educational programs, and the use of networks to support these activities. However, the networking needs of advanced research can no longer be met solely through the current Internet. To remain on the frontiers of research, Florida's universities require access to a high-performance network infrastructure with characteristics that are suited to the research environment-the NSF very High Speed Backbone Network Service (vBNS).
To address these challenges, seven of Florida's ten state universities and one of the state's leading private institutions have come together to develop a coordinated plan for building a high-performance statewide network for research and education. These institutions are: Florida A&M University, Florida Atlantic University, Florida International University, The Florida State University, the University of Central Florida, the University of Florida, the University of Miami, and the University of South Florida. The proposed network will link these institutions with each other, with other research universities in the southeast, and with the NSF vBNS. All of the universities represented in this proposal are members of the Internet2 consortium.
This proposal requests partial support for the FloridaNet project from the National Science Foundation in the amount of $2,800,000 to support the construction of a distributed gigapop linking the eight participating institutions and a high-speed connection to the vBNS. Matching funding from the participating institutions will exceed this amount. The University of Central Florida will serve as fiscal agent for this proposal.
FloridaNet will be developed in phases, with the initial phase configured as a star topology with the central gigapop located at the University of Florida in Gainesville (see Figure 3 on page 56). Each institution will connect to the gigapop at DS3 to OC3 speeds. The FloridaNet distributed gigapop will be connected to the Southern Crossroads gigapop at the Georgia Institute of Technology at OC3 speed. The Southern Crossroads gigapop will, in turn, have OC3 speed access to the vBNS, migrating to OC12 when available.
In addition to providing Florida's research institutions with access to the vBNS, FloridaNet will provide connectivity for commodity Internet traffic on a non-competing basis. This will create, for the first time, a true higher education network serving all of the needs of Florida's institutions of higher education. Using state and institutional resources, FloridaNet can be expanded beyond the configuration proposed herein to provide connectivity to all of the state's universities and colleges (ten four-year universities, 97 independent institutions and 28 community colleges). Of these institutions, only the eight participating universities will initially have access to the vBNS.
The institutions represented in this proposal are firmly committed
to the development of FloridaNet and the advancement of high-speed
networking to support leading edge research. This is evidenced
by the applications proposed, the levels of institutional financial
commitment, the participation of each institution in the Internet2
project, and the letters of institutional commitment and support
included in Appendix B.
Cover Sheet (Form 1207) 1
A. Project Summary 3
B. Table of Contents 4
C. Project Description 5
C.1 Statement of Work 5
C.2 Participating Institutions 6
C.3 Objectives and Expected Outcomes 6
C.4 Plan of Work 6
C.4.a FloridaNet Applications 7
C.4.b FloridaNet Network Design 53
C.4.c FloridaNet Network Engineering Plan 55
C.4.d Individual Campus Network Infrastructure Plans 58
C.5 Cost Sharing 66
C.6 Results from Prior NSF Support 66
C.7 Disseminating Results 67
D. References Cited 68
E. Biographical Sketches 68
F. Budget 75
G. Current and Pending Support 76
H. Facilities, Equipment and Other Resources 77
I. Special Information 78
J. Appendix A: Profiles of Participating Institutions 79
K. Appendix B: Letters of Support 88
C.1 Statement of Work
The need for a high-speed network infrastructure to support research and advanced services exists not only for tomorrow-it exists today. The commodity Internet, once the sine qua non of education and research applications has become increasingly congested and overloaded to the point it can no longer reliably support many advanced applications. Further, this network was not designed to support the emerging research applications that demand very high bandwidth and predictable network performance characteristics.
This proposal describes a plan to create a network not only to support advanced research, but also to support education, collaboration, distance learning, medical applications and other important uses.
FloridaNet will be developed in stages. The preferred design consists of dual SONET rings interconnecting all participating institutions. That topology cannot be created today in Florida because the required fiber optic paths are not available. Instead, FloridaNet will begin with a star topology homing to a gigapop at the University of Florida in Gainesville. The gigapop will be operated by the University of Florida on behalf of the participating institutions. The gigapop will be connected to the vBNS through the Southern Crossroads peering point at the Georgia Institute of Technology. This simple, yet effective design will provide each participating institution with direct access to the vBNS and to all other participating institutions. It will also allow individual institutions to increase the bandwidth of their connection to the gigapop as need and resources allow. Further, pairs of institutions can install additional inter-institutional links, further meshing the network and providing added redundancy. This design is robust, straightforward to manage, and will provide a suitable platform for migration to the eventual SONET ring topology.
Within the near future, additional state and independent institutions and industrial and government research partners are expected to connect to FloridaNet, providing increased opportunities for collaboration and enabling new and as yet unforeseen research applications.
The eight institutions participating in this proposal are: Florida A&M University, Florida Atlantic University, Florida International University, The Florida State University, the University of Central Florida, the University of Florida, the University of Miami, and the University of South Florida. Letters of institutional support from these institutions, and others, are included in Appendix B.
The following 1995-1996 statistics provide an indication of the scope and quality of the State University System of Florida:
National Merit Scholars: 902
National Merit & National Achievement Scholars: 1,350
(Increase of 99% over 1988)
Headcount Enrollment: 206,033
Baccalaureate Degrees Granted: 32,208
Master's Degrees Granted: 8,806
Doctoral Degrees Granted: 1,058
First Professional Degrees Granted: 1,002
Total Sponsored Research Trust Fund Expenditures: $493,589,283
With this application, the eight participating institutions respectfully
request funding from the National Science Foundation in the amount
of $350,000 each over two years. The total sum requested is eight
times $350,000, or $2,800,000. This amount will be matched by
$2,804,432 in direct project expenditures from the participating
institutions over the two year period of NSF support. In addition,
the eight institutions have reported plans to invest more than
$24,500,000 during the same period to upgrade and expand their
respective campus networks. The University of Central Florida
will serve as fiscal agent for this proposal.
C.2 Participating Institutions
The institutions participating in this proposal are, in alphabetical
order, Florida A&M University, Florida Atlantic University,
Florida International University, The Florida State University,
the University of Central Florida. the University of Florida,
the University of Miami, and the University of South Florida.
Institutional profiles describing each institution are included
in Appendix A. The President or similar high level official of
each institution has provided a letter of commitment and support
for this project. These are included in Appendix B.
C.3 Objectives and Expected Outcomes
FloridaNet is being created to establish a high-performance network infrastructure and services that will support research and educational applications that are not viable on today's congested Internet. The institutions participating in this proposal are among the charter members of Internet2. FloridaNet is designed to become an integral part of the emerging Internet2 architecture, as well as the current NSF vBNS. Phase one will provide connections to each institution at DS3 to OC3 speeds, with a shared vBNS connection operating at OC3. In addition, a gigapop and network management center will be constructed at the University of Florida, the convergence point for all phase one FloridaNet links.
FloridaNet will be engineered to carry not only high bandwidth
research traffic, but also commodity traffic destined for the
commercial Internet. Florida does not have an intranet for higher
education. Florida institutions obtain Internet connections from
a variety of providers, which has led to a number of problems
including unpredictable bandwidth, propagation and network performance
into the commercial Internet. With the creation of FloridaNet,
both sets of issues will be addressed. FloridaNet will provide
an intranet to support current and emerging high bandwidth research
applications, while at the same time providing an architecture
that will be capable of carrying the institutions' Internet traffic.
C.4 Plan of Work
The eight institutions participating in this proposal have spent many months in planning the development of FloridaNet-both at a coordinated state level, and as individual campuses. As this venture has progressed, the institutions have obtained the encouragement and support of the state Board of Regents, the Information Technology Program of the Department of Management Services, the Florida Information Resource Network, the state's telecommunications industry and participating vendors.
A state-wide Networking Task Force has been formed, with administrative and technical representatives from participating institutions and other agencies. The Task Force has coordinated all project activities to date. In cooperation with the Board of Regents and other applicable agencies, a permanent organizational structure for network governance, management and policy determination will be formed. This organization will be in place before FloridaNet begins operations.
The core of this proposal is the high bandwidth applications that
are listed in the tables below and described in detailed in the
section that follows.
C.4.a FloridaNet Applications
The meritorious high bandwidth applications to be supported by
this proposal are summarized in the table below and described
in detail following the table. These applications are supported
by a total of approximately $12,000,000 in research funding from
federal, state and private agencies.
Bandwidth |
Considerations | |||
622 Mbps Future | ||||
70 Mbps | 10 Mb per each concurrent instrument Current application |
Bandwidth |
Considerations | |||
Future application | ||||
40 MBps | 10 Mbps per Research Group Current application | |||
Current application | ||||
Current application | ||||
Current application | ||||
Current application | ||||
Current application | ||||
Future application | ||||
Current application | ||||
Video Compression |
| |||
|
Bandwidth |
Considerations | |||
| ||||
Multicast | ||||
Multicast | ||||
Multicast | ||||
Multicast | ||||
Multicast | ||||
South Florida Geographic Information System | Multicast | |||
Bandwidth |
Considerations | |||
Multicast | ||||
Multicast | ||||
Molecular Databases | 500 Mbps fut. | |||
for Real-Time Distributed Systems | ||||
to Molecular Reaction Theory | ||||
100 Mbps fut. | ||||
Management and 3-D Visualization Systems | ||||
Full descriptions of these applications follow.
University of Florida Brain Institute
The mission of the University of Florida Brain Institute (UFBI) is to provide a research, teaching, and clinical environment wherein a dedicated and evolving team of interdisciplinary researchers in the neurosciences will have the resources and freedom to efficiently and effectively focus their creative energies on fundamental, as well as clinical and commercial, applications of brain research.
With over 200 faculty from over 50 different departments, divisions, centers and programs of the University of Florida, the UFBI is clearly considered to be a major focus of the strengths of this university in the 1990's and well into the next century. To put this into perspective, a recent assessment of the neuroscience-related projects at the University revealed an annual total of well over $17 million in sponsored research. Some of these projects represent individual laboratory efforts, whereas others represent multi-laboratory, multi-departmental and even multi-college efforts. As a dramatic extrapolation of the latter, it is important to note that the $17 million annual total does not include the June, 1992, award of an $18 million competitive, peer-reviewed grant from the Department of Defense to help in the construction of a major new facility for various programs within the UFBI. This grant was matched by nearly $30 million in additional matching construction funds from the UF College of Medicine and Shands Hospital, and more recently by nearly $20 million in funding for infrastructure equipment from the Departments of Defense (DoD) and Veterans Affairs (VA).
The programs within the new UFBI building are numerous and will include a linear accelerator-equipped Radiobiology/surgery Research and Training Core facility, a Brain Tissue Bank facility, and a "Good Manufacturing Practice" Gene Therapy Vector Core Facility for human investigations. Of more relevance to the current initiative, the new building will also house the Center for Structural Biology. In addition to advanced optical and electron microscopy facilities, this latter center will include 7, 14.1, and 17.6 Tesla NMR magnet systems for studies of microscopic anatomy and macromolecular structure, and a 4.7 Tesla NMR magnet system for the study of animal models of human disease.
As part of the Center for Structural Biology, the UFBI will jointly operate a 12 Tesla magnet system, with the NSF-funded National High Magnetic Field Laboratory, as a national resource for MR imaging and spectroscopy. Also the UFBI will operate a unique clinical research and development teaching facility known as the Center for Advanced Practical Neuroscience (CAPN). The CAPN will include a number of computer-network linked lecture and demonstration facilities, as well a major, informatics-intensive cadaver dissection area. These facilities will house fully equipped animal neurosurgical operating rooms with computer workstations and access to the network for viewing MRI and other digitized imaging information, and for assistance with experimental stereotactic surgery. The CAPN will also be the hub of the UFBI's advanced informatics and scientific database management capabilities. Plans call for the installation of equipment to capture and deliver visual, auditory and digital (e.g., MRI) information over high-speed network connections throughout the UFBI, as well as to remote investigators. This will include the transfer of information to and from faculty offices, informatics/computer work rooms, core laboratories, faculty research laboratories, animal operating and testing rooms, teaching facilities, and the University's clinical research facilities.
As a university-wide and national facility, many of the faculty participants within the UFBI, and collaborating investigators from outside the University, will not be housed within the main UFBI structure, but they will still have access to all of the UFBI facilities and, through high-speed networks, they can participate in a "virtual institute" environment. In some cases (e.g., Electrical Engineering) this level of network participation, coupled with the need to retain a close proximity to others within their specialized discipline, is deemed the optimal relationship.
When the UFBI opens in the fall of 1998, the operation of the "virtual institute" will be extended to a national level with the inclusion of collaborations through the National High Magnetic Field Laboratory, the DoD and VA. As mentioned above, the UFBI will have several core facilities that will be operated remotely by investigators who have sent their subjects ahead of time. As an example, the use of the NMR systems by remote investigators will allow the study of macromolecular structure, excised tissue, and whole animals. To operate these systems, a remote investigator requires high-speed interactive instrument control. Once the sample is placed in the instrument by local UFBI Staff, the remote investigator will be given (and must have) direct control of the measurements. This can be provided by a high-speed network connection that allows input and nearly immediate feedback of information (spectra or image) from the instrument. This approach to resource sharing is efficient in personnel time and effort by effectively bring the "lab" to the investigator. In addition, the teaching and research resources (beyond core labs) of the UFBI will become available to remote participants once the high-speed network is in place. Therefore advanced training and research opportunities, both fundamental and applied, can be extended to a wider group of remote participants.
The technology employed in the UFBI core laboratories, and teaching
and research facilities, will require a range of network performance
from 25 Mbps to 1.2 Gbps. For example, remote NMR or MRI measurements
will require at least 150 Mbps to 250 Mbps bandwidths and, with
interactive audio and video, the requirement easily can increase
to 500 Mbps to 800 Mbps. In the surgical teaching facility, the
remote preview of deposited archived data, and fusion with live
video (possibly at HDTV quality), will require a bandwidth over
800 Mbps up to 1.2Gbps. Therefore, the operation of the UFBI as
a "virtual institute" demands the highest networking
capability available.
Networks to support inter-institution image processing and clinical diagnosis using digital medical images:
The Department of Radiology at the University of Florida produces large volumes of digital images, generally using direct digital acquisition devices such as Computed Tomography, Magnetic Resonance, and Nuclear Medicine. In addition, images that were traditionally produced as film are being captured in a digitized form and film is becoming less important to the diagnostic process. The department currently produces, stores, and displays 1 Terabyte/year of digital images that will increase to 3 Terabytes/year during the coming year. One of the reasons for implementing a digital radiology department is the opportunity to use digital image processing to enhance the diagnostic and treatment process as well as to follow disease progression to evaluate options for therapy. In addition, clinicians and scientists perform research to enhance imaging techniques to produce digital files to support the instructional mission.
A number of Sun Microsystem computers are available in the Department of Radiology that can be used for this project. The department owns software that may be appropriate for some of the required tasks. 3d-Viewnix, the Matlab image processing toolkit, IFL, and AVS software are available. In addition, software has been obtained for specific applications such as multi-sensor fusion with SPECT and MR images. Clinicians and researchers could benefit from collaboration among other institutions interested in image processing that will require high-speed networks to transfer these images. Digital mammography images are typically 40 Megabytes each with four to five images included in a study. New scanning protocols on Magnetic Resonance and computed tomography scanners allow image sets of 1000 or more individual images.
In addition to image processing research, this application will investigate new methods of collaboration among multiple institutions over the proposed high-speed network. The purpose of this work is to build the infrastructure for universal image display and analysis so digital methods can be used for all radiology and medical images.
Immediate image processing research problems include the following:
Key Participant: Janice C. Honeyman, Ph.D., Associate Professor,
Department of Radiology
High Field Whole Body Magnetic Resonance Imaging Network
The growth of new medical imaging techniques has required the development of a large image archiving capability that in turn places tremendous demands on the existing network to transfer images for clinical and research use. In particular, the development of magnetic resonance imaging (MRI) and magnetic resonance spectroscopy (MRS) at the University of Florida and the VA Medical Center have provided a wide range of research and clinical applications. Over the past ten years, the Department of Radiology has encouraged and supported the acquisition of a number of different MRI/MRS systems for both research and clinical development. The basic science support for these systems has been provided by the NMR physics group within the Department of Radiology through a number of NIH and VA grants. This funding and research has been critical to the successful development of MRI for its clinical application to diagnostic radiology
The university and VA Medical Center have recently invested over $3 million for a unique facility that houses a 3T whole body MRI/MRS system for which the support requirements will be more demanding and complex than any existing systems locally. This new system is intended to meet a range of clinical and research demands that will go well beyond the Radiology Department's boundaries and will include the interests of surgeons, internists, neuroscientists, cardiologists, oncologists, and many other basic science disciplines. Therefore, the need for a more effective high-performance network has never been greater.
The 3 Tesla facility was created for the purpose of expanding the already considerable effort in the fields of diagnostic imaging and neuroscience in particular through the use of emerging technologies in high field whole body magnetic resonance imaging (MRI) and spectroscopy (MRS). The development of functional brain imaging (fMRI) using high field systems with high-speed imaging capabilities greatly enhances research potential as well as the ability to obtain funding from NIH and other important funding agencies. A prime objective of the 3T program is to support the University of Florida Brain Institute (UFBI) in its efforts to fully explore the new frontiers of neuroscience. Since the UFBI is an essential aspect of the College of Medicine's (COM) strategic plan, the 3T program is in alignment with those same priorities. Primary areas of research to be done at this facility include but are not limited to:
In summary, it is clear that this application has the capability
of generating voluminous quantities of image data with the acquisition
technology available. Research to date has led to means of storing
these large data sets as well; however, at present the investigators
have a serious lack of network bandwidth that will enable sharing
data among major institutions of higher learning. The availability
of a new high-speed network could not be more timely for these
applications.
Key Participant: Jeffrey R. Fitzsimmons, Ph.D, Professor, Department
of Radiology
The University of Florida has several dedicated research programs closely interconnected to programs at other institutions nationwide, and more locally and more tightly to sister institutions in Florida, that can reach their full potential only through the high-speed high bandwidth networking that will be available with the proposed vBNS connection. These focused research efforts offer highly specialized and costly resources (such as the access to experiments at the National High Magnetic Field Laboratory (NHMFL), or sharing high density images in near real-time through Florida's Imaging Science Center (ISTC) in a shared network to provide cost effective use by a larger research community. Most importantly the high-speed network will expand scientific horizons by bringing new people and new ideas together on the most complex and challenging problems in a manner that will be particularly accessible to students. This will provide a particularly effective way to integrate research and education within the group of participating Florida institutions.
In addition, the proposed high-speed network will allow the University of Florida to significantly enhance its educational service and scientific exchanges with institutions in the Caribbean basin and Latin America. Highly specialized courses and data bank libraries at the large institutions can be accessed by participants connected to the network. To realize this capability the State of Florida will form a Florida Pipe of high-speed networking to reach out to member institutions and to integrate fully with the vBNS.
Key Participant: Neil S. Sullivan, Professor and Chair, Department
of Physics
National High Magnetic Field Laboratory (NHMFL)
High-speed networking for rapid on-line exchange of data between the remote sites of the National High Magnetic Field Laboratory (Tallahassee, Gainesville, Los Alamos) will enable investigators to create a unique interactive environment in which experiments can in many cases be carried out from remote sites. For example, a solid state physicist at Los Alamos can measure the electrical transport in a new semiconductor as a function of magnetic field by observing data on-line from an experiment being carried out in Tallahassee totally under computer control; or a biomedical sciences student in Tallahassee can analyze on-line magnetic resonance images of his specimens being studied by the unique 12 T, 40 cm small animal imaging facility in Gainesville.
Feedback from results of experiments at the NHMFL can be communicated to remote sites, to modify experimental parameters on-line and in near real-time. This capability will enormously enhance the experimental capabilities and the access to students. This is critical for optimizing the productivity and use of costly magnet time, and will also make the facilities available to a larger group of external users.
Videoconferencing capabilities are being introduced to maximize the integration of the advanced research with the educational needs of the institutions by bringing undergraduate and graduate students on-line to the experiments. The NHMFL is developing a magnetic resonance network first in Florida and then extending it to the nation; i.e. the National Magnetic Resonance Collaboratorium that will function as a virtual laboratory. This virtual laboratory will respond to science, users and educational activities critical to the mission of the NHMFL and the NSF.
The existence of the high-speed link will lay the groundwork for a qualitative shift in the NHMFL-user interaction for all outside users in the future. If such outside users also have nearby access to high-speed links (expected for more than 70%), the user experience can over time shifted from having to be physically at the site of the magnets to one where each can remain at their respective laboratories and perform experiments remotely. This will lead to a greatly improved and expanded use of the facilities and the quality of the research performed. Examples of research efforts that the Internet2 capability will enhance at NHMFL and provide remote capabilities include: studies of properties of new and novel materials in very high magnetic fields such as high-Tc superconductors and heavy fermion alloys, measurements of quantum confined structures, thin films and quantum heterostructures. The experimental technologies include: transport measurements, heat capacity studies, NMR and optical measurements. All of these studies require the accumulation and processing of large amounts of data and can only be carried out by users at remote sites when the speed of the vBNS connection becomes available to this national facility.
Key participants: UF: Neil Sullivan, John Graybeal, Tom Mareci,
John Eyler, Pradeep Kumar, Dwight Adams, Arthur Hebard, Kevin
Ingersent, Jack Sabin. FSU: Jack Crow, Robert Schrieffer, Bruce
Brandt, William Moulton, Zachary Fisk, Hans Schneider-Muntau,
Timothy Cross, Alan Marshall, Stefan Von Molnar. LANL: Don Parkin,
Laurence Campbell, Christopher Hammel, Joseph Thompson, Alexander
Lacerda.
Imaging Science and Technology Center
This new center involving a consortium of institutions (Universities of Florida, Central Florida, South Florida, Florida State University, and the Swiss Federal Institute of Technology), industrial partners in advanced computing and visualization (Silicon Graphics Inc., Lockheed Martin ) and specialized research centers (the National High Magnetic Field Laboratory, the Center for Research in Electro-optics and Lasers (Orlando), and the Moffitt Cancer Center (Tampa) is focused on developing real-time high resolution imaging for basic research and applications where information is needed at ultra-fast time scales or ultra-short length scales. The technologies include magnetic resonance imaging/spectroscopy (MRI/S), femtosecond laser spectroscopy, laser-induced X-ray microscopy and near-field optical microscopies. All of the applications depend critically on high-speed data acquisition coupled intimately to advanced image processing for real-time imaging; e.g. functional imaging of the brain, studies of dynamical processes in nanoscale devices, and on-line control processes in manufacturing.
The Internet connectivity at the much higher bandwidths of the vBNS is critical to enabling sharing of resources between institutions in areas of research such as medical imaging where high volumes of material need to be transmitted in short time-scales to be useful to on-line controls and interactions. The enhanced networking will enable participants to share and augment the capability of existing resources through reliable exchanges of benchmark references to test imaging processes and exchange raw and processed data streams between different sites. This will allow wider access to rare and costly facilities, and the State of Florida will develop a high-speed backbone for high-performance access of users on the Florida backbone to the high-speed vBNS connection.
The vision of the ISTC networking in Florida is to enable users to compare high volume images for testing and developing improved image processing for enhanced feature extraction and greatly improved reliability. The enhanced economy of scale will allow investigators to interact in a shared mode that is critical to research and applications where image gathering, analysis and processing high volumes of data at sufficiently high speed is the bottleneck to realizing major research objectives. As an example, investigators at one location using a new modality of MRI to produce a series of images of heart functions in a time sequence, can access benchmark reference images from a remote library to test and then refine their local image production and feature extraction methods. This capability will bring together the research scientists collecting images and the theoretical modeling experts to allow investigators to develop new algorithms and create new software/hardware environments to radically change the capability of high-speed imaging for real-time and near real-time applications.
Key participants: Neil Sullivan, Joseph Glover, Gerhard Ritter,
Edward Geiser, Tom Mareci, Samuel Trickey, David Wilson, Jeffrey
Fitzsimmons, Raymond Andrew, Reza Abbaschian, Jim Dufty, Mihran
Ohanian.
Elementary Particle Physics
High-speed Internet connectivity is of vital importance to the next generation of high energy physics experiments and their analysis. The UF high energy physics group conducts and leads major components of large-scale experiments at the large hadron Collider (LHC) at CERN in Geneva. This machine is now the only one worldwide to access the ultra-high energies where fundamental new particles and new physical forces can be expected to be discovered. Competition to participate in these experiments and also for the U.S. to maintain a presence in this field are such that teamwork and cooperation on experiments is the only way research can be done. The sharing of data and connectivity to different sites and the principal site is essential or individual universities cannot play a meaningful role. Large teams of users therefore collaborate on a few selected experiments that produce large amounts of data that need to be analyzed to detect possible new particles produced in high energy collisions. Each collision produces showers of hundreds of secondary particles, and the energy and momentum of each component needs to be tracked and analyzed in near real-time to optimize the experimental activity.
These outgoing particles bunch together into identifiable bursts called "jets." The "jets" reveal the underlying parton (quark and gluon) substructure of the event. The complexity of the events recorded is such that the data can only be treated effectively by multiple teams connected across several international sites through high-speed networks. This advanced capability is crucial to maintain a U.S. presence in this highly competitive field. As a particular example, Florida scientists have developed new applications of neural network theory to analyze complex "multi-jet" events in order to detect the signatures of new supersymmetric processes that are expected to be observable for the first time at these very high energies. International sites will be reached via the vBNS bridge to ESNET. Access to the vBNS network from Florida is essential to be able to carry out these fundamental research projects in collaboration with other institutions across the United States.
The University of Florida is also actively participating in collider experiments at Fermi National Accelerator Laboratory in Batavia, Illinois. The Accelerator at Fermilab will be upgraded in the next three years and will be able to deliver by one to two orders of magnitude more events of particle collisions at the highest energies available before the LHC comes into operation. Analysis of this data in collaborating institutions including UF requires transmission through Internet of huge amounts of data.
Key Participants: Guenakh Mitselmakher, Richard Field, Paul Avery,
Pierre Ramond, Andrey Korytov.
Laser Interferometer Gravitational Observatory (LIGO)
The University of Florida recently joined the California Institute of Technology and the Massachusetts Institute of Technology in the largest and most sensitive ever search for gravitational waves from the Universe, predicted by the Einstein general theory of relativity but never directly observed so far. This project called LIGO (Laser Interferometer Gravitational Observatory ) assumes construction by the year 1999 of two gravitational waves observatories in Louisiana and Washington. The data from these observatories will be transmitted through Internet to participating institutions, including the University of Florida. Again the amount of data will be very large and requires substantial improvement of Internet capabilities. Bandwidth required: 80 Mbps.
Key participants: Guenakh Mitselmakher, David Tanner, David Reitze,
Bernard Whiting
Quantum Theory Project (QTP). An Institute for Theory and Computation in Molecular and Materials Sciences
QTP is one of the largest and best-known groups world-wide in theoretical and computational chemical physics. The group consists of 10 faculty, one emeritus faculty, several adjuncts and international visitors, and typically 15 postdocs and another 15 graduate students. Among the membership there are many long-distance collaborations whose potential for innovation is hindered seriously by the bandwidth limitations of the current commodity Internet.
Examples include:
Jeff Krause: work with Kent Wilson (UCSD) and Ken Schafer (LSU) moves 100's of MB of visual data repeatedly by ZIP drive sent through U.S. Mail or FedEx. Hai-Ping Cheng: collaboration with John Gillaspy's experimental group (NIST) requires comparison of simulation and measurement that involves multiple fractional-GB file transfers. The current bandwidth-limited strategy is to minimize transfers (rather than optimize the research) and use FedEx as a crude alternative.
Rodney Bartlett: daily supercomputer use at Maui and Ohio State currently is limited to e-mail and ftp submittal of minimal input data files and batch execution scripts followed by retrieval of modest subsets of the total calculated data.
Sam Trickey: collaboration with Jon Boettger (Los Alamos National Lab) currently is limited to interchange of input data files.
Mike Zerner: very large files of biological molecule structure-energy data, are being communicated, slowly, in a collaboration with Klaus Schulten (University of Illinois).
The unexploited potential in these examples is for "meta-computing," the coupling of geographically dispersed machines and facilities to provide the suite of tools needed for complex calculations. What is being done today is a very limited version of what could be done with a qualitative improvement in usable bandwidth. For one example, Bartlett's group will be able to make far more effective use of calculated results if effective means were available to move detailed output data, computed at Maui, to UF and to the special graphics lab at Wright-Patterson AFB in Dayton, OH. Other examples of meta-computing have been conceived in QTP but not even tried because "back of the envelope" estimates of the linking bandwidth requirements made clear that the commodity Internet is a hopeless bottleneck.
NSF's recent PACI (Partnerships for Advanced Computational Infrastructure) awards are an unmistakable signal that aggressive use of metacomputing techniques and models is the direction for very-large-scale high-performance computing of the future. Without participation in Internet2 and similar forefront national research network initiatives, QTP's competitiveness will be sharply reduced. Increasingly, no access to advanced networking will mean no participation in computationally-based advanced research.
In summary, the Internet2 connection is vital for success of collaborations QTP members are undertaking now, not to mention those "on the shelf." These are not possible dream projects that someone might vaguely want to do some day. They are being done today, albeit sub-optimally. Bandwidth requirement: 155 Mbps (OC3) now, and 622 Mbps (OC12) in 18-24 months.
Key participants: Erik Deumens, Sam Trickey, Jeffrey Krause,
Hai-Ping Cheng, Rodney Bartlett, Michael Zerner, Yngve Ohrn, and
John Sabin.
Digital Library
Digital library services are provided by the Florida Center for Library Automation (FCLA) to the 50 libraries in the ten universities making up the State University System (SUS). FCLA is hosted by the University of Florida and located in Gainesville, but its mission and services extend to the entire SUS. Operating on large mainframe and UNIX platforms at the Northeast Regional Data Center, which is also located in Gainesville, FCLA provides software for the traditional library functions of acquisitions, fund accounting, catalog maintenance, circulation, serials control, and on-line public access to the catalogs. More importantly to this proposal, FCLA maintains digital library software that provides SUS students and faculty with electronic access to scholarly materials. Under its digital library component, FCLA provides gateways to remote databases and licenses proprietary databases for storage and retrieval from the large NERDC servers, and serves as a gateway to over a hundred copyrighted scholarly databases nation-wide. Even though these databases contain mostly ASCII data, the current state of the Internet has already proven inadequate for this relatively low bandwidth service.
Future digital library services will have a large portion of image data in the form of bit-mapped journal pages, photographs, maps and compound documents of text and image making the need for high network bandwidth even more critical. The data that will be selected in the future will be a mix of license copyrighted data, government public domain data and locally scanned library special collections. The common theme in future digital library offerings will be the value of the data for SUS research. A current project, already underway, is the licensing of the electronic version of 650 scientific journals from the Elsevier Publishing Company. At present, Florida and Ohio are the only two systems of higher education in the U.S. that are undertaking this massive delivery of Elsevier science literature. FCLA has the computer capacity for Elsevier and many other future projects. The sole obstacle to the rapid expansion of digital library services is network bandwidth.
Key Participant: Jim Corey, Director, Florida Center for Library Automation
National High Magnetic Field Laboratory (NHMFL)
The National High Magnetic Field Laboratory
(NHMFL) is charged by the NSF to provide high magnetic field research
facilities to scientists from all over the world, develop state
of the art magnets and instrumentation, and have a cadre of scientists
who use and drive the development of the high field research facilities.
The NHMFL's activities are distributed among three sites -- at
Florida State University, University of Florida, and Los Alamos
National Laboratory. The laboratory's distributed activities and
world-wide user community mean that it relies heavily on Internet
services. In 1996, the NHMFL had active collaborations with 53
national and international universities, laboratories, and high
technology corporations involved in NMR and general condensed
matter research. The following paragraphs describe ways that a
vBNS connection could improve existing research uses, make practical
processes that are possible but inconvenient now, and enable new
levels of remote control and interaction at a distance.
Key Participant: Bruce L. Brandt, Director,
Instrumentation and Operations, National High Magnetic Field Laboratory
NHMFL: Nuclear Magnetic Resonance - High-speed connections could dramatically change the way Nuclear Magnetic Resonance (NMR) spectroscopy is performed. There are currently seven NMR spectrometers in constant use at the NHMFL. Such machines can cost well in excess of a million dollars each and require researchers from around the world to expend large sums to travel to the laboratory to use them. The Internet provides a means by which these scientists can accomplish the same research by simply sending their samples to the laboratory and running their experiments remotely. A local staff member sets up the hardware and loads the sample, whereupon the researcher runs any number of experiments via computer control. Although desirable and barely possible with present technology, such a scenario does not often occur because of the limited bandwidth and unreliability of current Internet connections. Three specific areas in NMR that will benefit from a high-speed, highly reliable vBNS connection are described in detail below.
Remote Spectrometer Control. All modern NMR spectrometers are computer-controlled. Because only a limited number of companies make such instruments, only a few common software packages are in use to provide the user interface, and these run in an X-Window environment. Thus, these instruments can be run from a remote location, but this is seldom done because the spectrometer response is not transmitted to the user quickly enough, thus making essential adjustments (such as magnet shimming) impractical. A high-speed Internet connection with Quality of Service guarantees will eliminate this bottleneck.
Remote Processing and Data Visualization. A typical data set from a two-dimensional NMR experiment will be roughly 100 MBytes and there are much larger ones. These require computer processing (Fourier transformation, windowing functions, transposition, etc.) that is both CPU- and storage-intensive. In addition, the processing is semi-iterative. It demands graphical visualization as the user compensates for phase offsets and signal to noise problems. Commercial NMR spectrometers are thus equipped with the ability to perform "remote" data processing. The processing can, in theory, be done at any site as long as there is a means of transporting the large data sets and the remote facility has the processing software. All too often, one or both of these conditions is not met, so that "remote" ends up meaning that the processing is done on a computer that is in the same building as the spectrometer (one that essentially has a direct Ethernet connection). True remote data processing becomes feasible when the remote link has the 10 Mbps speed of the local Ethernet.
Remote Simulation.
A single NMR experiment can run for days, or even weeks. In such
cases it is critically important to set all experimental parameters
properly. Countless hours of valuable instrument time have been
lost from running "nonsense" experiments that result
from just one parameter set incorrectly. Computer simulation can
be used to prevent such occurrences. There are few areas in physics
where mathematics can predict experimental outcomes as well as
it does in the treatment of NMR. It is feasible to perform a simulation
using the exact interface used for the experiment. Thus, time
consuming experiments can be run through a simulation first in
order to ensure that various parameters are properly set, avoiding
the potential for wasting precious instrument time. Similarly,
a transparent interface to the instrument's own software can allow
novice users to plan their experiments through simulation and
make more efficient and effective use of their machine time. The
X-Window port must be quick enough to allow users to set up their
simulations, and the simulation results must be quickly transmitted
to the remote user for visualization and iterative analysis. Remote
simulation requires the same high data transfer speed and Quality
of Service guarantees as remote operation
NHMFL: Magnetic Resonance Imaging and Spectroscopy
- The NHMFL and members of the
Florida State University Chemistry Department are involved in
the Florida Resonance Imaging Network, a consortium developed
around the NHMFL imaging facilities located at UF and are part
of an Imaging Science and Technology Center under development
at UF. This consortium will tie together the three teaching hospitals
at University of Florida, University of Miami and University of
South Florida. The center will focus on the development and clinical
applications of advanced magnetic imaging. Its activities will
include research in image recognition, clinical protocols, image
improvement, and the application of spectroscopy in addition to
imaging. High-speed data transmission will enable real-time medical
image processing from remote sites.
NHMFL: Resistive Magnet Research - At
the moment, the remote visualization and collaboration environment
for users of the general purpose high field magnets is limited
to remote viewing of the data acquisition screen and audio conference
calls over regular phone lines. The available transmission speed
is not enough to allow a remote expert to observe sample lead
attachments via a video camera focused through a microscope. A
significant enhancement will occur with Quicktime, Mbone, or similar
products' ability to provide audio and visual information that
is simultaneous with the data acquisition. This will enable collaboration
by one or more remote researchers with the on-site experimenters.
Magnet time will be used more effectively because the experienced
Principal Investigator will be virtually present to guide an inexperienced
graduate student or post-doctoral fellow to solve problems quickly
and take data efficiently. It will save the PIs' time and travel
funds by allowing them to be at their home institutions when their
physical presence at the Magnet Lab is not needed. It will also
allow those who chose to be physically present for their high
field research to return virtually to their home labs when necessary.
The software and hardware for such collaboration is available
and inexpensive. The NHMFL regularly has four visiting research
groups working at the same time. Each one of these represents
two high-bandwidth real-time connections. A minimum of 10 Mbps
will be required for each of these four collaborations-a significant
fraction of our current internal capacity and much more than is
available remotely. Adequate bandwidth to support full-motion
video is critical to making this work and, while a latency of
several minutes could be tolerated in establishing a connection,
a guaranteed Quality of Service will be needed for consultation
on fine tuning the experiment.
Global Weather Forecasting
FSU has been involved in research on numerical weather prediction of hurricanes since around 1980. We have received major recognition for our research efforts. There are currently a large group of post-doctoral candidates and students involved in this activity. FSU scientists are engaged in a major study of recent hurricanes that have impacted southeastern U. S. and the Gulf of Mexico coast. The investigators have been experimenting with the highest resolution global numerical weather prediction model in the world. This experimentation is aimed at improving the prediction of the life cycle of hurricanes. This work is focused on developing a real-time hurricane prediction center at FSU.
Meteorological applications have always demanded high-bandwidth transfer of data sets from around the world. The need in real-time hurricane forecasting is to collect and transfer large data sets in a timely manner from the National Center for Atmospheric Research (NCAR, Boulder, CO), the National Meteorological Center (Washington, DC) and European agencies. At present, only NCAR is on the vBNS. These data sets range from a compressed size of 10 MB to several GB. The transfer of data files of these sizes can take several hours, especially during peak network usage (daytime transfers). To perform timely real-time forecasts, these diverse data products need to be assimilated efficiently into experimental models. Once this is achieved, forecast applications can be completed and final forecasts produced. When finished, forecast data are analyzed and subjective forecast products are then transferred to interested parties.
Ongoing meteorological research on supercomputers here in the United States and abroad also produces large data sets that are needed locally for assimilation, diagnostics and graphical applications, including 3-D animation. Hence, large data transfers are necessary for this purpose as well. Also, new remote-sensing platforms, especially satellite, continue to better resolve and provide more information, therefore greatly increasing the amount of data needed to be processed and transferred. This need will only increase with time as higher performance computing allows greater capabilities and increased model resolution.
High-bandwidth is essential to accomplish
these transfers in a timely manner and support the above meteorological
research tasks.
Key Participant: Tiruvalam N. Krishnamurti,
the Robert O. Lawton Distinguished Professor, Department of Meteorology;
Director, Cooperative Institute for Tropical Meteorology
High Energy Particle Physics (HEPP)
High Energy Particle Experimental Physics Research at FSU is done by physicists in the Physics Department and in the Supercomputer Computations Research Institute (SCRI). The major research centers at which the experiments are done are located at national or international laboratories, such as, Fermilab near Chicago, Illinois. and CERN near Geneva, Switzerland. Now most of the network traffic travels via the U.S. Department of Energy's ESNET. Traffic to non-ESNET sites are through the commodity Internet, where bottlenecks exist. SCRI was one of the original sites connected to the first trans-Atlantic satellite links for HEPP with CERN.
Key Participants: Vasken Hagopian, Professor, Department of Physics; Harrison B. Prosper, Associate Professor, Department of Physics; Saul Youssef, Research Scientist, Supercomputer Computations Research Institute
HEPP: Remote Data Bases - In a typical experiment, many terabytes of data are taken and stored on the computers at the national laboratories from where they are accessed by university groups that ran the experiment. Much of the initial processing of these data is done at the national laboratories. However, in addition to data acquired directly from experiments, data are also generated from extensive computer simulations of high energy particle collisions, and the evolution of the remnant particles of the interactions that were recorded in the apparatus. Simulations are done both at the national labs and at the home universities.
The Florida State University is a major center for this work for the D-zero Fermilab collaboration. At present large data-sets are moved between sites using magnetic media. This impedes the ability to respond quickly to new developments. Typically, when a hot topic is under study, rather than copy data, the investigators work via remote log-on sessions on the computers on which the data of interest currently resides. Unfortunately, this ties up computing resources at that site, while leaving local machines relatively idle. The availability of much higher bandwidths will allow making better use of highly distributed computing.
Ideally, data should never have to be copied. They should remain on whatever computer system was the primary processing site. For example, raw and processed data (processed to a form suitable for analysis) will remain, say, at the national laboratories and simulated data will reside at university sites. Every site will be responsible for keeping their data current. A distributed database approach accessible transparently as a single unified database from any site is needed. Local resources will be used to analyze data locally without regard to its physical location. The network must be sufficiently fast that the investigators could work with the data as if all were on a local computer. Of course, data that are used frequently will be cached. And it could be arranged for the cache to be updated only if more recent versions of the data existed on the remote site.
Another aspect of processing remote data involves
graphic images. The High Energy Group has sophisticated graphics
workstations that are well suited to the detailed study of high
energy events. It should be possible to view and manipulate these
events locally, using the data provided remotely; and to make
the graphics images available to colleagues at other sites. The
slowdown of the present network limits the speed with which analyses
can be performed. The need to constantly copy data to local machines,
in order to remain current, is a time consuming task that impedes
the research effort.
HEPP: Remote Collaboration and Operation of Facilities - In high energy physics, videoconferencing is an integral part of the data-analysis by which a group of widely dispersed collaborating physicists discuss an analysis of a given data-set. Breakdowns in video-links or video-links of poor quality have on many occasions caused misunderstandings and considerable delays. A high quality, reliable, video conference capability, including desktop-to-desktop links (e.g., Mbone) will reap enormous benefits. The chief benefit is better, timely and on-demand communication. The aim should be to mimic the kind of intense personal interaction between individuals physically present with each other that spawns great ideas.
Future experiments are being designed with the expectation that physicists from remote places will be able to take shifts via the network. An experiment typically requires peak data-taking rates of about 20 MBytes/s. Therefore, during the data-taking phase, frequently, raw on-line data needs to be transmitted to remote institutions for urgent on-line analysis, otherwise a broken component of the experiment may not be known until it is too late and major portions of the running time will be wasted. To do this-assuming 10% sampling-the experiments require data transmission rates of about 2 Mbytes/s. At the present time, this diagnostic function can be performed only by traveling to the national labs and being physically present for shifts. However, University teaching commitments place severe constraints on travel. Since the investigators cannot travel much more than they currently do, they must rely ever more heavily on the network.
High bandwidth with low latency is needed
to support full-motion videoconferencing and rapid experimental
sampling analysis.
High-performance Computing and Simulation (HCS) Research Laboratory
The High-performance Computing and Simulation
(HCS) Research Laboratory is located at the University of Florida
in Gainesville with a satellite facility at the FSU-FAMU College
of Engineering in Tallahassee. As the center for computer engineering
research in high-performance computing in North Florida, the HCS
Research Lab is conducting a number of research projects that
hope to exploit the benefits of a direct high-speed link between
UF and FSU. These projects include: 1) the scaleable cluster architecture
latency-hiding environment; and 2) parallel programming and cluster
architectures for heterogeneous computing. These projects can
only be successful if the needed high bandwidth for fast message-passing
and guaranteed Quality of Service are available.
Key Participants: Alan D. George, Associate
Professor, Dept. of Electrical Engineering; Professor, Dept. of
Electrical Engineering; David C. Kuncicky, Associate in Dept.
of Electrical Engineering
HCS: Scaleable Cluster Architecture Latency-hiding Environment (SCALE) - The goal of the SCALE research project is to investigate and develop a family of techniques by which cluster architectures can be constructed with latency-hiding, high-level parallel programming and coordination languages, and low-level lightweight communication protocols in a fashion that is open, portable, high-performance, distributed, fault-tolerant, and scaleable across different interconnects supporting Uniform-Memory-Access (UMA), Non-UMA (NUMA), and Cache-Coherent-NUMA (CC-NUMA) shared memory as well as (and in tandem with) message passing. These scaleable cluster architectures leverage the constantly rising levels of performance and dropping costs in Commercial-Off-The-Shelf (COTS) hardware including state-of-the-art workstations, high-performance interconnects, adapters, switches, single-board computers, etc.
The parallel architecture of the SCALE system will be constructed by combining high-performance interconnects in a multilevel fashion. The scaleable topology proposed is a cluster that starts with shared-bus, shared-memory Symmetric MultiProcessors (SMPs) with UMA. These SMPs are combined via Scaleable Coherent Interfaces (SCI) with NUMA and CC-NUMA across local-area distances, and then clusters of these SCI-based multiprocessors are combined via ATM across metropolitan-area and wide-area distances. The software aspects of the SCALE system will emphasize latency-hiding mechanisms including distributed-directory cache coherence, instruction and data prefetching, relaxed memory consistency, shared virtual memory, and multiple contexts via multithreading. High-level parallel programming and coordination language implementations will be developed to exploit latency-hiding mechanisms to achieve high-performance and scalability across multiple tiers in the SCALE system, such as multithreaded MPI and Linda. Low-level, lightweight communication protocols will be developed to exploit the low-latency and high-throughput potential of the underlying high-performance interconnects. This approach will achieve flexibility and portability without the performance drain associated with TCP/IP, such as with the first multithreaded implementations of HCS_LIB (a lightweight parallel communications protocol developed by the HCS Lab) and Active Messages (originally from UC Berkeley) for SCI and ATM/SONET. While SCI and ATM are highlighted as target interconnects for this project demonstration, and in some sense represent opposite ends of the high-performance interconnect spectrum, systems constructed with many other high-performance interconnects can also be addressed with this technology including Fibre Channel, Gigabit Ethernet, HIPPI, Myranet, etc.
The SCALE project is designed to bridge the
gap between the research worlds of parallel computing and high-performance
interconnects by employing concurrent research methods that are
both theoretical and experimental in nature. Future high-performance
computing systems, for both general-purpose and embedded applications,
must be "able": scaleable, portable, dependable, programmable,
and affordable. The SCALE project will contribute to this end.
HCS: Parallel Programming and Cluster Architectures for Heterogeneous Computing - By using different types of parallel processors, processing elements, and interconnection paradigms, heterogeneous computing has the potential to maximize performance and cost-effectiveness for a wide variety of challenging scientific computing problems. Such applications typically have computation and communication requirements that not only vary between applications but dynamically within applications. The specialist parallelism approach of heterogeneous computing is particularly attractive, where the processors and computers are scheduled for subtasks of the application based on their fit to the requirements of the subtask instead of merely using load balancing to spread out the problem evenly or proportionately. For example, those aspects of a program that are inherently vector processing oriented will be mapped to a vector machine, those that are data parallel in nature might be mapped to a Single Instruction Multiple Data (SIMD) machine, etc. The suitability of the tasks must first be evaluated, and only then is load balancing used among selected machines for the final assignment based on these suitabilities. Research topics in heterogeneous computing are diverse and involve issues related to connectivity, bandwidth, granularity, high-level orchestration tools, and programming paradigms.
The objective of this project is to study the use of emerging
parallel programming paradigms and workstation cluster architectures
as enabling technologies for future scheduling extensions in heterogeneous
computing. In particular, with the emergence of ATM and others,
the interconnects and networks that form the backbone of the cluster
concept hold the potential to revolutionize the connectivity,
bandwidth, and granularity of LAN- and WAN-based Virtual Heterogeneous
Machines (VHMs) and environments. Concomitant with this emergence
is the increasing need for lightweight protocols to support these
interconnects and networks so that the software bottleneck associated
with interprocessor communication does not continue its trend
of being the most contributing factor in interconnect latency.
Finally, in order to improve fault tolerance and software development
capabilities of VHMs, the impact of parallel programming paradigms
such as Linda/Piranha and ISIS may play a pivotal role in concert
with these lightweight protocols.
Human Brain Project
This project will benefit greatly from a high
bandwidth Internet connection. The PI of this team is Dr. David
Rottenberg, head of Neurology at the VA Medical Center in Minneapolis.
The project "Spatial and Temporal Patterns in Functional
Neuroimaging" has received a 5-year NIH grant (1P20-MH 57180-01).
Research on this project is in visualization of brain datasets
(whole-brain MRI, PET and fMRI scans). These large (10MB) datasets
are transferred and worked on interactively among team members
in Tallahassee, Akita-Japan, Copenhagen-Denmark, Minneapolis,
Chicago and Boston. The investigators plan to use the Internet
as a communication medium as much as possible. (Two of the collaborating
sites are unlikely to be part of vBNS/Internet2 in the foreseeable
future.) A high bandwidth will support interactive use of remote
visualization computers and software and to exchange these large
datasets in real-time during peak hours. Such frequent uploads
and downloads now take on the order of minutes, which makes interactive
visualization among collaborators extremely difficult. For example,
one part of the project (based in Copenhagen) involves the use
of VRML techniques to interactively view and manipulate (slice,
highlight, etc.) whole-brain datasets in real-time. This data
exploration is currently impossible to do remotely with current
Internet technology.
Key Participant: De Witt L. Sumners, Professor,
Department of Mathematics
EHEP: CMS Regional Computing Center
The Experimental High Energy Physics (EHEP) group at the Supercomputer Computations Research Institute (SCRI) consists of 7 Ph.D. physicists, doing research at the European Laboratory for Particle Physics (CERN) in Geneva, Switzerland, and at Fermilab in Batavia, Illinois. The work of this group is responsible for the multiple T1 lines of ESNET connectivity currently used by Florida State University. FSU has been connected to ESNET since the days when it was 56 Kbps circuits and named MFEnet.
The EHEP group at SCRI is a member of the ALEPH and CMS collaborations at CERN and the D0 collaboration at Fermilab. SCRI is a major site for computing for all these experiments, acting in a sense as a remote computing center. The EHEP group at SCRI provides computing and infrastructure support remotely to its CMS collaborators in the US and worldwide. It is responsible for the maintenance and distribution of the Windows/NT version of the CMS software, and the local NT cluster is heavily used for simulations of the CMS detector. Good network connectivity is important for two-way access to simulated data sets.
The data access requirements of CMS are unique, and have been identified as the most difficult aspect of the CMS computing problem. Expectations are for more than 3 PetaBytes/year written to tape and accessed by geographically dispersed physicists. The data will be treated as a single logical entity but will reside on disk and tape at various physical locations (federation of Object databases). These locations will be in the form of regional computing centers connected by high-speed links to universities collaborating in these experiments. SCRI is one of the few candidates that currently qualify for a regional computing center. Since the NSF is funding a substantial part of the US CMS project and participating University groups, connectivity to SCRI via the vBNS will be particularly appropriate.
Key Participants: Christos Georgiopoulos, Scientist, SCRI; Martyn Corden, Scientist, SCRI
Digital Medical Imaging Program
College of Medicine
The Digital Medical Imaging Program at USF is nationally recognized for its achievements in medical image analysis, in particular in the fields of brain tumor MRI, nuclear imaging and mammography. In mammography, DMIP has focused its efforts on image segmentation and mammography (fewer missed cancers) and specificity (fewer unnecessary biopsies). Also currently under development are image interpretation algorithms for the early detection of lung cancer in chest xrays. Evaluation of the developed computer techniques requires centralized access to large numbers of images, generated in multiple medical centers. The DMIP currently has research collaborations with the Uniformed University of the Health Sciences located at the Naval Hospital in Bethesda, the University of South Carolina, the NASA SETI group and the University of Pennsylvania, and is planning expansion of the collaborations. Fast exchange of data between the originating sites and the research site is vital for the success of the project. A mammogram examination generates approximately 200 MBytes of image data, necessitating high bandwidth networking. Chest xrays may be 120 MBytes. Both image types are produced in large numbers as they allow screening of the population for early signs of cancer. Other areas where high bandwidth is required are: 1) Mammography computer algorithms provide an important tool in the teaching of mammography interpretation, which due to the size of the images and the difficulty of obtaining ground truth is best organized centrally with the opportunity for sites nationwide to interact with the teaching files via the vBNS. 2) Similarly, the image data base of DMIP may provide an important resource nationwide for researchers in medical image processing. In the past, a small database with lowresolution mammograms has been distributed by USF on tape. High resolution images with ground truth may be distributed using the vBNS. Direct digital xray systems are now commercially available and are likely to change the way in which radiology is practiced. Computer assisted methods will be a vital part of these developments, for diagnosis (image analysis), storage and transmission (compression), and retrieval (by a wide variety of image features). All these areas are addressed by the research at USF and its collaborators at other institutions.
Key Participants: Robert
Velthuizen, L.P. Clarke
Marine ScienceRemote Sensing Facility
Department of Marine Sciences
The Remote Sensing Facility requires higher network speeds to obtain and distribute data sets collected by satelliteborne remote sensors. These projects, which are funded by NSF, NASA, and IMMS, are 1) the CARIACO Program, which establishes and maintains a time series of in situ observations in the Southern Caribbean Sea of wind, current, seasurface temperature, and ocean color satellite data sets that are only obtainable via network links to federal laboratories or other universities. 2) MMS (Dept. of the Interior's Minerals Management Services) sponsored program to analyze satellite data covering the Gulf of Mexico and the West Florida Shelf. The MMS program involves transfer and redistribution of ocean topography, sea surface temperature and ocean color data sets obtained daily from satellites. In the next 5 years the application will require the transfer 10 Gbytes/day of satellite data from the NASA Goddard Space Center.
Key Participant: Frank E. Muller-Karger
Next Generation Video Compression
College of Engineering
USF is developing a next generation video compression algorithm based upon the concept of "multicandidate block motion estimation", and the use of a modified Hopfield Neural Network. This project seeks to use the vBNS as a test vehicle for the compression algorithms with the goal of providing a comprehensive set of tools for video compression both to (a) the user community (including industrial, academic and government), and to (b) the research community for further advances in the field on a 'remotemulticollaborative' basis.
The bit rates required from the vBNS for this endeavor range from a few Mbps for some applications to 1.2 Gbps for HDTV quality. These are the rates for downstream, i.e., to USF, or more generally to any compression site. Of course, what is downstream for the compression site will be upstream for a user. The upstream rates (for the compression site) could be a factor of 10 to 500 lower, depending upon the particular compression goals. In a multicollaborative situation, however, each site will need to support the higher bit rates (10 Mbps to 1.2 Gbps) both for upstream and downstream-since each site could be a compression center with its own unique experimental features as well as a user.
Key Participant: V. K. Jain
MHLP Data Center
Louis de la Parte Florida Mental Health Institute
The Department of Mental Health Law and Policy in the Louis de la Parte Florida Mental Health Institute has established the MHLP Data Center to improve the quality of life for mentally disabled people and the efficiency and effectiveness of the services provided to them, by conducting appropriate and timely research examining the systemwide dynamics of existing services utilization, the related outcomes of those dynamics, and the impact of service system changes on outcomes and utilization. This research requires the compilation, management and examination of large secondary administrative data sets containing information such as personlevel service claims data and facilitylevel survey/reports data. The initiation of deinstitutionalization of persons with mental illness have forced public and private mental health systems to create and finance appropriate community care alternatives. Continuing changes in the organization and financing of mental health services have made research and planning exceedingly complex. Large and powerful information systems are essential to monitor and manage the service delivery system as well as advise policy makers who are setting the direction of innovation in the system.
The MHLP Data Center is poised to become the primary mental health data repository and manager for the State of Florida in the next year. To realize this objective, the Center will require high-speed FTP capability to acquire data from US Census Bureau and other sites (see collaborative sites listed below) and to provide access to information in our repository for State of Florida and national agencies. The Center's data system currently contains over 20 million records, which requires several gigabytes of disk space. Ultimately these files will contain in excess of 100 million records with estimated sizes in excess of 100 Gigabytes. To compile the data and manage this interactive system effectively will require technologies allowing high-speed longdistance interactive data analysis/reporting, and computerized conferencing. The MHLP Data Center currently has research collaborations with the Center for Mental Health Policy and Services Research at the University of Pennsylvania, the Sheps Center for Health Services Research at the University of North Carolina, the Oregon Health Sciences University, and the Department of Psychology at the University of Hawaii, all of which are Internet2 universities. The MHLP Data Center also anticipates expanded collaborations with researchers at the Department of Social Work at the University of Washington, and with the National Association of State Mental Health Program Directors Research Institute in Alexandria, Virginia. Data exchange via high-speed FTP with these sites and the integration of these data must occur at levels impossible today in order for the Data Center to effectively realize its mission of informing physical and mental health care research and reform.
Key Participants: Paul G. Stiles,
J.D., Ph.D., William D. Kearns, Ph.D.
Ocean Optics, ECOS Modeling, Ocean Modeling Projects
Department of Marine Sciences
COBOP Project: The Coastal Benthic Optical Properties Project (COBOP - funded by ONR) involves the creation of an interactive database through transmission of real-time data and video from our Autonomous Underwater Vehicle (AUV) and Remote Operated Vehicle (ROV). The AUV project will require vBNS to implement remote control of the vehicle and instrument systems, real-time data downloading, and visualization of images. Remote control of the ROV will allow researchers at ONR or other vBNS sites to direct the cameras on the ROV to examine and record data from different targets. The vBNS will also enable collection and distribution of Physical Oceanographic Real-time System of data buoy and towers around Tampa Bay and the NOAA oceanographic data buoys in the Florida Keys.
Key Participant: Kendall Carder
The ECOS Modeling project requires vBNS bandwidth to transmit satellite data to initialize numerical models on supercomputers at the NASA Goddard Space Flight Center and Florida State University in Tallahassee. The resulting numerical models are quite large and must be retrieved and transferred to USF either for visualization, or downloaded for additional analyses on local machines. The ECOS facility is currently supported by NASA and NSF to model the Caribbean Sea, Gulf of Mexico, North Sea, the Arctic Ocean, and the Southern Ocean.
Key Participant: John J. Walsh
The Ocean Modeling and Prediction project is currently engaged in an the evaluation of Mixed Layer Dynamics and Upwelling in the Arabian Sea. This project, funded by the Office of Naval Research, requires large numerical model integrations to be run on remote supercomputers at the Naval Oceanographic Office at Stennis Space Center, MS, and rapidly transferred to USF. A second project underway is the development of an Integrated EndtoEnd Marine Contaminant Information Management System, funded by the Environmental Protection Agency. This project involves the combination of realtime oceanographic data, a numerical ocean circulation model, a contaminant spill trajectory model, and a GIS database and is sponsored by the Florida Department of Environmental Protection (FDEP) and the NOAA National Ocean Service (NOS). The DMS seeks to link its computing facilities to oceanographic databases located at NOS in Silver Spring, MD; the large size of downloads from these databases will necessitate vBNS connectivity. A third project underway is HydroWire, an Electronic Information Service for the Marine and Aquatic Sciences: This WorldWideWebbased newsletter for the international ocean science community, with sponsorship of the four primary scientific societies, is very graphicsintensive.
Key Participant: Mark Luther
Digital Libraries Project
USF Libraries
The USF Libraries' Digital Libraries Project seeks connectivity to the vBNS to enable a new generation of applications in the area of media integration. The considerable bandwidth requirements, especially for projects employing video and audio materials, make transmission of these materials over the commodity Internet impossible.
Each project will be available to the international research community through a WWW browser. Specific projects will be cataloged in SGML and will provide hyperlinks, metadata, and indexing through a number of access points. Standards and protocols will include SGML, Z39.50, HTTP, and TCP/IP. There are servers at each of the libraries to support multimedia projects, and there are editing suites for audio and video supporting a variety of formats and enhanced video productions. The Florida Center for Library Automation at the University of Florida may serve as an archiving facility for these projects:
Dion Boucicault Theater Collection (actor, playwright, stage director, manager, producer). Storage Requirement: 1.34GB;
The Florida Sentinel Bulletin (has been the voice of the AfricanAmerican community in Tampa from 1945 to the present). Storage Requirement: 15GB;
Ringling Collection (historically significant estate of Charles Ringling). Storage Requirements: 1 GB;
USF electronic dissertations and theses. Storage Requirement: Requirements: 2.0 GB, annually;
Florida Slave Narratives (containing interviews conducted some 70 years after slavery ended). Storage Requirement: 700MB;
John. C. Briggs Collection (focuses on ichthyology and natural history illustration 3,000 items). Storage Requirement: 6,950MB;
Louis de la Parte Florida Mental Health Institute Collection (consists of multimedia programs and keynote addresses from the Institute's national and state conferences). Storage requirement: 540GB;
Harry T. Moore Collection (focuses on the FBI files and related materials concerning the bombing and subsequent deaths of Harry T. Moore and his wife Harriet). Storage requirement: 263 MB;
History of the Written Word Collection (includes pages from medieval manuscripts and similar examples of early written and printed items). Storage requirements: 30 GB
Poynter Collection (the papers of the longtime owner and editor of the St. Petersburg Times (Florida), founder of the Poynter Institute for Media Studies, and founder of Congressional Quarterly). Storage requirements: 5500 MB;
Jacobs Collection (consists of photographs of historic St. Petersburg, Florida during the 1920's and an extensive collection of oral history interviews). Storage Requirement: 15.6 GB.
Key Participants: Samuel Fustukjian, Ardis Hanson, Landon Greaves, Beverly Shattuck, and Joan Pelland.
Each is an active member of the USF Digital Libraries Workgroup that is dedicated to increasing the amount and quality of library materials available on-line. The results of this group's efforts may be viewed at the following web sites:
http://www.lib.usf.edu/ Tampa Campus Main Library
http://www.nelson.usf.edu/ St. Petersburg Campus Library
http://www.fmhi.usf.edu/library/newest.html FMHI Research Library
http://www.med.usf.edu/HSC/index.html Health Sciences Center Library
http://www.sar.usf.edu/~library/ Sarasota
Campus Main Library
MathematicPhysical Engine Project
College of Engineering
The MathematicPhysical Engine (MPE) Project requires access to supercomputing facilities using parallel processing techniques to solve and display in real time the solution of sets of ordinary or partial differential equations. The objective of the project is to comprehensively model various physical phenomena, ranging from fluid flow to electromagentic field dynamics, thermal patterns inside a semiconductor wafer, thermal patterns on a geophysical scale, combustion inside an automotive cylinder, and ion transport across a biological membrane. Different versions of the MPE range from a single commercial parallel machine, to a stackedwafer dedicated machine, to networked supercomputers. Access to expensive remote MPE supercomputing facilities is presently hindered by network bandwidth restrictions. vBNS connectivity will support parallel processing among supercomputing facilities engaged in MPE research and enhance the ability of scientists and engineers to accelerate basic research on the one hand, and product development on the other.
Key Participant: A. D. Snider
Efficient Operation of the Internet2Measurement and Reduction of Traffic SelfSimilarity
College of Engineering
Traffic in highspeed networks is bursty and selfsimilar. The superposition of selfsimilar traffic streams does not result in statistically smooth traffic, but results in an even burstier traffic stream. Uncontrolled bursts cause packet or cell loss at intermediate nodes (e.g., in ATM switches) and result in network degradation. Highspeed networks, such as Internet2, are expected to transport multiple, aggregated traffic streams and the measurement of traffic selfsimilarity on the Internet2 is the primary objective of this research. The secondary objective is the reduction of selfsimilarity via sourcelevel traffic shaping. Internet2, as a highspeed operational network carrying many different types of traffic, provides a unique and invaluable facility for this research. Characteristics of selfsimilarity, or long range dependence (LRD), have been measured in many types of traffic and networks. Thus, it is likely that the traffic on Internet2 will also be LRD and thus problematic to control. Unlike slowerspeed networks, traffic measurements on Internet2 cannot assume that statistics on the full population of packets can be made. Instead, methods of traffic sampling must be developed. Existing works in traffic sampling do not validate sampling methods for measuring characteristics of LRD. In this research, traffic sampling methods based on an adaptive sampling scheme will be developed and validated for measuring LRD characteristics. To reduce traffic selfsimilarity and the associated ATM cell loss at intermediate nodes, selected traffic bursts are shaped at the source. This research will explore the applicability of these new methods, in concert with existing control mechanisms, to insure the efficient operation of Internet2. In summary, this research proposes to develop methods to measure and control traffic to insure efficient operation of Internet2. Internet2 is the only network that will provide 1) the high speed, and 2) the large mix of traffic types necessary for this research. The results from this research will extend beyond Internet2.
Key Participant: Ken Christensen
Real-Time, Reliable, Multicast Applications
A fundamental communication mechanism for distributed applications is multicast addressing. An example of a Large-scale, Multicast Applications (LSMA) is Distributed Interactive Simulation. DIS began with the development of the Simulation Networking (SIMNET) project funded by DARPA, and has been an avid topic of research for over 10 years. The proprietary SIMNET solution soon led to the creation of a standards development organization and process that in turn developed the IEEE 1278 standards. DIS has since evolved into both an architecture and a protocol.
Historically, these Virtual Reality (VR) simulations run over a local internet subnet using subnet broadcasting. While the appropriate use of subnet broadcasting by DIS is arguable, one conclusion was evident: broadcast messaging was limiting the scaleability of the overall simulation exercise. Simulation hosts were responsible for reacting to each and every protocol data unit (PDU) transmitted. A broadcast communication mechanism was not scaleable because the number of PDUs transmitted grew more than linearly with the number of entities being entertained by a given simulation exercise. In short, DIS scaled at best, O(N2).
By itself, static multicast addressing, in lieu of subnet broadcasting, is only marginally successful. Simple address-based filtering using static multicast addressing showed limited results because the underlying architecture was principally unicast-based. Likewise, the DIS "heart beat" or replay of static entity state information (used to reset dead reckoning algorithms) was also not scaleable and furthermore resulted in additional increased congestion. A new methodology will be needed to resolve LSMA operational issues.
This new methodology, recently adopted by the U. S. Department of Defense (DoD) Defense Modeling and Simulation Office (DMSO), is called the High Level Architecture (HLA). HLA provides a better communication mechanism than either SIMNET or DIS in that only data that has changed is transmitted. Furthermore, data are distributed using a dynamic Data Distribution Mechanism (DDM) that employs dynamic multicasting and could benefit from resource reservation in order to ensure real-time transmissions.
However, the adoption of new methodology is only as good as the underlying infrastructure's ability to support it. In the case of HLA, new demands are being placed on the routing and switching fabric of the Internet. LSMAs now require that thousands, even tens of thousands of simultaneous multicast groups be utilized. Furthermore, these groups need to change at a rate of 1-5% in under 0.5 seconds. Similarly, these group changes need to be propagated end-to-end in order to be useful across the Internet.
Bandwidth reservation is also needed in order to ensure that specific data rates for some data can be achieved. In order to reserve a flow between two hosts today, either specific networking technology (i.e., ATM) or experimental software (i.e., RSVP) needs to be employed. Switched Virtual Circuits (SVCs) are not available today. Nor is a standard RSVP transport protocol available within the Internet Protocol Suite. Existing implementations utilize lengthy setup and tear-down times, which are too highly latent to support the dynamics of real-time LSMAs.
To support LSMA research, the University of Central Florida requires the availability of a high-speed backbone testbed to continue its research in the development of LSMAs. Specifically, this application intends to study multicast communications, bandwidth reservation, and the dynamics and aggregation of both in light of a large-scale, real-time simulation event. These Virtual Reality applications require both reliable and best-effort delivery of digital data, as well as the occasional use of streaming audio and video data.
The existing Commodity Internet does not support native end-to-end
multicast communications, flow reservation, or the real-time transfer
of reliable data. This applications research area therefore describes
not the environment of a specific VR application, but rather the
underlying network's ability to support this combination of such
requests. This is currently an active area of research by UCF/IST
personnel, the Internet Engineering Task Force, and the Reliable
Multicast Research Group of the Internet Research Task Force.
Key Participant: Dr. Michael Myjak, Senior Research Scientist,
Institute for Simulation and Training
Virtual Reality Tool for 3-D Dynamic Anatomy
A connection to the vBNS is required to support existing collaborations with other research institutions, specifically the University of North Carolina at Chapel Hill (UNC-CH) and the University of Michigan in Lansing. The common application is that of 3-D scientific visualization using various 3-D displays and visualization platforms such as the mirage system (developed at the Institute for Simulation and Training at the University of Central Florida), head-mounted displays located in the Vision, Graphics, and Image Laboratory (VGILab) of CREOL, and in the future cave-like platforms using 3-D color displays.
One specific application in development at this time is that of the design and testing of a virtual reality tool (VRDA tool) for teaching 3-D dynamic anatomy. This project is supported in part by the Office of Naval Research and by the National Institutes of Health. A First Award from NIH to Dr. Rolland will start effectively in July 1997 and will continue for the next 5 years.
Dr. Donna Wright from UNC-CH, co-founder of the project will provide medical knowledge to the development of the tool, as well as Dr. Winfield, Director of the Arthritis Center at UNC-CH. It will be cost- and time-effective to be able to use the vBNS connection to have them visualize the anatomy animation in real-time rather than having them travel to UCF. Especially, it will allow closer collaboration during each stage of the development with increased feedback on progress.
Dr. Frank Biocca, Ameritech Professor of Telecommunication and director of the M.I.N.D. laboratory at the University of Michigan Lansing will assist in the testing of the tool developed in the VGILab. A vBNS connection will allow direct access from the M.I.N.D. Lab to the animation done in the VGILab. Because of bandwidth requirements, it is not possible to support the application on the current Internet. The animations are real-time based on optical tracking data. Replication of the work in the M.I.N.D. Lab is not possible at this time since they do not have the optical tracking equipment. So the other option beside the vBNS connection is to transport part of the VGILab to Lansing Michigan, which would interfere with teaching duties and student supervision; or to have Dr. Biocca come to UCF, which will not be possible for him. Moreover, doctors from the medical school on campus in Michigan are more readily available than at UCF and running studies with subjects in Michigan and data at UCF will be highly facilitated by a high-speed network connection.
When Drs. Rolland, Wright, and Biocca first conceived the project the three investigators were at the University of North Carolina at Chapel Hill in the departments of computer science, radiology, and communication, respectively. They have won the NIH award but are now located at three different sites. Access to the vBNS link is required to supply the high bandwidth data flow needed to support the continuation of this collaboration among UCF, UNC Chapel Hill, and the University of Michigan.
Key Participant: Dr. Jannick Rolland. Center for Research and
Education in Optics and Lasers (CREOL)
Distributed Immersive Environments - Virtual People
Today the client-server model prevails over Internet2. Client application to and server application interactions operate over low speed, connection-oriented circuits supporting variably latent, albeit reliable, connections. However, human-to-application and eventually Human-to-Human interaction will require support by true distributed immersive environment applications. These applications will continue to evolve over the next decade as they begin to take full advantage of Internet2, eventually becoming synonymous with the Next Generation Internet.
Distributed Immersive Environment applications will provide a new metaphor by which people will interact with one another. Ultimately, this will change our society from what it is today. These environments utilize high-speed multi-channel communications links requiring a continuously variable array of Quality of Service support. This support will come as shared, distributed middleware, relieving the application from requesting specific attributes like high-bandwidth or low-latency links between peering hosts. Relieved of this burden, distributed applications will focus on higher level interactions with people. Eventually, this will become the norm, and not the exception among NGI applications.
Distributed Immersive Environments promise to change the way people live, interact, learn, work and play for the better. Geographically dispersed locations will become closer than ever before provided by airplane. Web-based interactions will, for example, manipulate multiple profiles for an individual, emphasizing his or her habits, and representing one's likes and dislikes as he or she interacts with others through cyberspace. The result of this data fusion is that to the casual observer, distributed assemblages of the sort cited above will one day become commonplace.
The technologies are already available today to construct distributed interactive simulations, electro-optics, digital signal and image processing, and immersive environments. These technologies will conjointly evolve into a new communication architecture, utilizing reliable, high-bandwidth, low-latency network technology of tomorrow.
In this example, the visual display model is generated from a
reflection of the real subject integrated with previously stored
construction information. This aggregated information is then
communicated over the WAN space and eventually displayed using
pseudo and/or holographic projection systems. Of course, sound
and video are already capable of being conveyed over great distances
with the technology of the Commodity Internet today. Even then,
video and audio traffic occur only in limited, and predominately
experimental use. However, the conveyance of tactile sensations,
three-dimensional sounds and images, possibly as holograms, interacting
over the next generation internet in real-time, and between multiple
sites is another matter. The availability of high-speed, quality
network services is absolutely necessary to accomplish this.
Key Participant: Dr. Michael Myjak, Senior Research Scientist,
Institute for Simulation and Training
Live-Fire Instrumentation Using Internet2 QoS
From a military briefing, to a high technology live-fire demonstration, to final disclosure and debriefing, short bursts of data are generated, recorded, processed, analyzed and ultimately reported. During this time, the data are highly susceptible to various sources of corruption, contamination or loss. Perhaps the most detrimental of these to simulated live-fire testing is the variance and jitter witnessed in end-to-end latency. Perceived latency between any two devices can be caused by a variety of factors, including complicated irregularities in instrumentation to errors in simple timing synchronization. In order to create and establish a distributed