Please consider the following response to your RFP for INTERNET2 PKI LABS. The enclosed document in Microsoft Word 97-89 format is a formatted version of the ASCII text below. Thank you, Robert Brentrup Dartmouth College voice: (603)646-1413 fax: (603)646-2810 e-mail: Robert. J. Brentrup@dartmouth.edu =============================== Proposal to Establish Internet2 PKI Lab at Dartmouth College David Nicol, Sean Smith, Edward Feustel, Robert Brentrup, William Taylor Dartmouth College and its new Institute for Security Technology Studies (ISTS) are pleased to propose the establishment of an Internet2 PKI lab at Dartmouth College. Why Dartmouth? ------------- At Dartmouth College, the Computing Services group, the Computer Science Department and the Institute for Security Technology Studies are pursuing a research agenda identifying and addressing the security and privacy issues in the current and emerging computational infrastructure. Together we make a compelling site for the proposed PKILab. By way of background, our Computing Services team (an Internet2 Middleware Early Adopter) has already deployed a limited PKI. It operates a Kerberos infrastructure for security and rights management for a large number of users, spanning organizations as diverse as other universities, medical centers, and a local non-profit ISP. It has implemented a distributed authentication and authorization protocol (IDAP) to share information resources among multiple campuses. Several institutions jointly license database resources that are mounted on servers at Dartmouth College. Many current administrative and library applications requiring authorization are supported at Dartmouth with various combinations of Dartmouth Name Directory, Kerberos and IDAP. It is recognized that the current ad-hoc nature of determining user attributes limits the development of additional applications. We are also finding that commercial groupware applications like calendaring require directory and/or PKI middleware to function. Providing this infrastructure using standard protocols are the objectives of Dartmouth's Early Adopters project. (see: http://www.internet2.edu/middleware/earlyadopters/ ) Dartmouth is home to the new Institute for Security Technology Studies, ISTS. Funded at $15M/year, ISTS brings to the table a unique resident group focused on security technology, as well as existing collaborations with a dozen other leading universities, as well as research labs in and out of government. The proposed PKILab would be leveraged from the user group and practical experience of Dartmouth's Computing Services group, and the resources (technical and administrative) of ISTS, not to mention the benefit of its exposure and potential for application in the on-going ISTS work in diverse core areas such as network security, performance/effectiveness studies of security systems, security models, as well as advanced application areas such as secure multi-media training and incident response. Lastly, Dartmouth's teaching mission provides a sophisticated, questioning user base, to push the limits of what we build and provide requirements for new directions. For all these reasons Dartmouth will be an excellent site for an Internet2 PKILab: our interests already coincide with the research agenda, the PKILab can leverage from a significant existing base of interest and support, and we have the right mix of people, tools, and ideas to carry out a full spectrum of research: from immediate deployment and performance issues, to longer-term speculative work, and everything in between. Dartmouth presents an optimum point in scale as well---it is large enough to be supporting a full range of applications over a very substantive distributed network, but at the same time is small enough for us to be able to do comprehensive controlled experimental studies. Technical Interests ------------------- AUTHORIZATION CERTIFICATES AND DELEGATION are central to the security of distributed computations. Different authorization techniques are feasible, including identity-based, role-based, capability-based, or combinations of these. Applications of workflow and information dissemination management in a distributed environment require allocation of access rights between owner and user. Lampson discusses this delegation of rights in a famous paper. Bull, et.al. have a different view of how this should be done. PKI technology permits generation and distribution of certificates representing identity and authorizations. The challenge is find how best to allocate information to certificates, and what information ought be contained. For example, Dartmouth Computing Services' first PKI based project is an Electronic Payroll Authorization workflow system to support basic Human Resources actions involved in processes to hire, promote, transfer, terminate, and adjust the salary or account attributes of college employees. Establishing an audit trail of electronic signatures by authorized users is key to replacing a paper based system. The Library anticipates applications of PKI authorization in the inter-mediation of access to vendor supplied information products. This would allow revisiting the vendor agreements controlling content and provide the option to use PKI instead of IP address checking. The existence of suitable middleware would also enable the development of a Library Intranet that would adapt its presentation and behavior based on demographic or identity information available from the end user. We would survey current PKI technology, authorization protocol, and API work, concentrating on how identity and authorization certificates might be combined, or used in delegation as described by Lampson or Bull. Open standards will be employed, whenever feasible, and a prototype constructed using facilities provided by the grant. The prototype will leverage existing applications developed by Dartmouth's Computing Services group, and evaluated for inter-operability and scalability within its framework. Certificate management among servers and users will be a key focus. A SECURE COPROCESSOR is a physically secure, general-purpose computer with high- performance cryptography and the ability to prove "it's the real thing, doing the right thing." The IBM 4758 family (see: http://www.research.ibm.com/secure_systems/scop.htm) represents the state of the art in this area, and its security architect just joined the Dartmouth faculty---and is bringing many prototypes and tools with him, as part of a joint research agreement with IBM. The security of any PKI depends on the integrity of the root cryptography and related computation. Hardening the infrastructure by integration with secure coprocessors enables many exciting research directions. - PRIVACY. Private user data and sensitive computation can remain unaccessible even to someone with root privileges on the main server. - SCALABILITY. Rather than confine sensitive computation to a few machines behind locked doors, we can distribute to safe havens throughout the infrastructure, even to potentially adversarial sites. - CONNECTIVITY. This distribution also frees us from always requiring connection to the back-end secure machines. - LEGAL. Federal applications are REQUIRED to use FIPS-validated modules where they exist, and the IBM 4758 is the sole programmable device validated the FIPS 140-1 Level 4 (the highest level of security). For example, consider the problem of certifying user key-pairs for encryption. Students want to share and transfer their data, and sometimes forget their private keys; universities and government sometimes have legitimate need to recover (sometimes with a strong evidentiary trail) certain sets of data. Balancing the legitimate need for data recovery and transfer, with legal and social privacy constraints is a serious issue (and one that perhaps ONLY a university is in a position to explore). Secure coprocessing provides a unique technical slant: we CANNOT see the private keys; only the pre-approved programs can. With secure coprocessing, we can decide and implement (and evolve) policies COMPUTATIONALLY, instead of a priori via the cryptographic protocol. This work is coordinated with the ongoing Packet Vault work funded by ISTS. For another example, consider using short-lived keypairs to avoid CRLs. In a large distributed system, how do we efficiently and securely provide new keypairs? How do we efficiently and transparently achieve continuity of function across the discontinuity of key changes? Distribution of secure coprocessors with high cryptographic throughput throughout the infrastructure enables many possible solutions. Secure coprocessing is a unique and exciting area that offers solutions to problems in manageability, trust models, and scaling. One of our practical objectives will be to investigate how our secure coprocessors integrate with and can enhance the functionality of the PKI software provided by the grant. With the IBM prototypes, we plan to implement and test these approaches (and IBM Research has offered to provide additional support for a broader pilot, should our proposal be funded). BUILDING A SECURE, USABLE BROWSER. To be effective, a security infrastructure needs to give users tools that enable them both to participate in their applications, as well as to easily make reasonable trust judgments about their partners in these interactions. Although current browsers appear to meet the first goal, they fall far short in the second: - Features (such as JavaScript) and implementation flaws continue to render the security GUI meaningless: adversarial web sites can spoof the SSL "lock" and pop-up windows, and can even change the alleged URLs the browser displays.. - Even sophisticated users cannot easily define what they should conclude about the identity (and security attributes) of a server from the existence of the SSL lock. - And, sadly, studies show that most users don't care, and will happily dismiss warning boxes if doing so lets them proceed with their e-commerce purchase. An interesting area of work will be to augment a browser so that the user can clearly, unambiguously, authenticate identity (and attributes) of who they're connected to, without risk of spoofing. (This work also raises the "software engineering" and formal verification aspects of security: why are we going to get it right when the commercial vendors keep getting it wrong?) Given the fact that customers are happy without security, it is arguable that only a university has motivation to carry out this work---and the resulting tool would fill a hole in the current infrastructure, as well as be a building block for future work. PKI TOOLS FOR PERVASIVE COMPUTING. Much current PKI work focuses on people: users have keypairs, and the system needs to authenticate them. However, in emerging world of pervasive computing, the SYSTEM with which the users (and their machines) interact is not only distributed, but has stopped being a monolithic security entity. Conference rooms, libraries, lounges, coffee shops all will have servers and providers of information and computing services. Each of these machines may be owned by a different entity, and may have (or lack) additional security attributes. User machines will interact with this infrastructure (and often this interaction will be transparent). This pervasive computing vision raises serious security and privacy problems. As some of our longer-term research, we hope to help make manifest the necessary infrastructure to solve them. Some elements may include: - a PKI for MACHINES, in order to effectively indicate what entity owns that machine, how exposed the machine is to attack, and what additional security measures (such as integration with a secure coprocessor) may have been taken. (Perhaps this could be based on an identity certificate and a set of attribute certificates.) - On the client end, a way to express, reason about, and implement reasonable policy decisions about each action with each machine. - Potentially, a user interface to clearly express the trustworthiness of information and connections (which ties in with the secure browser work above). The Dartmouth Computing Services framework and the systems provided by AT&T with the grant will be a fertile backdrop for dealing with questions about integrating pervasive computing into the PKI infrastructure. CONCLUSION ---------- Dartmouth College provides a unique testbed for the PKILab. It has already significant experience in development of authorization infrastructure and is very much in the business of promoting and improving it. It has the Institute for Security Technology Studies which provides immense leverage, exposure, and national application for the proposed PKILab. It has fresh forward-looking ideas to investigate in the context of the PKILab. It has a long history of sophisticated computer users who willingly experiment with new software systems. Finally, it has the right scale for the proposed implementations and study---small enough to manage yet large enough to be interesting. BACKGROUND ---------- The Institute for Security Technology Studies was recently established to serve as a national center for research in security technologies related to cyber-security and information infrastructure protection. Funded at $15M/year, it is comprised of researchers in computer science and engineering at Dartmouth along with partners in leading universities and laboratories. See http://www.ists.dartmouth.edu . Dartmouth's Computing Services has a long history of developing innovative networked services and effectively deploying them across the entire campus. Examples of networked services developed at Dartmouth include the Potlatch file server (circa 1985 high- performance AppleShare and FTP server), Dartmouth Name Directory (circa 1986 campus- wide white pages and authentication directory), BlitzMail (circa 1987 institutional e-mail system), Fetch (circa 1989 Macintosh FTP client), the Dartmouth College Information System, DCIS (circa 1991 campus-wide client-server information "dashboard"), and InterMapper (circa 1996 network management application). All of the above services are in active campus-wide use in 1999 and their developers are available as technical resources. See http://www.dartmouth.edu/pages/softdev/ INVESTIGATORS -------------------- Edward A. Feustel just started as a Research Associate at Dartmouth's Institute for Security Technology Studies. Ed's research focuses on security issues in distributed computing applications and infrastructure. He has represented Prime and IDA as technical liaison to the Object Management Group and to OSF and X/Open. He comes to Dartmouth from the Computer and Software Engineering Division of the Institute for Defense Analyses, Alexandria, Virginia where he was a Research Staff Member with interests in Security and Distributed Systems and Applications. Prior to IDA Alexandria, he worked at Prime Computer as a Principal Technical Consultant, Rice University as a tenured Associate Professor of Electrical Engineering and Computer Science, IDA Princeton as a Systems Programmer, Lawrence Livermore Laboratory as a Sabatical Researcher, and California Institute of Technology as a Research Fellow. He is a graduate of Princeton University (MA 1966; Ph.D. 1967 in Electrical Engineering) and MIT (BSEE and MSEE, 1964). His 1971 IEEE Transactions of Computers Paper on the Advantages of Tagged Architectures set the stage for network exchange of tagged data such as CORBA and application exchange of tagged data such as XML. Sean Smith (http://www.cs.cmu.edu/~smith/) just started as an Assistant Professor of Computer Science at Dartmouth College. Sean's research focuses on both practical and theoretical aspects of security and privacy in distributed systems. He comes to Dartmouth from the Secure Systems group at IBM T.J. Watson Research Center, where he designed the security architecture for (and helped code and test) the IBM 4758 secure coprocessor, and then led the formal modeling and verification work that earned it the world's first FIPS 140-1 Level 4 security validation. Prior to IBM, he worked in the Computer Research and Applications Group at Los Alamos National Laboratory, performing e- commerce security reviews and designs for a wide range of public-sector and private-sector clients. He is a graduate of Carnegie Mellon (M.S., CS, 1988; Ph.D., CS, 1994) and Princeton (A.B., Mathematics, 1987). David Nicol (http://www.cs.dartmouth.edu/~nicol) is Professor and Chair of the Department of Computer Science at Dartmouth College, as well as the Technical Coordinator for the Cyber-security program at ISTS. He has long worked on issues in systems, distributed (and parallel computing), and is kept nice and practical by his consulting work with IBM, NASA, AT&T, and Sandia National Laboratory. He publishes prolifically, and serves as the Editor-in-Chief of ACM Transactions on Modeling and Computer Simulation. He holds a B.A. in Mathematics from Carleton College (1979), and a Ph.D. in Computer Science from the University of Virginia (1985). Robert Brentrup is the Director of Information Systems at Dartmouth College. He has been the project director for the Dartmouth College Information System (DCIS) since 1990 and manages Library Systems. Available at http://www.dartmouth.edu/~dcis/ or by using locally developed desktop applications, DCIS delivers bibliographic sources, electronic reference works, multimedia, general campus information and administrative data from a wide variety of college departments to thousands of users daily. DCIS includes extensive authentication and authorization features which interface to the institutional directories of a multiple campus consortium who have jointly licensed some of the databases. Robert previously worked at Lotus Development Corp., where he was a principal engineer involved in the Lotus Jazz, Notes and Improv products, Spartacus Computers and Raytheon Co. He is the author of a number of professional papers and was a member of the Northeastern University faculty. Robert holds a B.S. with high honors in Electrical Engineering from Michigan Technological University (1977) and a M.S. in Computer Engineering from Boston University (1981). William Taylor is the Director of Technical Services at Dartmouth College responsible for the network and computing infrastructure of the College. He has overseen the development of campus-wide email, file service, print spooling, and network management systems as well as the College's directory and authentication services. Previously he was a department manager at Computer Sciences Corporation where he was project leader on large software development projects funded by NASA and the Department of Defense. He holds a B.S. in Engineering from Duke University and a M.S. in Applied Mathematics from Johns Hopkins University.