CENTER FOR THE NEW ENGINEER


http://www.cne.gmu.edu

Final Report, DARPA Project
HPC in the Curriculum

Peter J. Denning, PI
Daniel A. Menasce, co-PI


Joseph Gerstner, CNELINK Director
Sam Wyman, CLIN Director

DARPA Contract DABT63-93-C-0026
May 1993 through July 1997
Total funds $3.1M


This project was begun in May 1993 under the auspicies of the DARPA High Performance Computing Infrastructure program. The overall objective was to help integrate knowledge gained from R&D in high performance computing and communication into the engineering curriculum and into the normal practices of interaction between engineering schools, industry, and K-12 schools. For the first two years, we focused on building infrastructure that would help us construct and disseminate knowledge modules, involve faculty, make HPCC labs accessible to students, and link to a consortium of K-12 schools that got their first contact with Internet through this project. CNELINK, the network of K-12 schools, was designed and operated in partnership with CLIN (Community Learning and Information Network).

In 1996 DARPA discontinued the HPCC program and transferred our project into the auspicies of the CAETI (computer aided education and training initiative). After the transition, we concentrated our full attention on the tasks that supported tools and systems for on-line, distributed learning, self-assessment, and certification. We enlisted real customers from within DoD, professional societies, George Mason University student body, and the local Fairfax Country Public Schools system.

Overview of the Project

The Statement of Work committed us to four main tasks:

At the start of the project, we founded the Center for the New Engineer, a research center in the School of Information Technology and Engineering at George Mason University devoted to educating engineers for the 21st century. CNE served as the focal point for all the activities of this project, plus a companion project funded by NSF to create learning modules to help engineers understand complex systems.

The remainder of this report gives more details on our work and accomplishments under each task. In addition to postive results, we will report on dead-ends or ideas that did not work out in practice, in the hope that others can learn more this way. At the end is a summary of the major accomplishments and publications of the project.

Task 1: Integrating HPC into the CS Curriculum.

Task 1.1: Curriculum Development: Modules, Workbenches, and Exhibitions

We began with a requirements analysis to determine what approaches to curriculum improvement would have the highest leverage and would be most likely to be adopted by faculty within accreditation guidelines. We came to three conclusions, in order of priority. (1) On-line (Internet) tutorial modules on focused topics would be the easiest to adopt because they could be used as supplements to existing courses. (2) A senior design exhibition with an outside industry customer for each student team would be an excellent method to insure that students gain system-integration knowledge and experience in dealing with real-world project concerns. (3) A programming practicum, in which students exhibit their programming skills, would be a good gateway to the upper division curriculum.

(1) Modules and Workbenches. During 1994-96 we implemented a library of on-line tutorial modules. Because these modules consisted of a collection of objects hyperlinked in the World Wide Web, we began to call them hyperlearning modules in 1996. Early on, we invented a subway-map paradigm to assist students in navigating these modules. The subway maps represented the non-linear connections among the objects while suggesting preferred routes of travel. The user can click on any subway station and be taken to the Web page on that topic. The modules library covers eight topics in computer science and engineering, one on general engineering, and two college-entry-level refreshers:


The math and statistics refreshers were implemented for the Defense Acquisition University in 1996-7, to assist their students enrolled in cost-accounting courses. These modules use many examples from cost accounting to illustrate concepts, and contain clear statements of what the student can expect to get from each topic element. They also contain an early version of an on-line self-assessment system.

In 1995 we turned attention to the problem of implementing workbenches for inclusion in the modules. A workbench is the simulation of a system from a control panel accessible to the user. The user can perform experiments with the workbench and use it to attain an understanding of a complex behavior that would be difficult to describe in words and static diagrams. Our initial attempts at workbenches were based on an attempt to present the user with a remote interface that acted as if the system were running on the user's PC. This approach did not work well because commercial software for remote desktops was slow and buggy. We finally abandoned it in favor of Java applets, which could be downloaded to a user's PC and run locally. This approach was practical and gave good performance. The downside is that the workbenches had to be programmed from scratch and could not serve as remote interfaces for complex systems. We implemented a library of sixteen workbenches in six categories:

The hyperlearning modules and workbenches were used extensively in operating system, networking, distributed-system, and parallel system courses in engineering and in computational science. The instructors reported that the modules saved them class time on these topics, allowing more time to cover other material and to answer students' questions. Students reported that the modules and workbenches helped them understand concepts that their books and papers did not adequately explain. Our modules have been successful on an international scale. Our Web site receives over 4000 hits a week from people accessing our modules from all over the Internet; it received a Magellan three-star rating. We receive frequent emails from users praising the modules.

(2) Senior Design Exhibitions. In 1993 we put together a pilot version of design exhibition course. The objective was to offer teams of seniors the opportunity to do a project with an outside industry customer. Their grade depended on showing (exhibiting) to the faculty advisor and industry customer that they understood how to integrate complex system knowledge from separate computer science courses into a working system that met the customer's requirements. Professor Menasce offered the pilot course in Spring 1994. It was a smashing success, earning 5.0/5.0 ratings from all students and from the industry customers. The Computer Science faculty subsequently adopted the course.

We prepared a senior design exhibition module that explained the objectives of senior design projects in engineering and linked it to the four undergraduate departments offering senior design projects for their majors. With support from the Century Club, a local organization of companies who assist the university, we put together a database system that permits prospective companies to register projects. The course coordinator can then match students with projects, form teams, and put each team in contact with its industry customer.

(3) Programming Practicum.We proposed to the faculty that they create a new exhibition course for juniors. The programming practicum, serving as the gateway for entry into the upper division curriculum, would require students to demonstrate their ability to solve problems with programs. The course proposal was adopted by the Computer Science faculty but could not be implemented because the department was unable to get the capital resources needed to support the extra staff needed to supervise the course.

Task 1.2: Extensions to High School Faculty and Students

On an experimental basis we wanted to make our modules and workbench technologies available to high school teachers and students. This would give advanced students seeking to major in computer science and engineering an early start, and would establish relationships that might encourage these students to come to George Mason University. We subcontracted with CLIN, the Community Learning and Information Network, to assist in this task and we appointed Joseph Gerstner as the director of this task.

CLIN canvassed regional schools in Alexandria, Arlington, Fairfax, and Prince William Counties and identified a set of schools that were interested in participating in the experiment. The two most pressing needs they identified were affordable full Internet access, and training. To meet the first need, we designed and implemented an ISDN network based on a hub at George Mason University (in our CNE lab) and direct BRI connections to each of the dozen schools who agreed to participate. With assistance from the George Mason University networking group, we arrived at a configuration that would cost each school $1000 to connect (via a Digiboard connected to a server on their LAN) and $40 a month in ISDN line charges. Due to lack of sufficient funds, we abandoned an initial plan to provide real-time ISDN TV to all the sites. To meet the second need, we organized summer training workshops for teachers in 1993, 1994, 1995, and 1996, and offered occasional consulting to them in between. CLIN also offered them a range of software packages of possible use in high-school classes for evaluation.

We created the name CNELINK for the fifteen schools connected to Internet via the ISDN links to the CNE Lab. The schools (and their counties) were:


Oakview school's parents, teachers, and children produced such an outstanding web page that they were selected as one of five schools honored in a book sponsored by Adobe.

CNELINK completed its work by May 1996, when all the schools had been transitioned from our ISDN network into commercial ISDN or Internet providers. Managing a network of schools was far more time-consuming than we originally expected. We were glad that professional ISP's were able to take over at the end of the experiment and allow the schools to continue their good work on their own.

Task 1.3: Hyperlearning Meter System

The Hyperlearning Meter system grew out of earlier work in using CNE modules for real courses and training, where we learned that users placed considerable value on ubiquitous self-assessment opportunities throughout the modules. Our customers also wanted to know if we could provide on-line certification of the knowledge imparted by a module. We created an early version of a self-assessment system for the DAU math and stat refresher modules. This system presented users with a series of questions drawn at random from a database, with randomized parameter values, and with random ordering of answer choices. On answering the questions, the user got an immediate report of correct and incorrect answers, together with links to the module elements to which the user could refer for more information. The database of test questions and the test-generator engine all ran on a single server (in the CNE lab).

In our work with the CAETI community, we learned that self-assessment must work in a community of students and instructors course instructors distributed around the Internet. In Fall 1996 we embarked on a new version of the system that would meet these objectives: question templates could be stored in many Internet-accessible databases, enrolled students could access tests from anywhere in the Internet, test-generation should occur locally on the student's client to avoid server bottlenecks, and the whole system should be easily scalable to handle thousands of students and hundreds of instructors. The result was the current Hyperlearning Meter System.

A hyperlearning module covers a space of concepts that must all be mastered by a student seeking certification in the knowledge promised by the module. We can draw a diagram of the concept space as a map showing boxes, one for each concept, arranged as a tree. Weights on the branches tell the relative importance of subconcepts to a concept. The concept map is used by the Hyperlearning Meter system to generate automatically complete tests whose questions cover all the concepts in relative proportion to their importance. The concept map is used by the instructor to declare and connect the concepts of the module. It is related to, but not the same as, the subway map, which is a navigational aid for the student; the subway map recommends learning paths among the concepts, represented as stations.

The Hyperlearning Meter consists of the following components:

To test ease of use of the Hyperlearning Meter system, we trained six third- and fifth-grade elementary teachers from Fairfax County Public Schools. Within half an hour, they were able to create PATs of questions that will be offered to their students in Fall 1997. The feedback from these teachers led to some design improvements (mostly to provide missing functions or to improve access to important functions).

To test the robustness of the system for a large database, we developed a set of over 200 PATs for an operating system course. In Fall 1997, students will use this system to test their knowledge of course concepts. The instructor (Menasce) will measure the change of productivity enabled by the system in terms of the depth of knowledge displayed by students and the amount of material covered compared to previous semesters.

To test the system in a professional application, we collaborated with ACM (Association for Computing) on the prototype of a Professional Knowlege Module on Data and Network Security. The module consists of a collection of 15 seminal articles, editorial overviews, two dozen secondary articles, study questions, links to other Web resources, and a search engine. A certification test at the end measures reading comprehension of all the articles. The test generator gives the student 2 questions from each article, 2 questions about relationships among articles, and 2 questions applying articles to current situations; the certifier issues the certificate to anyone who scores 30/34 with at least one correct for each article. ACM plans to develop a major new professional education program around PK modules.

Validation of the Hyperlearning Meter system is a major open research issue. The question is, what does the person who passes the test actually know? The question cannot be answered in the traditional way for other written tests because the student may have repeated self-assessment tests several times each prior to the final certification. Moreover, the path that a student follows in concept space may constitute a signature revealing the student's level of expertise in the domain.

Task 2: Bringing HPC Technology to a Large Number of Students at a Price per Student Approximating that of a PC.

This task focused on implementing an architecture for distributing the services of an HPC lab to group of students and participating faculty. The architecture consisted of three layers: (1) The HPC lab served as the core layer. It consists of a Sun server and a local network of Sun, Dell, and Mac workstations. Computations can be performed on the server or the workstations, or they can be distributed among several workstations using the PVM system. (2) Sun workstations installed in the offices of participating faculty served as the faculty layer. This allowed participating faculty to create course materials on the server and for use by students. (3) Individual student PCs served as the student layer. With these PCs, students could log in to the core layer and engage with software prepared by faculty on the faculty layer. The facilities of the core and faculty layers have been operated and maintained reliably and securely since their inception.

In addition to these hardware layers, we planned to partition HPC tasks into client and server parts. The server parts would run in the core layer and the results would be delivered to the student layer. This strategy would allow the heavy computational work to be done in the lab without having to require students to purchase top-of-the-line PCs.

The core and faculty layers were implemented by the beginning of 1994. The student layer was designed in 1995 but not implemented. Two events conspired to cause us to abandon implementation of the student layer. One was the transition of the program within DARPA to CAETI, which called for different emphases of our effort among the four Tasks; tool development and field testing were emphasized (Tasks 1 and 4) and infrastructure and K-12 outreach de-emphasized (Tasks 2 and 3). The other was the rapid development of commercial technology; by 1995 students could purchase sufficiently powerful PCs at reasonable prices and we had learned how to distribute computational tasks to their PCs via Java applets.

Task 3: Establish a Feedback Loop from HPC Research Community into Undergraduate Education.

This task was successfully completed by a variety of means:

(1) Faculty participating in HPC research programs were put into closer contact with undergraduate students. This included faculty from Computer Science, Electrical and Computer Engineering, Operations Research, Systems Engineering, Applied Engineering Statistics, and the Institute for Computational Science and Informatics. Save for a module on genetic algorithms, all the modules designed by this team were completed and are available to the Internet in the CNE modules library.

(2) With assistance from a companion NSF Grant ("Educating Engineers to Design Complex Systems") we were able to hire extra students to assist in the construction of the cross-disciplinary hyperlearning modules.

(3) The CNE hyperlearning modules enabled more faculty to supplement their course materials and bring students into contact with the material on parallel and distributed high-performance computing.

(4) About two dozen graduate students were employed by CNE during the project period. They were brought in to the circle of people skilled in HPC.

Task 4: Establishing a Partnership with Government and Business Organizations in the HPC Curriculum

As a first step, we created the Consortium for the New Engineer, an advisory group for the George Mason University HPC education program. Its board consisted of faculty, industry, government, and school-district representatives. We established partnerships with Action Technologies, Course Technology, Lotus, Media General, and the Morino Foundation. We also created a liaison with NSF through our companion project funded by an NSF grant.

We involved a number of regional companies in the Senior Design Exhibition projects, as customers for the student teams. All participants, students and companies, found this to be very fruitful. A larger circle of companies became involved through George Mason University's Century Club, which assisted in designing a database to allow companies to register projects available to students in all six departments of the School of Information Technology and Engineering.

Through CLIN, we were allied with companies and government organizations interested in using distance-learning technology to deliver on-site training.

We entered into a collaboration with DARPA/DISA AITS-JPO for joint studies assessing and promoting technology transfer of HPC technologies applicable to the DoD. This work was funded with an increment of $80,000 to the contract amount, and was coordinated by Walt Corliss.

We enlisted several major organizations to serve as customers for our hyperlearning modules and hyperlearning meter. The first was the Defense Acquisition University (DAU), for whom we developed two refresher modules and a first-generation self-assessment system. The second was Association for Computing (ACM), for whom we developed a Professional Knowledge Program module on Data and Network Security; this module is a prototype of a family of modules ACM plans for its members to help introduce them to hot topics through intensive reading programs with certification that the person has read all the materials. The third partnership is with the Fairfax County Public Schools, who have put the hyperlearning meter to work in their third-grade classrooms as a way of helping students prepare for Virginia's Standards of Learning tests.


Major Accomplishments

Establishment of the Center for the New Engineer and its successful Web site (http://www.cne.gmu.edu). The Web site received (unsolicited) a Magellan three-star award.

Creation of the HPC Lab, with core, faculty, and student computing layers, as part of the Center.

Formation and implementation of CNELINK, the network connecting fifteen K-12 schools to Internet before ISPs were able to do the job at a cost the schools could afford. One of the schools (Oakview in Fairfax) received a special recognition as an outstanding K-12 school Web site.

Design and implementation of an architecture for hyperlearning modules, including subway-map navigation, culminating in a library of eleven modules and sixteen Java workbenches on the Web site.

Design and implementation of the Hyperlearning Meter system, a distributed system for self-assessment and certification based on tests generated on the fly from parametric templates based on a concept map of the domain.

Application of the HL modules and HM system in the training programs of Defense Acquisition University (DAU), Association for Computing (ACM), and Fairfax County Public Schools (FCPS).



Personnel

In addition to the PI (Peter Denning) and co-PI (Daniel Menasce), these faculty participated in the projects of the Center:

Christopher Dede -- Graduate School of Education
Kenneth DeJong -- Computer Science
Lynn Fontana -- Instructional Development
Eugene Norris -- Computer Science
Mark Pullen -- Computer Science
Sanjeev Setia -- Computer Science
Pearl Wang -- Computer Science
Elizabeth White -- Computer Science

These faculty participated in the NSF project advisory group that contributed to the Center and helped select topics for the modules:

Ophir Frieder -- Computer Science
James Gentle -- Computational Sciences and Informatics
Menas Kafatos -- Computational Science and Informatics
Kathryn Laskey -- Systems Engineering
Andrej Manitius -- Electrical and Computer Engineering
George Michaels -- Computational Science and Informatics
George Tecuci -- Computer Science
John Wallin -- Computational Science and Informatics
Edward Wegman -- Computational Statistics

These students were employed as Graduate Research Assistants during the contract period:

Agrawal, Amit
Apsani, Jyothi
Bangalore, Anantha
Bujamella, Srikantha
Deshbandu, Naval
Ellis, Ellen
Fort, Celese
Hieb, Michael
Jones, Alfred
Koneru, Srilatha
Le, Hai
McAndrews, Priscilla
Nandigama, Satish
Park, K.
Patil, Vanita
Puli, Sekhar
Quisumbing, Edgar
Rao, Amar
Reddivalam, Satish
Sinha, Mahua
Sunkara, Anil
Tata, Srinivasa
Telukutla, Aruna
Tran, Ngon
Winkler, Jeff
Valuri, Srinivas
Vundavali, Sudhakar
Zhou, Ying
Zhou, Zhen

These people were employed as system or project administrators during the contract period:

Dybala, Thomas
Kaull, Nina
Mundur, Padma
Schultz, Alan
Potter, Mitch


Publications

The publications of the investigators and students are listed below in two categories: (1) those related directly to the Center and the DARPA project, and (2) those related to individual modules or components of the Center.

Publications Directly Related to the Center

P. Denning. 1992. "Educating a new engineer." Communications of ACM 35, 12 (December), 83-97.

P. Denning. 1993. "Designing new principles for sustaining research in our universities." Communications of ACM 36, 7 (July), 99-104.

P. Denning, D. Menasce, and J. Gerstner. 1995. "Re-engineering the engineering school." In Proc. ASEE Conf. (June), 1037-1042.

P. Denning. 1996. "Business designs for the new university." Educom Review (Nov-Dec), 20-33.

P. Denning. 1996. "Undergraduate education in computer science and engineering." AMS Notices (June), 677-680.

P. Denning. 1997. "A new social contract for research." Communications ACM (Feb).

P. Denning. 1997. "How we will learn." In Beyond Calculation: The Next 50 Years of Computing (P. Denning and B. Metcalfe, eds.). Copernicus Books, 267-286.

P. Denning. 1997. "The new university." The Leading Edge, published by the Society of Exploration of Geophysics (July), 1013-1016.

P. Denning. 1997. "The new engineer revisited." (Interview format with Charlotte Thomas, ed.) Graduating Engineer, Spring 1997. Full text of interview at http://cne.gmu.edu/cne/perspectives/thomas.html.

P. Denning. 1997. "Quantitative practices." In Why Numbers Count (Lynn Arthur Steen, ed.) College Board Press, 106-117.

P. Denning. 1997. "A glimpse into the future." (Interview format with John Gehl, ed.), Educom Review (July 1997), 14-27.

P. Denning. 1997. "Getting hyper over learning." (Interview with Deborah Coppula, ed.) ASEE Prism (May), 12-17.

Publications Related to Modules or Components of the Center

A. K. Bangalore and D. A. Menasce. 1994. "An Experimental Study of the Effects of Heterogeneity and Data Partitioning on the Performance of Parallel Applications." First International Workshop on Parallel Processing, Bangalore, India (December 27-30).

P. Denning and Raul Medina-Mora. 1994. "Case Study of Course Scheduling in a University." In New Tools for New Times: The Workflow Paradigm, published by Future Strategies, Inc.

P. Denning. 1994. "Worldnet." In Studies in Computer Science (J. Rice and R. DeMillo, Eds.), Plenum Press, 15-27.

P. Denning and R. Medina-Mora. 1995. "Completing the Loops." ORSA/TIMS Interfaces 25, 3 (May-June), 42-57.

P. Denning. 1995. "Can there be a science of information?" ACM Computing Surveys (June 1995), 23-25.

P. Denning and P. A. Dargan. 1996. "Action-Centered Design." In Bringing Design to Software (T. Winograd, Ed.), Addison-Wesley.

P. Denning. 1996. "Workflow in the web." In New Tools for New Times: Electronic Commerce (L. Fischer, ed.), Future Strategies. 45-58.

P. Denning. 1997. "The Internet after thirty years." In Internet Besieged: Countering Cyberspace Scofflaws (D. Denning and P. Denning, eds.), Addison-Wesley and ACM Press.

P. Denning. 1997. "Secure electronic commerce." In Internet Besieged: Countering Cyberspace Scofflaws (D. Denning and P. Denning, eds.), Addison-Wesley and ACM Press.

P. Denning. 1997. "Before memory was virtual." In In The Beginning: Personal Recollections of Software Pioneers. Robert Glass, Ed. IEEE Press.

P. Denning. 1997. "Virtual memory." CRC Handbook of Computer Science and Engineering 1448-1451.

H. Gomaa, D. A. Menasce, and Larry Kerschberg. 1996. "A Software Architectural Design Method for Large-Scale Distributed Data Intensive Information Systems." Journal of Distributed Systems Engineering 3, 162-172.

L. Kerschberg, H. Gomaa, D. A. Menasce, and J. P. Yoon. 1996. "Data and Information Architectures for Large-Scale Distributed Data Intensive Information Systems." Proc. of the Eighth IEEE International Conference on Scientific and Statistical Database Management, Stockholm, Sweden, (June 18-20).

D. A. Menasce and Stella C.S. Porto. 1993. "Scheduling on Heterogeneous Message Passing Architectures." Journal of Computer and Software Engineering, Ablex Publishing Co., New Jersey, Volume 1, Number 3.

D.A. Menasce, S. H. Noh, and S.K. Tripathi. 1993. "A Methodology for the Performance Prediction of Massively Parallel Applications." 5th IEEE Symposium on Parallel and Distributed Processing, Dallas, Texas (December 1-4).

D.A. Menasce, V.A.F. Almeida, and L. Dowdy. 1994. Capacity Planning and Performance Modeling: from mainframes to client-server systems. Prentice Hall, Englewood Cliffs, New Jersey.

D. A. Menasce, Y. Yesha, and K. Kalpakis. 1994. "On a Unified Framework for the Evaluation of Distributed Quorum Attainment Protocols." IEEE Transactions on Software Engineering 20, 11 (November).

D. A. Menasce, S. C. S. Porto, and S. K. Tripathi. 1994. "Static Heuristic Processor Assignment in Heterogeneous Multiprocessors." International Journal of High Speed Computing 6, 1 (March), 115-137.

D. A. Menasce and V. A. F. Almeida. 1994. "Two-level Performance Models of Client-Server Systems." Computer Measurement Group Conference, Orlando, Florida (December).

D. A. Menasce, D. Saha, S. C. da Silva Porto, V. A. F. Almeida, "Static and Dynamic Processor Scheduling Disciplines in Heterogeneous Parallel Architectures." and S. K. Tripathi. 1995. Journal of Parallel and Distributed Computing 28, 1 (July), 1-18.

D. A. Menasce and A. Bangalore. 1995. "A Theoretical and Experimental Assessment of Data Partitioning Strategies on Networks of Heterogeneous Workstations." Special issue on Parallel Computation of the Journal of the Brazilian Computer Society 1, 2 (July), 5-12.

D. A. Menasce. 1995. "Capacity Planning in Client/Server Environments." Proc. of the 8th International Conference on Modelling Techniques and Tools (PERFORMANCE TOOLS '95), Heidelberg, Germany (September 18-21).

D. A. Menasce, H. Gomaa, and L. Kerschberg. 1995. "A Performance-Oriented Design Methodology for Large-Scale Distributed Data Intensive Information Systems." First IEEE International Conference on Engineering of Complex Computer Systems, Southern Florida, USA (November 6-10).

D. A. Menasce, D. Dregits, R. Rossin, and D. Gantz. 1995. "A Federation Oriented Capacity Management Methodology for LAN Environments." Conference of the Computer Measurement Group, Nashville, Tennessee (December).

D. A. Menasce and D. Dregits. 1995. "On the Design and Implementation of a Capacity Management Tool for LAN Environments." Conference of the Computer Measurement Group., Nashville, Tennessee (December).

O. I. Pentakalos , D. A. Menasce, Y. Yesha, and M. Halem. 1995. "An Approximate Performance Model of a Unitree Mass Storage System." Fourteenth IEEE Symposium on Mass Storage Systems (Second International Symposium), Monterey, CA (September).

D. A. Menasce, O. Pentakalos, and Y. Yesha. 1996. "An Analytic Model of Hierarchical Mass Storage Systems with Network-Attached Storage Devices." Proc. of the 1996 ACM Sigmetrics Conference, Philadelphia, PA (May).

D. A. Menasce and A. Rao. 1996. "Performance Prediction of Parallel Applications on Networks of Workstations." Computer Measurement Group Conference, San Diego, CA (December).

O. I. Pentakalos and D. A. Menasce. 1996. "Automated Clustering Based Workload Characterization for Mass Storage Systems." Fifth NASA Goddard Space Flight Center Conference on Mass Storage Systems and Technologies, College Park, MD (September 17-19).

O. I. Pentakalos, D. A. Menasce, and Y. Yesha. 1997. "Pythia: A Performance Analyzer of Hierarchical Mass Storage Systems." Performance Tools'97, 9th International Conference on Modelling Techniques and Tools for Computer Performance Evaluation, St. Malo, France (June 2-6).

O. I. Pentakalos, D. A. Menasce, M. Halem, and Y. Yesha. 1997. "Analytical Performance Modeling of Hierarchical Mass Storage Systems." IEEE Transactions on Computers 46, 10 (October), 1103-1118.

O. I. Pentakalos, D. A. Menasce, and Y. Yesha. 1997. "Pythia and Pythia/WK: Tools for the Performance Analysis of Mass Storage Systems." Software Practice and Experience (October).