1. May 19, 2005 Presentation General overview - Barry Kurtz
Set up of grid – Barry Wilkinson
Distributed MATLAB – Rahman Tashakkori
Course overview – Barry Kurtz
Cryptography course - Shan Suthaharan
Image processing course - Rahman Tashakkori
Future of grid computing - Barry Wilkinson
Intelligent decision making course – Dave Powell
Remaining courses – Barry Kurtz
Future Plans – Barry Kurtz

2. Overview
Barry Kurtz

3. A Consortium to Promote Computational Science and High Performance Computing
Lead Institution: Appalachian State
Partnering Schools: Elon University, High Point University, Lenoir-Rhyne College, North Carolina A & T, UNC Charlotte, UNC Greensboro, UNC Pembroke, Western Carolina University
Funding: $650,000
Duration: July 1, 2004 – June 30, 2006

5. Mission
Our mission is to provide the opportunity for undergraduate students at comprehensive universities to study computational science and high performance computing at a level comparable to students at Research I institutions, to promote faculty research by involving undergraduate students in cutting-edge research projects, and to stimulate economic development by promoting grid computing methodologies throughout North Carolina.

6. Vision
Our vision is that by pooling knowledge, resources, and courses at eight collaborating comprehensive universities, by establishing a shared grid computing network, and by exporting grid technology to local IT companies, we can satisfy our mission statement.

7. Goal
Our goal is to provide students at comprehensive universities the opportunity to take advanced courses in computational science and high performance computing at the undergraduate level and allow these students to participate in cutting-edge team-oriented research projects in support of faculty research programs.

8. Faculty and Institutions
Barry L. Kurtz and Rahman Tashakkori
Appalachian State
David Powell and Joel Hollingsworth
Elon University
Bill Hightower and Roger Shore
High Point University*
Dick Hull and Bjarne Berg
Lenoir-Rhyne College*
Yaohang Li and Stephen Providence
North Carolina A & T
Barry Wilkinson
UNC Charlotte*
Shan Suthaharan and Sue Lea
UNC Greensboro
Doek-Hyun Hwang and William Campbell
UNC Pembroke
Barry Wilkinson and Mark Holliday
Western Carolina
* Joins the consortium in year two

9. Hardware Acquisitions
A Computer Cluster
Eight Dell Precision 360 or 370 workstations
High Speed Switch
Firewall
There are currently eight clusters purchased; a ninth cluster at UNC Charlotte will be added soon

10. Original Equipment Projections
Year One
$15,000 per cluster budgeted for six schools
$10,000 for partial cluster at two schools
Total allocation: $110,000
Year Two
$5,000 to complete cluster at two schools
Total allocation: $10,000
Actual Equipment Costs
Six months after the projected costs were estimated, the actual equipment (identical to the promised equipment) only cost $12,000 per cluster

11. Actual Equipment Expenditures
Year One
$12,000 per cluster for eight schools (the two partial clusters are made complete)
An additional cluster for Barry Wilkinson who is returning to UNC Charlotte (our ninth partner)
$2,000 is still left in reserve
Year Two
$10,000 to complete the partial clusters is available
Combined with the $2000 in reserve we can fund a tenth cluster for another UNC school that is willing to join our consortium

12. Software Grid Computing Software MATLAB MPI Globus Condor
Toolboxes: Wavelets, Signal Processing, Image Processing, Neural Networks, Statistics, Bioinformatics
Distributed MatLab
MPI
Standard MPI
Grid-enabled MPI

13. Set up of grid
Barry Wilkinson

14. Grid Support
Goals
Create a working grid with clusters at every university in the consortium.
Use that grid in courses that are jointly taught by faculty from several of the universities.
Participating Sites

15. Current Status
Summer 2004: Initial installation of the NMI Distribution including Globus Toolkit 3.2 and Condor-G
Summer 2004: Developed projects testing the distribution (web services, grid services, GRAM, Condor-G)
Summer 2004: Distribution CD developed
Summer 2004: First load testing in summer workshop at ASU
Participating Sites

16. Current Status Fall 2004: Clusters acquired and installed at all sites
Fall 2004: All clusters operational with basic functionality: Linux operating system and cluster software
Participating Sites

17. Current Status
Fall 2004: Intense testing of the grid software in the Grid Computing course taught by Barry Wilkinson
Fall 2004: Extended the software being used to include a MPI project developed by Barry Wilkinson and a grid workflow editor, GridNexus, developed at UNC-Wilmington
Participating Sites

18. Current Status
Spring 2005: Installation of a newer NMI Distribution (still GT3.2) at Elon and WCU
Spring 2005:Testing of the new version of the software in the Intelligent Decision Making course at Elon and WCU
Spring 2005: Joel Hollingsworth developed a new grid services project
Participating Sites

19. Future Plans
Summer 2005: Installation of a newer NMI Distribution (changing to GT4!)
Summer 2005: Redo all the projects using GT4; develop more projects
Summer 2005: Retest everything in the summer workshop
Summer 2005: Create a certificate authority at Elon to allow multi-site operation
Participating Sites

20. Future Plans
Fall 2005: Test the new installations and project versions in Barry Wilkinson’s Grid Computing course
Fall 2005: Further test the grid software in the capstone course and the other courses
Spring 2006: Revise the projects (and perhaps reinstall) based on the Fall experience
Spring 2006:Use in the spring courses
Participating Sites

21. Summary It has been a learning experience for everyone involved.
It has been a successful experience with students getting working experience with grid software.
Teamwork and inter-university collaboration have been the keys to our success.
Participating Sites

22. Summary Much remains to be done transition to GT4 project revisions
additional projects
certificate authority and multi-site grid operation
Participating Sites

23. Distributed MATLAB
Rahman Tashakkori

24. Distributed Toolbox
The MathWorks web site was the main source of the MATLAB related information on this presentation
The Distributed Computing Toolbox works with the MATLAB Distributed Computing Engine to execute coarse-grained MATLAB algorithms and Simulink models in a cluster of computers
It allows prototyping and developing applications in the MATLAB environment and then use the Distributed Computing Toolbox to divide them into independent tasks.
The MATLAB Distributed Computing Engine evaluates these tasks on remote MATLAB sessions.

25. Distributed Toolbox: Key Features
Distributed execution of coarse-grained MATLAB algorithms and Simulink models on remote MATLAB sessions
Control of the distributed computing process via a function-based or an object-based interface
Distributed processing on both homogeneous and heterogeneous platforms
Support for synchronous and asynchronous operations
Access to single or multiple clusters by single or multiple users

26. Distributed Toolbox

27. Creating and Submitting Jobs with the Distributed Computing Toolbox
The toolbox includes functions for defining jobs, dividing them into tasks, sending them to the MATLAB Distributed Computing Engine for execution, and retrieving the results. The complete process includes five steps:
Finding a job manager
Creating a job
Creating tasks
Submitting the job to the job queue
Retrieving results

28. Problems Network has to be able to do multicast
It is not X-term friendly, best to run the jobs from console
All machines should have the same version of MATLAB
It requires all Unix-based machines (manager and workers) to run the same version of kernel
In case that a firewall is used, specific ports should be left open to establish workers-server and server-license manager connectivity

29. Course Overview
Barry Kurtz

30. Courses Cryptography and Network Security Fall 04-05 Shan Suthaharan
Barry L. Kurtz
Digital Image Processing
Rahman Tashakkori
Sue Lea
Grid Computing
Barry Wilkinson
UNCW faculty
Intelligent Decision Making
Spring
05-06
Dave Powell, Joel Hollingsworth, Mark Holliday
Monte Carlo Methods
Yaohang Li
Parallel Algorithms
Steven Providence
William Campbell
Bioinformatics
Summer
Doek-Hyun Hwang
Capstone Research Course
Fall 05
Mark Holliday, et. al.

31. Research Outcomes Publications Presentations Projects
2 in print
2 under review
Presentations
7 completed
Projects
13 completed
2 planned for summer
Follow Up Grant Proposals
3 planned for the National Science Foundation
1 planned for another agency

32. Cryptography and
Network Security
Shan Suthaharan

33. Cryptography and Network Security
Host Institution – UNC Greensboro
Dr. Shan Suthaharan, primary instructor
Fall 2004, 15 students at UNCG
Remote Classrooms
Dr. Barry Kurtz, AppState, team instructor
2 students at AppState

34. Elliptic Curve Cryptography
Stephanie Rednour worked under the direction of B. Kurtz on this special research project

35. GUI for RC5 Encryption
Ramu Pulipati worked under the direction of Shan
Suthaharan on this special research project.

36. Digital Image
Processing
Rahman Tashakkori

37. Digital Image Processing
Host Institution – ASU
Dr. Rahman Tashakkori
Fall 2004, 20 students at ASU
Remote Classrooms
Dr. Sue Lea, UNCG
2 students at UNCG

38. Major Topics Introduction Mathematical Background Fundamentals
Intensity Transformations and Spatial Filtering
Frequency Domain Processing
Image Restoration
Image Compression
Wavelet Analysis

39. Image Processing Projects, ASU
These projects were completed under the direct supervision of Dr. Tashakkori
Investigating content-based search techniques in medical images of different modalities – Steve Heffner
This project lead to Steve’s honor’s thesis
An Efficient Medical Image Content-based Search Approach, May 2005
Parallel Implementation of lifting schemes – Tim Racz
This project lead to Tim’s M.S. thesis
Parallel implementation of Hexagonal Lifting Schemes, In process

40. Image Processing Projects, ASU
Other Projects:
A Java-based enhancement Toolbox. This toolbox includes several different modules that are developed by 8 students throughput the semester.
A C++-based enhancement Toolbox. This toolbox includes several different modules that are developed by 5 students.
Histogram equalization implementation of color images
A morphing and edge detection Toolbox. This toolbox includes three different modules that are developed by 3 of our students throughout the semester
An image cropping Toolbox.
A high- and low-pass filtering Toolbox

41. Image Processing Project, UNCG
David Waizenegger and Sarah Parker worked on this project under Dr. Lea’s supervision
The students investigated time sequences of (ocean) wave crest images. The wave crest data is not clean data; crests may be erroneously connected to form contours. Students were asked to come up with an automatic technique for determining the angle through which the image needed to be rotated in order to make the crests approximately parallel to the bottom of the rotated image. Accomplishing such a rotation simplifies analysis of the sequence of images to infer the bottom depth near the shore.

42. Future Plans – Fall 2005
Continue teaching the course in the consortium
Expand and improve existing Toolboxes
Create new Toolboxes
Establish a team-based research problems to includes collaboration across campuses
Utilize the grid to solve some of the image processing problems

43. Student Presentations
Steve Heffner, Investigating content-based search techniques in medical images of different modalities, The 8th Annual Celebration of Student Research and Creative Endeavors, April 19, Appalachian State University

44. The Future of
Grid Computing
Barry Wilkinson

45. Intelligent Decision
Making
Dave Powell

46. Intelligent Decision Making
Elon Instructors: Powell, Hollingsworth
WCU Instructor: Holliday
Using Inexpensive High Performance Computing to design products faster, better and cheaper.

47. Overview
Motivation: Need for High Performance Computing for Global Competitiveness
Course Enrollment, Topics and Projects
Ideal Capstone Course
Key Benefits
Key Pedagogies for NCREN
Key Challenges
Student Survey Suggestions
Research Outcomes
Future Work

50. Optimization Customers
Airbus
AlliedSignal
Black & Decker
Boeing
Bombardier
Bridgestone/Firestone
Corning
Daimler
Delphi Packard
Ford
General Electric
General Motors
Hitachi Computer
Kodak
Lockheed-Martn
Mitsubishi
Motorola
NASA
Nippon Sheet Glass
Northrup Grumman
Otis Elevators
PPG
Pratt & Whitney
Raytheon
Rolls Royce
Sony
Toshiba
TRW
U.S. Army, Navy, & Air Force
Voith
Xerox
York

53. Key Questions For Engineer and Student
How do I pose the optimization problem? (theory)
What type optimization algorithms are available? (theory)
How do I pick the correct algorithm? (theory)
What are some of the available optimization codes? (public domain)
How can I quickly apply code to my “customers” problem? (wrapping)
How can I use grid computing to quickly arrive at a solution?

54. Course Details Host Institution – Elon University Student Breakdown
Team taught: Hollingsworth and Powell, Elon University, Holliday, Western Carolina
Student Breakdown
Eight students at Elon
Three students at Western Carolina
Five students at Appalachian State
Average GPA:

55. Major Topics A Mathematical Programming Language (AMPL)
Object wrapping with adapter of facade design pattern
Condor-G for multiple and parallel job submission
Grid Services (Globus 3.2)
Introduction to Formulation and Classification of Optimization Models
Elements of Improving Search Based Optimization Algorithms
Formulation and Classification of Linear Problems
Simplex algorithm
Sensitivity analysis
Formulation of Unconstrained NLP
Golden Section and Gradient Search
Formulation of Constrained NLP
Penalty Methods
Formulation of Mixed Integer Problems

56. Programming Projects
Used the Façade Design Pattern to wrap the AMPL commercial code and called it from a customized Java Swing GUI.
AMPL Java Wrapper
AMPL
Gui Application Specific Interface
APIs
User Library

57. Programming Projects
Coupled a third party nonlinear constrained optimizer
AMPL Java Wrapper
AMPL
Gui Application Specific Interface
APIs
User Library
APIs
Student Optimizer

58. Programming Projects
Develop optimization grid service and validate/demonstrate with two separate client GUIs.
Grid at Elon
AMPL Java Wrapper
Gui Application Specific Interface
User Library
AMPL Java Wrapper
AMPL
Client Machine at ASU
Student Optimizer

59. ASU Submission

60. Programming Projects
Developed Condor-G scripts to submit and execute optimization from multiple starting points on Elon 8 node grid running Globus 3.2.

61. Ideal Capstone Course Languages/Formats AMPL C Fortran Java Soap XML
WSDL
WSDD
Concepts
Client/Server
Wrappers
JNI
Certificates Pairwise Programming SOA
Proxy
Reuse
Applied Math
Finite Differences Taylor Series
Lagrange Multipliers
Simplex
Penalty Functions
Convex/Concave
Courses
Calculus 1 and 2
Linear Algebra
Programming 1, 2, 3
Programming Languages
Environments/Tools
Linux
Apache/Tomcat/Axis
Web Services
Grid Services
Condor
Ant
Abstractions
Optimization Formulations

62. Key Benefits Students Faculty/University
Course that would not otherwise be taught.
Expertise of faculty.
A grid node with 8 machines.
Push outside comfort zone.
Comparison with peers.
State of the art exposure.
Faculty/University
New Computing Equipment
New NCREN Equipment
Dialog with colleagues
New skills

63. Key Pedagogies for NCREN
Pairwise Programming
Blackboard
Digital Dropbox
Discussion Board
Grades
Assignment Posting
Captured and posted class for video streaming
Semester project
Define, Specify, Design, Impl, Test
Netmeeting
Skype
Instant Messenger
HW Solution Posted immediately

64. Key Challenges Different start and end times
Different number of credit hours.
Different spring breaks.
Different off days (e.g. Easter)
System administration.

65. Student Surveys More frequent switching of project teams.
More individual work.
Course was too much work. Felt like two courses in one.
Better course description to emphasize amount of Math involved.
More interaction/presentations between the schools to hear more from the other students.
Same spring break for all campuses.

66. Research Outcomes
Hollingsworth and Powell submitted paper, “Globus Grid Computing and AMPL: a Pragmatic Educational Environment for Real World, Engineering Design Optimization” to “The Fifth IASTED International Conference on Modeling, Simulation and Optimization”. Status: Accepted.
Hollingsworth and Powell submitted paper, “Leveraging Grid Computing in an Intelligent Decision Making Course” to IASTED International Conference on Computers and Advanced Technology for Education Status: Accepted.
Extended AMPL Commercial Tool Capable for being extended to do multiple objective optimization. (standalone and as grid service)
Coupled multiple objective optimization package, NSGA2, with AMPL package.
Coupled simulated annealing optimization package, ASA, with AMPL package.

67. Future Plans
Install the new release of Globus Toolkit, GT4 and investigate assortment of third party parallel job submission tools from Platform Computing and Sun Microsystems.
Extend NSGA2 AMPL coupling to evaluate populations in parallel.
Investigate BPEL for workflow control of hierarchical optimization, simulation code evaluation, grid services and web services.
Extend Eclipse Plug-ins to support grid computing

68. Other Courses
Barry Kurtz

69. High Performance Computing and Monte Carlo Methods
Host Institution – North Carolina A&T State University
Dr. Yaohang Li, Primary Instructor
Spring 2005, 10 students enrolled at NCAT from Departments of Computer Science, Mechanical Engineering, Computational Science and Engineering, Physics, and Biology

70. Research Outcomes High Performance Computing and Monte Carlo Methods
Students completed term papers on the following topics
Computational Financing
Monte Carlo Computation on the Grid
Ray Chasing Problem using Monte Carlo method
Network Simulation
Protein Structure Prediction using Markov Chain Monte Carlo
Ab initio Monte Carlo

71. Parallel Algorithms Host Institution – North Carolina A & T
Dr. Stephen Providence, instructor
Spring 2005, 5 students at NCA&T
Remote Classrooms
2 students at Appalachian State supervised by Barry Kurtz
1 student at UNC Greensboro supervised by Shan Suthaharan

72. Outcomes Parallel Algorithms
Grid-enabled MPI was installed and used on the local computer clusters
Students completed a sequence of assignments out of Barry Wilkinson’s textbook:
Parallel Programming : Techniques and Applications Using Networked Workstations and Parallel Computers (2nd Edition)

73. Bioinformatics Computing
Host Institution – UNC Pembroke
Dr. Deok-Hyun Hwang, primary instructor
Summer I, June 2005
Remote Classrooms
2 students at Appalachian State University supervised by Barry Kurtz
Enrollment at other institutions is not known at this time

74. Research Outcomes Bioinformatics Computing
Set up grids on Dell machines as well as Compaq machines
Set up a stand alone Blast system with database
Design and build globus-enabled HTC Blast on a grid
Implement parallel MPICH-Blast on a grid

75. Capstone Research Course
Coordinator: Mark Holliday
Goal: Provide an opportunity for a student to explore one of the course areas in further depth
Offered: Fall 2005
Participating Sites

76. Capstone Research Course
Implementation:
Student works with a faculty member in a specific course area
Project presentation to the entire class using NCREN
Mark Holliday provides grid support and overall coordination
Participating Sites

77. The Future
Barry Kurtz

78. What Remains to be Done Welcome three new schools into the consortium
UNC Charlotte, High Point University, Lenoir Rhyne College
Invite additional schools to join the consortium
Teach the Summer 2005 Workshop
Intended for other UNC schools and IT industry
At Appalachian State July 29-31, 2005
July 29: MatLab, MPI; July 30: Grid Setup; July 31: Grid Applications
Teach the Capstone Research Course
Hosted at Western Carolina by Mark Holliday
Teach seven other courses for a second time
Emphasis use of HPC hardware and software
Involve student in cutting edge research projects

79. Future Plans
Institutionalize the high performance computing courses at the host institutions
Continue research work in high performance computing
Sponsor student project work in high performance computing
Apply for external funding from the NSF and other funding agencies