Goals and Purpose

Our project and goals included building a small-scale cluster, and learning both implementation and application levels of clustered networking. We also created a reference for undergraduate computer scientists interested in this topic. The reference materials can be used in an undergraduate curriculum in courses such as computer architecture, operating systems, or parallel processing/programming topics.

Research Process

We were given equipment from TCNJ Information Management that allowed for the development of a homogeneous system of 6 PC's with Pentium 75 MHz processors, 32 Meg of RAM, and various hard drive sizes. For networking, each machine contained a 10Meg ISA network card, and we were also given a 3-Com 100Meg Ethernet switch.

We originally chose Debian Linux for our operating system; however, we were eventually forced to abandon Debian due to a bug in the installation program. The second distribution we selected was Red Hat Linux 6.2. We were easily able to do a successful install on most of the nodes. Then experimented with installing only the minimum required packages in order to successfully install Red Hat on the nodes that had 500 M hard drives. We then networked the machines, and configured them to be nodes on the campus network with campus IP addresses, and mounted a common directory across the cluster with NFS.

We began experimentation of clustering software with PVM (Parallel Virtual Machine). We installed the program and the appropriate daemons, and successfully ran example programs. After further research, we discovered the PVM is no longer supported by many research institutions and had been replaced by MPI (Message Passing Interface).

We decided to replace PVM with MPI as the intra-cluster communication method. This involved reconfiguring all the machines with the MPI program, and configuring it to work with SSH (Secure SHell). We also configured the machines to communicate with each other without authentication during a parallel program.

After reconfiguring the nodes, we spent the remainder of the year on running and analyzing parallel programs in C using the MPI libraries.

Through out the process of creating the cluster, we ran into many problems and spent a great deal of time dealing with older obsolete hardware. We had to overcome problems like failed hard drives and CD-Roms, as well as upgrading RAM.

Through each step of the building and configuring of the cluster, we documented our work and created a Linux cluster HOWTO for other undergraduates.

Conclusion and Results

Our research this year resulted in having a 6-node cluster up and running that will be utilized for future projects. Next semester, the cluster will be used for students interested parallel programming research. Two students have signed on to continue this project.

Although we did not finalize learning modules, we created a full website featuring our research papers on the project and HOWTO for other undergraduates. The site can be found at http://www.tcnj.edu/~crew

The dissemination of our project included participation at various poster presentations. We were selected to present at the Consortium for Computing in Small Colleges Northeastern Region this year on April 19th at Middlebury College, Middlebury, VT. We also presented our work locally at TCNJ on April 25th for the Celebration of Student Achievement, and on April 2nd at Spring Day for prospective freshman computer science majors. We also presented a status report to the Department of Computer Science in December 2000.

Students and Faculty Mentor

Our home institution is The College of New Jersey in Ewing, NJ. Both student researchers, Dana Drag and Lorraine Juzwick were senior Computer Science majors in the 2000/2001 school year for the duration of the CREW project. Dr. Deborah Knox is an Associate Professor of Computer Science (knox@tcnj.edu).