2008 Symposium for Graduate Research

The Symposium for Graduate Research (SGR) is a conference for graduate-level computer science research run by graduate students. The SGR 2008 will be held at North Carolina State University by graduate students in the Department of Computer Science. The goal of the SGR is to disseminate emerging and established graduate research from across the United States to graduate students, faculty, and industrial partners, thus providing the opportunity for collaboration and external feedback.

Paper Abstracts

Contributions of Formal Computational Approaches to Designing and Evaluating Human-Machine Interfaces
Maria Vicente Bonto-Kane and Robert St. Amant

This research shows the use of formal computational approaches to design and develop more usable human-machine interfaces. Traditional methods in the field of Human-Machine Interaction (HMI) have used task analyses, cognitive walkthroughs, live user testing and the GOMS family of modeling techniques. While decades of research show the success and usefulness of these techniques, they still do not address all areas of concern within HMI usability. This research seeks to examine a less popular approach, the use of formal computational approaches. Formal computational approaches are better able to make an impact in the early stages of design, where usability initiatives can make the most intervention. Also, formal computational approaches allow one to track usage patterns and develop formal probabilistic models of how a task is accomplished on a device. This research shows that formal models help develop more usable interfaces that are facilitative of the most highly probable operations of a task.

Adding Agile Practices on an Embedded Systems Project
Leyna Cotran, Susan Elliott Sim, John Noll

With the emergence of Agile approaches, aerospace projects in particular are examining Agile and exploring its feasibility in defense projects. Aerospace projects generally follow strict processes and standards as required by government contracts. The contribution of this experience report addresses Agile's practicality in large scale settings such as aerospace. A traditional defense program would inherently have a difficult time adopting Agile approaches because of the characteristics of these large scale systems. These systems are entirely single trial in their production system, meaning that frequent releases of software, for instance, is usually not possible. The complexities of the software in large scale systems are directly impacted by characteristics of the hardware which does not allow for frequent software releases due to the cost and resources required. This experience report incorporated Agile techniques to mitigate the customers concerns on the software's experience. Through the use of Agile techniques, the project's outcome was successful.

Scalable I/O on Distributed File Systems for Parallel Genomic Searches
Heshan Lin, Xiaosong Ma, Wuchun Feng

Parallel I/O techniques have traditionally been designed with respect to parallel numerical simulations and floating-point computation. In contrast, today's computational biology and bioinformatics applications (and arguably, data mining in general) rely more on large scale data comparison, thus requiring re-examination of I/O strategies for efficient file access. Consequently, in this paper, we uncover many common I/O characteristics of parallel genomic searches, which form an important class of workloads in the life science. These characteristics - such as highly irregular, computation-time distributions, dynamic file layout, and the wide use of distributed file systems - may result in poor I/O performance with common sequential or parallel I/O interfaces. Further, existing parallel I/O optimizations do not work well for these applications. To address this problem, we propose an asynchronous, distributed, two-phase I/O strategy. It avoids the implicit synchronization and I/O contention introduced by collective I/O and largely eliminates the processing overhead and memory requirements brought by centralized I/O. We evaluate our I/O strategy with popular BLAST tools on three clusters and show that it significantly improves both the applications' I/O performance and their overall performance.

For more information, email the Organizing Committee at csc-sgr-org AT lists.ncsu.edu.

Sponsors: