Students

Dissertation title: Scalability Issues in the Development of Tools for Massively Parallel Systems.

Chu's dissertation addresses the development of support tools for debugging and performance tuning of massively parallel applications.

Chu passed his proposal defense in January 1998.

Dissertation title: Using Kernel Extensions to Decrease the Latency of User-Level Communication Primitives.

Rolf works at Sandia National Labs where he is the technical lead on the C-plant project. His dissertation addresses approaches to kernel extensibility from the perspective of high performance, massively parallel machines. Rolf's proposal is available in html and postscript.

Rolf passed his proposal defense in September 1996.

Patricia is looking at high-performance implementations of commodity prtocols, i.e., TCP/IP. Along with Todd Underwood (see MS students below), she has developed an implementation of IP for Gigabit Ethernet that offloads fragmentation and reassembly to the NIC. This implementation achieves high bandwith with low and predictable latency while greatly reducing CPU utilization.

Patricia passed her general exams in January 2001.

Gabriela is attending UNM on a Fullbright scholarship. This semester, she's mostly taking classes preparing for the comprehensive exams. Gabriela is also doing some background research looking into reliable transmission protocols for highly reliable networks.

Gabriela is planning to take the general exams in August 2001.

Ron has an MS degree from Mississippi State and has just started in the PhD program at UNM. Right now, he's taking courses getting ready to take the comprehensive exams. Like Rolf, Ron works at Sandoa National Labs on the C-plant project. We used to call Ron "the MPI guy," but lately it seems like he's doing a lot more than MPI implementations.

Ron is planning to take the general exams in January 2002.

Jared's proposal is titled Theory Development for Sensor Deception Analysis. His dissertation will address the development of a theory of deception from remote, embedded sensors.

Jared defended his proposal in May 1998.

Thesis tile: Improving IP Performance by Offloading Fragmentation and Reassembly

Along with Particia Gilfeather (see PhD students), Todd has implemented a NIC control program that offloads IP fragmentation and reassembly from the host processor. This implementation achieves bandwidths comparable to the standard implementation (990 Mb/s) while reducing the CPU uitlization from 80% to 40% as measured by netperf.

Todd defended his MS thesis proposal in September 2000.

Thesis title: Measuring the Impact of Varying Communication Overhead on Application Performance

Edgar is currently working as a student intern at Intel in Santa Clara, but is still running tests remotely.

Edgar has modified a version of GM so that he can vary communication timing parameters (e.g., host processor overhead and latency). He has also ported several of the NCSA community codes to run on a small cluster that uses his modified GM drivers.

Thesis title: An Implementation of Portals 3.0 on Gigabit Ethernet

Wenbin is writing a control program to implement Portals 3.0 on the Alteon ACE Nics.

Bill just started working in the group and is presently looking for a thesis topic.

Dena is developing a test suite for Portals 3.0.

Riley is working on collecting and developing performance evaluation tools.

Using Execution Disciplines to Speed-up Computer Programs

Phil used PDW constructs to examine execution variability under control-driven, data-driven, and demand-driven executions.

Persistence in Distributed Object Systems: ORB/ODBMS Integration

Francisco spent most of his time at Los Alamos National Labs where he worked on the Sunrise/Telemed project. Check out Francisco's home page, it has links to the project's he's involved in.

Francisco is back in Brazil now, working as an Assistant Professor in the Department of Computer Science at the University of Sao Paulo in Brazil.

Representing and Extracting Program Dependences

When Krishna finished his degree, he went to work for (what used to be) Cray Research in Santa Fe. After spending his first year working on the final version of Cray's Pascal compiler, it looked like Krishna would finally get a chance to start working on program dependence again. Then he decided that SGI/Cray wasn't for him anymore.

Krishna is now working on smart cards for Schlumberger in Austin, Texas.

Stephen is now employed at Intel in Chandler AZ.

In August of 1995, Stephen was named outstanding alumnus of the CS department at UNM. In December of 1994, Stephen lead the group that set the world record (281 Gflops on MPLINPAK, and 328 Gflops on an LU factorization code). Stephen was also a member of the team that won the 1994 Gordon Bell Award.