Brandon Salmon: Adapting and Thriving on Key in ECE


February 2, 2004

Although Brandon Salmon studies adaptive systems, the second year ECE graduate student and winner of the 2003 NSF Graduate Research Fellowship does much more than simply fit into his surroundings - he thrives on solving complicated networking problems. Outside of the classroom, he sings baritone with the Bach Choir of Pittsburgh.

He was drawn to ECE for its strong program, helpful students and sense of community, and sings high notes of the Pittsburgh music scene and his advisor, Greg Ganger, associate professor of ECE and CS and director of the Parallel Data Lab (PDL), a storage systems research center at Carnegie Mellon.

"I enjoy the freedom it gives you, and the kind of problems you could solve," Salmon explained how he chose his area. "Since I enjoyed systems, I wanted to work with Greg to get closer to the hardware and networking, so I went to computer engineering."

Together they are creating a new Special Topics in Computer Engineering course in graduate algorithms, File Systems Survey, and plan to publish the results.

"Brandon is developing an interesting mix of AI and software systems expertise, and applying it to close out long-standing problems in computer systems. I have big expectations for his continuing research," Ganger said.

Probing the challenges facing engineers today, Salmon took Security and Cryptography with Mike Reiter, professor of ECE and CS and technical director of the Carnegie Mellon CyLab.

"His project for the class was quite strong, demonstrating critical thought, knowledge of the relevant literature, and insight into an important problem," Reiter established.

Trained to investigate how systems can fix problems with their surroundings without being told, Salmon works toward a solution when he sees a need, and is examining adapting a disk layout to its workload to help users store and access data. This line of inquiry explores continuous data reorganization with automated tuning.

Salmon also inspects expressive storage interfaces that foster communication and enable storage systems to work with less effort and more speed. Additionally, he is part of a team defining self-* storage systems (pronounced "self-star"), which simplify storage administration.

"One of the challenges is using machine-learning techniques and applying them in systems," he observed. Machine-learning proponents can spend lots of resources to find exact solutions, while systems problems are often solved in a limited resource environment, Salmon summarized.

A rewarding moment for Salmon was coauthoring a white paper, "A Two Tiered Software Architecture for Automated Tuning of Disk Layouts" with Ganger and classmates Eno Thereska (ECE) and Craig Soules (CS). They presented their findings at the First Workshop on Algorithms and Architectures for Self-Managing Systems, part of the Federated Computing Research Conference (FCRC) in San Diego.

Additionally, he participated in the 2nd USENIX Conference on File And Storage Technologies (FAST '03) in San Francisco with Ganger. Salmon suggests that graduate students attend seminars on campus, through the department and the PDL: "They are a really broad overview. You can see what other problems people are solving to apply to your research." He advises undergraduates to consider research opportunities, too.

Growing up in the California Bay Area, he always loved computers and tinkering with systems; this fascination led him to earn a B.S. in computer science at Stanford. After finishing his Ph.D. at Carnegie Mellon, Salmon would like to pursue industrial research.


Related Groups: