IBM: PERCS
DARPA-funded IBM research project to develop easier-to-use parallel computers.
Links:
IBM: PERCS: 10x Productivity Baseline vs. Legacy Systems
The competition required us to prove that PERCS would be significantly easier and faster for development than common practices circa 2002. I set up research plan to replicate 2002 behavior for a user study, analysis, then ran six studies with experienced supercomputer programmers on legacy computers to observe how they did things “in the old days”. Analysis of logged programmer action data, and compared results to observational data. I applied the findings to compare to work being done for the new PERCS tools. This baseline was instrumental for showing that PERCS improved programmer productivity by 10x to satisfy the contract. I identified unsupported tasks in the new X10 tools, and convinced the development team to put in some of those features.
Also did analysis of help desk issues during 2002 to understand primary problems that were being reported then.
PERCS: Qualitative Study
The competition required us to prove that PERCS would be significantly easier and faster for development than common practices circa 2002. I set up research plan to replicate 2002 behavior for a user study, analysis, then ran six studies with experienced supercomputer programmers on legacy computers to observe how they did things “in the old days”. Analysis of logged programmer action data, and compared results to observational data. I applied the findings to compare to work being done for the new PERCS tools. This baseline was instrumental for showing that PERCS improved programmer productivity by 10x to satisfy the contract. I identified unsupported tasks in the new X10 tools, and convinced the development team to put in some of those features.
Also did analysis of help desk issues during 2002 to understand primary problems that were being reported then.
Evaluate X10 Programming Language Tools
Did two cross-language studies to understand both how the X10 language would help, and how tools improved productivity for any language. Did one of experimental design for one study and observational component for all, led four researchers running study and conducting interviews, and worked with host institutions (Rice, Pittsburgh Supercomputer Center) to set up study. Each study involved 15-20 programmers, conducted over a week at a host institution. Came up with great results of how X10 would help. My research results also encouraged changes to the Eclipse Parallel Tools Platform.
MDAT: Multiple Device Authoring Tools
MDAT was an IBM tool for building applications that could be written to run on multiple mobile operating systems, and highlighted the code tweaks that differed. With Ingrid Erickson, I did a video and observational study of two different teams learning to MDAT for coding. We discovered that the developer’s idea of “model-view-controller” didn’t match the MDAT developers’ understanding of the term, leading to confusion when all thought that the concept behind MDAT was understood by all the same. From our feedback, the development team realized they needed to rework their documentation because common concepts weren’t really common.
Links:
Papers:
- Banavar et. al, "An Authoring Technology for Multidevice Web Applications". IEEE Pervasive Computing 3(3) 2004.