Abstract

DCTV is a trace analysis toolkit aimed at letting investigators pose succinct, comprehensible, and expressive questions to system trace files and get back interpretable answers useful as the basis for further analysis. It provides a custom dialect of SQL that extends the traditional relational algebra with special operators and syntax for analyzing the heterogeneous time series one finds when analyzing system performance. Coupled with comprehensive dimensional analysis and a custom columnar database back-end for high performance analysis of larger-than-memory datasets, DCTV provides new power in trace analysis. This talk is a brief introduction to the design, motivation, and future direction of DCTV.

Bio

Daniel Colascione is a staff software engineer at Google working on the Android core performance team. He previously worked on mobile performance at Facebook and the Windows Phone core team at Microsoft. (The common theme is a lot of time spent puzzling over long trace files.) Daniel is also a core contributor to GNU Emacs and author of the popular “Buttery Smooth Emacs” article from a few years ago.