Previous approaches to visualize distributed system logs focused mostly on detailed views at the scale of single message traces or short groups of log lines, however, a full distributed log can be as large as a terabyte. While detailed views are quite useful, the problem of deciding which traces, nodes, or time-spans to inspect remains unsolved. We created VectorVis, a tool that enables interactive exploration with coordinated detail and summary views. The premise of the tool is that it is meant as a pre-processing step to filter large log files and output a reduced log to ShiViz, a sophisticated distributed log visualization tool that focuses on inspecting the “happens-before” relationship. We contribute (1) a task analysis that motivates the major design choices of VectorVis; (2) a detailed description of our design process that articulates design trade-offs and requirements; (3) a prototype that demonstrates key interaction motifs necessary for mediating summary and detail views.

I created this application together with Paul Bucci for the Distributed Systems grad course (538B) at UBC.