Behavior on read
Scuba's strategy of analyzing behavior as it rapidly scans data from disk. Scuba goes beyond row-level field extraction and summarization — we aggregate behavioral insights per-actor and for the population as a whole; we maintain state and gather insights across actor journeys. Our strategy doesn't depend on pre-aggregating data: pre-aggregation data techniques limit flexibility and slow the exploration process. Our Behavior-on-Read method gives less technical users more flexibility and power by making queries on newer/changed constructs just as fast as those that were previously defined. Lastly, Behavior-on-Read strengthens our ability to work with multiple types of actors within the same dataset and not limit behavioral analysis to a single population of "user."