Val Tannen's Talk on DB Seminar

Title: Provisioning for Autonomous Exploration of Query Answers under What-If Scenarios
Val Tannen
EPFL and University of Pennsylvania

In business intelligence and in data-driven scientific investigation a user is seldom satisfied with the result of just one fixed query as it runs on the current state of the data sources. Instead, she would
often also like to see how that result would be affected by hypothetical updates by and query modifications, i.e., by different combinations of ``what-if'' scenarios. Performing this kind of
exploration by simply repeating queries and/or updates and rollbacks is often impractical or even impossible. In this context we ask if it is possible to support the lightweight materialization of a
representation of a space of query answers that is autonomous of the data sources but still allows easy recovery of answers corresponding to any combination of what-if scenarios.

Specifically, we assume that the user will specify an initial query and multiple what-if scenarios in advance. From them and from an initial query a system would generate a special kind of query that we
term a *provisioning query*. This query would run just once, and will produce a special kind of answer that we term a *provisioned representation* which, crucially, is autonomous of the data sources. This representation can be explored with a tool where combinations of what-if scenarios are dialed and set as desired. In some ways this can be seen as generalized OLAP: a ``cube'' of scenarios.