Optimizing Flows for Real Time Operations Management

Modern data analytic flows involve complex data computations that may span multiple execution engines and need to be optimized for a variety of objectives like performance, fault-tolerance, freshness, and so on. In this paper, we present optimization techniques and tradeoffs in terms of a real-world, cyber-physical flow that starts with raw time series sensor data and external event data, and through a series of analytic operations produces automated actions and actionable insights.