Differential dataflow: a model and implementation
Differential dataflow is a model for incremental computation which shifts from sequences of changes to partial orders of changes. This shift provides substantial benefits in the incremental maintenance of iterative computations, such as Datalog, graph algorithms, and other more general settings. I’ll lay out the model of differential dataflow, detail the important features of the current data-parallel implementation, and demonstrate its performance on a range of what I hope are thought-provoking tasks. The talk is meant to be understood, and interactive, so bring your most pressing questions!
Differential dataflow is available at: https://github.com/TimelyDataflow/differential-dataflow
Bio: Frank McSherry is Chief Scientist at Materialize, Inc. Frank shared the Godel prize for the invention of differential privacy, the TCC test of time award for its theoretical development, and the SIGMOD test of time award for its initial implementation. He led the award-winning Naiad project at Microsoft Research, which led to the invention of timely and differential dataflow. Frank’s laptop is similarly accomplished.