Big Data Management Theory
We are studying the foundations of efficient Big Data management, which leads to new fundamental results regarding the complexity of various ad hoc data transformations in modern massive-scale parallel systems.
Our model of computation is an extension of the Massively Parallel (MPC) model, where computation is performed by p severs that alternate between local computation and re-shuffling steps, and each server is limited to O(n/p) space, where n is the size of the data. We study the time/space complexity of data transformations and extensions to cope with failures.
