JPDI: a runtime for parallel execution of macro data flow graphs

These days Multicore processors are ubiquitous, structured parallel programming models for these systems are generally not flexible to describe tasks with arbitrary dependency. Macro Data flow programming model is an effective way to represent parallelism, in a way where the computational logic is modeled as a Directed Acyclic Graph, and the computations are coarse grain. We present a Java8 Library to support the execution of task graphs. The library is aware of scheduling and management of micro-tasks on a multi core architectures. The scheduling policy takes into account data and control dependency, and helps us to achieve good performances. We demonstrate the efficiency of our approach by implementing synthetic applications and validate our proposal. Some experimental results give an idea of computational grains that can be exploited on current state of art multi core systems