Section: New Results

Memory Management and Distributed Computing with StarPU

Task-based programming models manage to abstract away much of the architecture complexity while efficiently meeting the performance challenge, even at large scale. While computation scheduling has been well studied, the dynamic management of memory resource subscription inside such run-times has however been little explored, despite the fact that the lookahead, anticipative capabilities offered by the decoupled task submission/task execution steps may sometime come with a high memory cost, especially in distributed context where buffers for receiving incoming contributions have to be accounted for. We therefore studied the cooperation between a task-based application code and a run-time system engine to control the memory subscription levels throughout the execution. We showed that the task paradigm allows to control the memory footprint of the application by throttling the task submission flow rate, striking a compromise between the performance benefits of anticipative task submission and the resulting memory consumption.