ARTE: an Application-specific Run-Time Management Framework for Multi-core Systems

Programmable multi-core and many-core platforms increase exponentially the challenge of task mapping and scheduling, provided that enough task-parallelism does exist for each application. This problem worsens when dealing with small eco- systems such as embedded systems-on-chip. In fact, in this case, the assumption of exploiting a traditional operating system is out of context given the memory available to satisfy the run-time footprint of such a configuration. An efficient Run-time Resource Management (RRM) becomes of paramount importance to dispatch tasks to the cores by taking into account the task-parallelization options that each application provides. State-of-the-art approaches to RRM try to allocate re- sources to maximize the instantaneous throughput while meeting a power budget constraint. In this paper, we will show that queuing theory can be an alternative yet effective way of solving resource allocation by presenting ARTE, an Application-specific Run-Time managEment framework. The framework exploits few assumptions about the target many-core computing fabric such as the availability of performance (throughput) information about the platform applications. We will show that this information can be combined, at run-time, with queuing models to enhance the response time of the applications by pounding the actual effect on the system power consumption better than previous approaches. Experimental results show that, compared to reference state- of-the-art RRM techniques, ARTE is able to efficiently improve system performance by pro-actively reducing the response time while meeting the same power consumption requirements. Be- sides, we will show that the run-time overhead of ARTE does not significantly impact neither the system performance nor the on-chip-memory occupation