Showing papers by "Todor Stefanov published in 2015"

Energy-efficient mapping of real-time streaming applications on cluster heterogeneous MPSoCs

Abstract: In this paper, we propose a novel polynomial time algorithm, called Frequency Driven Mapping, to map real-time streaming applications specified as cyclo-static dataflow (CSDF) graphs onto a cluster heterogeneous MPSoC. The objective of our mapping approach is to reduce the energy consumption and guarantee latency and throughput constraints. The main novelty in our mapping algorithm is twofold: 1) By using hard-realtime scheduling of CSDF graphs, we propose an efficient way to determine a suitable processor type for each task in a CSDF graph, where the energy consumption is minimized and throughput and latency constraints are met; 2) According to an initial mapping derived by a first-fit-decreasing heuristic, we propose a remapping approach, where some tasks are remapped to unused clusters in order to further reduce the energy consumption of the system by cluster dynamic voltage/frequency scaling (DVFS). The experimental results show that the proposed algorithm finds more energy efficient mapping compared to existing approaches. The energy savings due to our proposed algorithm are up to 34%.

Improved hard real-time scheduling of CSDF-modeled streaming applications

Abstract: Recently, it has been shown that hard real-time scheduling theory can be applied to streaming applications modeled as acyclic Cyclo-Static Dataflow (CSDF) graphs. However, that approach is not efficient in terms of throughput and processor utilization. Therefore, in this paper, we propose an improved hard real-time scheduling approach to schedule streaming applications modeled as acyclic CSDF graphs on a Multi-Processor System-on-Chip (MPSoC) platform. The proposed approach converts each actor in a CSDF graph to a set of real-time periodic tasks. The conversion enables application of many hard real-time scheduling algorithms which offer fast calculation of the required number of processors for scheduling the tasks. We evaluate the performance and time complexity of our approach in comparison to several existing scheduling approaches. Experiments on a set of real-life streaming applications demonstrate that our approach: 1) results in systems with higher throughput and better processor utilization in comparison to the existing hard real-time scheduling approach for CSDF graphs while requiring comparable time for the system derivation; 2) gives the same throughput as the existing periodic scheduling approach for CSDF graphs but requires much shorter time to derive the task schedule and tasks' parameters (periods, start times, etc.); and 3) gives the throughput that is equal or very close to the maximum achievable throughput of an application obtained via self-timed scheduling, but requires much shorter time to derive the schedule. The total time needed for the proposed conversion approach and the calculation of the minimum number of processors needed to schedule the tasks and the calculation of the size of communication buffers between tasks is in the range of seconds.