Profiling-based task graph extraction on multiprocessor system-on-chip

This paper proposes a profiling-based method to extract a task graph, which describes the system behavior of a multiprocessor system-on-chip with Android OS. The proposed method computes the resource usage of each task and extracts dependency among tasks using the run-time system profiling results. The proposed method calculates CPU resource usage and I/O waiting time of each task by analyzing CPU scheduling and frequency logs. Also, the proposed method extracts the intra-process task dependency using the log of wakeup operations, and extracts the inter-process task dependency using the log of binder transactions. The proposed method describes system behavior without using the source code of the scenarios running on the MPSoC, has low intrusiveness, and guarantees the extraction of thread-level parallelism.